Uncertainty/Sensitivity analysis of the transport model ... · The uncertainty in the calculations...

Date

April 08, 2011

Authors

Charis Kouridis

Ioannis Kioutsioukis

Thomas Papageorgiou

Stephen Mills

Les White

Leonidas Ntziachristos

Client

European Commission

Climate Action DG

Directorate A: International & Climate Strategy, Unit A4: Strategy & Economic Assessment

1049 Brussels, BELGIUM

Final Report

EMISIA SA Report

No: 11.RE.01.V3

Uncertainty/Sensitivity analysis of the transport model TREMOVE

EMISIA SA ANTONI TRITSI 15-17

SERVICE POST 2 GR 57001 THESSALONIKI GREECE

tel: +30 2310 473352 fax: + 30 2310 473374 http://www.emisia.com

Project Title

Uncertainty/Sensitivity analysis of the transport model TREMOVE

Call for Tender No:

ENV.C.5/SER/2009/0017

Report Title

Final Report

Contract No:

07.0307/2009/538771/C5

Project Manager

Giorgos Mellios

Author(s)

Charis Kouridis, Ioannis Kioutsioukis, Thomas Papageorgiou, Stephen

Mills, Les White, Leonidas Ntziachristos

Coordinator: EMISIA SA

Sub-contractor LWA Ltd

Summary

This is the final report for the Climate Action DG of the European Commission project designed to estimate the uncertainty of the TREMOVE output and its sensitivity to the input variables. This report includes a summary of the methods used, the results of the uncertainty and sensitivity analysis for the baseline and three scenarios, and a list of conclusions and recommendations. The report is accompanied by a DVD with a summary of the output in aggregated form and a full dataset of all modelled output. For the analysis, United Kingdom was examined as a test case and TREMOVE v3.3.1 was used. The study identified 14 variables that were found to be most important for the uncertainty of the model output. It also identified linear associations between output and input variables in several instances. Elasticities between intermodal shifts and other choices (vehicle types, fuels, etc.) appeared limited. Choices to decrease model uncertainty include better estimates for key input variables and simplification of the model structure.

Keywords

Uncertainty, Sensitivity, Monte Carlo, TREMOVE, Statistics, Emissions

Internet reference

http://www.emisia.com/gui/unc.php Version / Date

Final Version / 08 April 2011

Classification statement

PUBLIC No of Pages

233

No of Figures

24

No of Tables

23

No of References

19 Approved by:

Emisia is an ISO 9001 certified company

Contents

Executive Summary ........................................................................................ 7

1 Introduction ................................................................................... 17 1.1 Background ............................................................................... 17 1.2 Objectives of the study................................................................ 19 1.3 Structure of this report................................................................ 19

2 Uncertainty ranges of input variables and modelling parameters ............ 20 2.1 General..................................................................................... 20 2.2 Input variables description........................................................... 21 2.3 Emission factor modelling ............................................................ 47 2.4 Input variables and parameters not varied ..................................... 53 2.5 Changes over interim report......................................................... 54

3 Modelling theory / approach ............................................................. 58 3.1 General..................................................................................... 58 3.2 Methods.................................................................................... 58 3.3 Parameterisations of input data .................................................... 62

4 TREMOVE software modification and update........................................ 74 4.1 General..................................................................................... 74 4.2 Software code modification .......................................................... 74 4.3 Software code added .................................................................. 75 4.4 New features ............................................................................. 77 4.5 Guidance to use the software ....................................................... 78 4.6 Differences between the two steps................................................ 78

5 Variance of the baseline output ......................................................... 80 5.1 General..................................................................................... 80 5.2 Screening uncertainty and sensitivity analysis ................................ 81 5.3 Variance-based uncertainty and sensitivity analysis ......................... 96 5.4 Discussion................................................................................109

6 Uncertainty and sensitivity analysis of scenarios .................................112 6.1 General....................................................................................112 6.2 Methodology.............................................................................112 6.3 Ownership tax increase ..............................................................112 6.4 Effect of fuel cost ......................................................................118 6.5 HDV Euro VI .............................................................................123

7 Conclusions and recommendations ...................................................129

References .................................................................................................135

ANNEX I: uncertainty estimates of the baseline per vehicle category ...................137

ANNEX II : uncertainty estimates of the scenarios per vehicle category ...............161 Scenario 1.........................................................................................162 Scenario 2.........................................................................................185 Scenario 3.........................................................................................208

ANNEX III: Description of the DVD contents ....................................................231

ANNEX IV: Description of the full dataset ........................................................233

7

Executive Summary

This is the final report of the study entitled “Uncertainty/Sensitivity analysis of the transport

model TREMOVE”, funded by the European Commission. TREMOVE has been the main model

used in Europe for impact assessments of road transport related policies. Recent applications of

the model include impact assessments related to the Eurovignette directive, the Euro VI

emission standards for heavy duty vehicles, CO2 regulations for passenger cars, etc. Aim of this

study was to characterize the uncertainty in the output of the TREMOVE model (v3.3.1 – June

2010), i.e. to estimate the variance in the activity and emission results and to identify the

factors which are most important for this variability. Specifically, the study aimed at:

- Identifying the variance of the input data to TREMOVE.

- Determining the uncertainty range of the basecase.

- Determining the uncertainty range of three indicative scenarios.

- Conducting a sensitivity analysis to identify the most important factors in terms of

uncertainty.

TREMOVE covers all transport modes (road, rail, aviation, inland waterways, maritime) for all

EU27 member states plus Switzerland, Croation, Norway, and Turkey. The uncertainty in the

calculations should in principle depend on the country considered, as different sets of

parameters are important in each case. However, in this study we addressed uncertainty only

in the case of UK as an example. Characterising the uncertainty for all countries would have

been impossible due to time and cost constraints. UK was selected of all countries, as we could

obtain access to estimates of primary data variance. Uncertainty estimates are derived for the

transport activity, vehicle population, fuel consumption, pollutant emission factors (PM, CO,

VOC, NOx), and cost components.

This report outlines the main findings of this study and makes recommendations on how the

quality of the output may be improved. It follows up on an inception report clarifying the

targets and the approach implemented and on an interim report discussing the selection of the

statistical method used.

Approach

TREMOVE aims at assessing the environmental impact of different policy options. Each new

policy is associated with a marginal cost which drives the demand. This is simulated in the

demand module. The new demand also leads to variations in vehicle choices, which are

estimated in the stock module. Then, an emissions and fuel consumption module calculates the

environmental impact of the vehicle stock operation. The calculation is done on an iterative

process, as the costs initially assumed and finally calculated should match. Following the

iterations, a welfare and a well-to-wheel module calculate the total cost-benefit (including

external costs) and upstream fuel production emissions respectively.

8

The model is built around a basecase which is exogenous. This exogenous basecase offers

transport quantities and costs which are used to calibrate the price elasticities of the model,

based also on assumptions regarding the elasticities of substitution between different transport

options. The elasticities of substitution themselves are also either estimates receiving a value

while calibrating the model, or constant empirical values. In general, several elasticities of

substitution are selected during calibration so that the derived price elasticities receive values

that are close to the literature values. After calibration, the price elasticities may be used to

simulate alternative scenarios. This scheme offers intrinsic difficulties in characterizing the

uncertainty of the demand, when using any TREMOVE version:

1. The model uses detailed exogenous data (from SCENES, PRIMES, TRANS-TOOLs, etc.)

to define its baseline. As a result, the uncertainty of the basecase is largely defined by

the uncertainty in the output of such higher-order models.

2. All demand functions are calibrated to this exogenous basecase. During simulations,

the model uses these calibrated values. Individually changing any of the price

elasticities of the model to assess its impact on output uncertainty will bring the model

off-balance.

3. Tremove uses elasticities of substitution which are constant with respect to the income

while, in scenarios, the income is assumed constant relative to the basecase. In

principle this means that the demand module can be used to calculate changes in

demand only for small changes in the (generalized) cost. The definition of ‘small’ in

this last sentence is arbitrary. The expected error in the calculation of the demand

using the Tremove demand tree structure increases the more distant are the scenario

costs from the basecase costs. This introduces an uncertainty which is exogenous to

the model and cannot be estimated. However, it largely defines the total uncertainty of

the calculation.

Individually characterising the uncertainty of the TREMOVE demand module would mean to

quantify the output variance, using realistic variability indicators for the input data. For the

reasons outlined above, it was decided that a realistic uncertainty analysis of the demand

module is not possible. Neither the input data can be independently varied (since each model

version is calibrated around a fixed basecase), nor the output variance produced would be

realistic (due to the constant elasticities of substitution effect). The only useful approach that

can be recommended to assess the demand module uncertainty is to compare its output using

the projections of alternative models as input. This can be considered to encapsulate all

uncertainties, ranging from uncertainties in input data, modelling approach, assumptions for

parameters, etc. However, this was clearly not the target of this study.

The analysis therefore focused on the characterization of the uncertainty produced by the

vehicle stock and the emission modules. The parameters of the two modules are perturbed

across their range and the output variance is observed. This is done for the basecase and for

three alternative scenarios built to reflect options related to taxation policy, fuel costs, and

introduction of a new emission standard. The statistical method chosen also results to a

detailed sensitivity analysis, i.e. demonstrates which individual model variables or variable

combinations are important (i.e. having large impact on model output) or not.

9

The approach followed is considered to calculate not a theoretical uncertainty, but the actual

uncertainty of the model output when applying it to examine the impact of a new policy.

Therefore, our approach is considered to reliably address the following points:

1. The uncertainty induced by the model to the exogenously defined baseline.

2. The confidence by which activity and emission differences can be distinguished

between different scenarios and the baseline.

3. The key variables/parameters which can be better estimated to reduce the uncertainty

of the output.

Statistical methods

Various methods are available to evaluate the model output uncertainty and quantify the

importance of the input factors. The selection of the appropriate method is a function of the

system’s uncertainty and the stakes involved. For the case of models with a direct policy

orientation like TREMOVE, global sensitivity analysis methods are preferable. The global

sensitivity analysis methods involve multiple evaluations of the model using Monte-Carlo

simulations, where values for the input variables are selected according to specific sampling

strategies. Here, we have adopted variance-based techniques, and in particular the extended-

FAST, that display a number of attractive features like the exploration of the whole range of

variation of the input factors and the consideration of interaction effects. This approach allowed

us to gain useful insight in the TREMOVE processes in order to assess the dependencies of the

individual variables as well as the quality of its estimates, with the view to better defend policy

messages.

The high computational load of TREMOVE as well as its large number of uncertain input

variables is tackled through a screening sensitivity analysis experiment prior to the extended-

FAST setting. The relative importance of the uncertain input factors is explored with a design

based on quasi-random LPτ sequences. The estimated sensitivity coefficients, calculated in the

nth-D space defined by the input factors, identified the non-influential input factors that were

fixed in their nominal values in the subsequent variance-decomposition analysis.

Software implementation

The TREMOVE model code was modified in order to be able to execute the thousands of Monte

Carlo simulations that were necessary. The code was modified in several ways with regard to

its execution but it was not affected with respect to its operation. In particular, it has been

made possible to decouple scenario from basecase execution that has greatly reduced

processing time. New pieces of code have been added to induce perturbations to the input

variables across their variance range. Depending on the variable considered and the exact

formulation, this was made possible either by feeding externally alternatives, or by introducing

an error factor in the calculation of the variable, or by replacing the variable value after its

calculation by the code (e.g. emission factors). A new graphical user interface was also

developed to allow the Monte Carlo execution of TREMOVE.

10

Uncertainty of the input variables

All the input factors were examined in order to understand their role in the structure of the

model. Out of the more than 100 input variables, those who received fixed values were

excluded in the analysis. Also variables related to the demand, welfare, well-to-wheel and non

road modules were removed from the analysis. This filtering procedure led to 33 input variables

for which uncertainty ranges were estimated.

The estimation of the uncertainty ranges for the input variables was done for the case of UK

using different available sources, including public authorities, clubs and association information,

vehicle manufacturers, research institutes, the public domain, and expert judgment in some

cases where published information was not available. Uncertainty ranges were collected for

vehicle related information (costs, specifications, scrappage rates, existence of air-conditioning,

etc), vehicle operation (mileage, speed, trip distance), fuel parameters (composition, heating

value, specifications), cost related issues (purchase, maintenance, ownership, labour, etc.),

environmental factors (temperatures), and infrastructure (availability of CNG stations).

With regard to internal model parameters, only the emission factor uncertainty has been

characterized, based on the available experimental data used to derive them. For hot emission

factors and fuel consumption, log-normal probability distributions were developed around the

factors for fourteen different speed classes. In the absence of robust experimental data for cold

start, the standard deviation over mean of the hot emission factors has been also used, also

assuming log-normal probability functions.

Uncertainty and sensitivity of the basecase

Based on the rationale outlined in the ‘approach’ section, uncertainty characterisation of the

basecase means to quantify the variance induced by the model variables uncertainty to the

fixed exogenous projection. It does not mean to quantify the uncertainty of the projection per

se, as this would require determining the uncertainty in the demographic and macroeconomic

data assumed to develop it. This is not relevant for TREMOVE but for the models used to derive

this projection. Also, the uncertainty of the basecase is limited to the model variables. For

example, TREMOVE includes no hybrid nor electric vehicles that are expected to become quite

popular up to 2030. Inclusion of such technologies would have increased the uncertainty of the

baseline but this was not possible to quantify, as these are not part of the model formulation.

The uncertainty of the basecase was assessed by formulating ‘baseline’ scenarios around the

fixed basecase. The scenarios were built by perturbing the values of the input variables along

their range collected from literature data. The mean value of the uncertainty range is very

close to the value assumed in the TREMOVE basecase for several variables. However, in other

cases (more predominantly for the purchase cost and ownership tax of two-wheelers) the

TREMOVE basecase value was much lower than the literature values. In these cases, literature

values have been used for baseline uncertainty characterisation. This led to some divergence

between the TREMOVE basecase and the mean output of the baseline scenarios, in particular

for two wheelers. This is recognised but it has no impact on the conclusions related to

uncertainty and sensitivity of TREMOVE.

In order to assess the uncertainty, a screening test (512 runs) was first performed to identify

the most influential variables. The screening procedure identified 14 input variables which

11

explain most of the uncertainty of the model output. Using the uncertainty of this limited set of

variables, TREMOVE was executed for 5950 runs in a Monte Carlo analysis to calculate the total

uncertainty of the model and to quantify the contribution of each variable to the output

uncertainty.

The uncertainty of the baseline in the years 2010, 2020, and 2030 expressed as a coefficient of

variance (cov) is shown in Table ES1. The main conclusions drawn from the uncertainty

analysis are:

Table ES1: Median value and coefficient of variation (cov) of the baseline output in TREMOVE

shown in a descending order according to the cov2030 ranks.

Output Variable Units Median 2010 Median 2020 Median 2030 cov 2010 cov 2020 cov 2030

CO Ton 252,000 119,402 119,439 69% 51% 51%

PM Ton 10,750 3,400 3,421 25% 27% 27%

VOC Ton 46,927 30,279 31,559 37% 26% 24%

TAXrest M€ -8,454 -8,768 -9,365 21% 22% 22%

NOx Ton 331,319 170,500 158,458 19% 17% 17%

COSTinsurance M€ 24,922 38,291 44,671 6% 13% 14%

TAXinsurance M€ 1,260 1,934 2,255 6% 13% 14%

VATfuel M€ 8,041 9,418 11,412 13% 13% 13%

TAXfuel M€ 33,301 41,490 48,064 12% 12% 12%

COSTfuel M€ 30,727 32,106 40,202 11% 11% 12%

TAXownership M€ 6,166 10,810 12,070 8% 11% 12%

FC Ton 46,618,917 55,591,702 60,043,602 11% 11% 11%

COSTrepair M€ 59,798 73,706 86,483 3% 10% 10%

VATrepair M€ 7,293 9,090 10,691 3% 10% 10%

COSTlabour M€ 10,869 14,896 16,607 9% 9% 9%

COSTlabourtax M€ 11,634 15,944 17,773 9% 9% 9%

VATpurchase M€ 10,841 11,566 13,107 5% 9% 9%

COSTpurchase M€ 84,178 99,323 114,467 4% 8% 9%

TAXregistration M€ 22.5 24.2 27.4 7% 8% 8%

VATrest M€ 1,252 1,293 1,377 5% 5% 5%

Costs M€ 323,333 396,036 458,228 2% 4% 4%

Vehicles # 33,652,081 37,918,723 40,997,888 2% 3% 3%

Vehkms ×106 km 585,653 665,914 720,553 2% 3% 3%

COSTrest M€ 41,193 44,395 47,695 2% 2% 2%

- The uncertainty is large for the emission of pollutants, mostly due to the uncertainty in

the emission factors. Cov’s are in the order of 20-30% but can reach up to 50% in the

case of CO.

- Fuel dependent variables (fuel costs and consumption) are second with regard to

output uncertainty with cov values in the order of 10-15%.

- Total cost figures exhibit uncertainty ranges in the order of 4-10%, i.e. they are rather

little dependent on the variance of the input data.

- Finally, population and activity data are found to be associated with very small

uncertainty, in the order of 2-3%. The uncertainty per vehicle category is of the same

magnitude. This means that large fluctuations in the input data (i.e. costs and other

variables) will result in relatively small changes in the activity and population data.

12

The contribution of each individual input variable to the uncertainty is summarized in Table

ES2. The first order dependencies (ΣSI’s) show how much of the output variance is explained

by the variance of each single input variable. A value of 1 would mean that the total output

uncertainty is explained only by first-order dependencies to the input variables.

Interdependencies of input variables become less and less important in explaining the output

variance the closer the value moves to 1. The table also shows the one or two most influential

input variables per model output. The following conclusions may be drawn based on the

sensitivity analysis conducted:

Table ES2: Summary of first-order (ΣSI) dependencies of output variance to input variance, in

a decreasing order according to the dependencies in 2030.

Output Variable Most Important Input Variable ΣSI2010 ΣSI

2020 ΣSI2030

COSTrepair eRREPMAINTCFRACTION, eRPCSBASE

0.97 0.98 0.99

VATrepair eRREPMAINTCFRACTION, eRPCSBASE

0.98 0.99 0.99

FC eEFfc 0.96 0.97 0.98

COSTfuel eEFfc 0.96 0.97 0.97

TAXfuel eEFfc 0.96 0.97 0.97

VATfuel eEFfc 0.96 0.97 0.97

COSTlabour RLABOURC 0.96 0.96 0.96

COSTlabourtax RLABOURTX 0.96 0.96 0.96

TAXrest PUBLICCOSTCOV 0.96 0.96 0.96

VATrest PUBLICCOSTCOV 0.96 0.96 0.96

PM eEF 0.97 0.96 0.96

COSTpurchase eRPCSBASE 0.97 0.95 0.95

TAXownership ROWNTX 0.96 0.95 0.95

COSTinsurance RINSCFRACTION 0.94 0.94 0.95

TAXinsurance RINSCFRACTION 0.94 0.94 0.95

COSTrest PUBLICCOSTCOV 0.97 0.96 0.95

NOx eEF 0.96 0.95 0.95

VATpurchase eRPCSBASE 0.97 0.94 0.94

TAXregistration uparaBT 0.89 0.92 0.92

CO eEF 0.91 0.92 0.91

VOC eEF 0.93 0.91 0.91

Costs eRPCSBASE, eEFfc 0.88 0.88 0.88

Vehicles eRPCSBASE, eEFfc 0.89 0.88 0.88

Vehkms eRPCSBASE, eEFfc 0.89 0.88 0.88

- The hot emission factors (eEF) influence most the variance of the emissions (VOC,

NOX, PM, CO) while the basic road vehicle purchase resource cost (eRCPSBASE)

controls the variability of the stock and activity variables (vehicles and vehicle-kms).

On the other hand, many input factors are responsible for the variability of the cost

related output.

- All model outputs exhibit high linearity to input variables. The least amount of

explained-by-single-contributions variance estimated was 88% and corresponds to the

output variables Costs, Vehicles, and Vehkms. Only the remaining fraction depends on

higher order interdependencies between the input variables.

13

- The linearity of the output variables is generally constant in time. On the other hand,

the total effects are either constant or decreasing in the future. This implies that the

reduction of the variance of single input variables will be always effective in decreasing

the uncertainty of the estimates.

- Most of the interaction effects were observed for TAXregistration and these are

attributed to interdependencies of all input variables. Least dependencies on second

and higher order terms have been observed for COSTpurchase and VATpurchase. This

gives the opportunity to work only on the uncertainty of the RPCSBASE in the future,

in order to reduce the variability of the purchase cost related items.

Scenario uncertainty

The uncertainty of three alternative scenarios was quantified in order to understand the

uncertainty and sensitivity of the model in simulated applications of policy impact analysis. The

three scenarios were formulated in such way as to potentially activate different paths of

uncertainty of the model. The three scenarios were:

1. Increase the ownership tax of passenger cars to demonstrate mostly shifts to other

modes of transport within the road sector.

2. Increase the road fuel price to demonstrate general drop in road transport activity.

3. Introduce a new emission standard (Euro VI heavy duty vehicles) to demonstrate drop

in total emissions.

Those simulations led to the following observations:

In Scenario 1, providing a much higher ownership tax for passenger cars (three times higher

than the basecase in 2030) affects fifteen out of the 24 output variables, mainly the cost-

related ones. The substantial increase in ownership costs increases total road transport costs

by 2.4% and this leads to an almost equal decrease in vehicle number and vehicle kilometres

of passenger cars. Despite the high increase of car operation costs, no substantial intermodal

shifts between the road vehicles or between road transport and other modes were observed.

The main impact of the cost increase was a proportional drop in the activity of passenger cars.

In Scenario 2, the base cost of road fuel was assumed to range within ±30% of its mean value

from 2010 onwards. The mean fuel price did not change in the scenario compared to the

basecase. This variation only affected the confidence interval of the fuel cost output variable in

a statistically significant manner. The fuel tax is independent of fuel cost. Interestingly, the

VAT of fuel was not seen to vary significantly between the two cases. This was because the

VAT is applied on the (fuel cost + fuel tax) value and the constant fuel tax range attenuated

the impact of the larger fuel cost variability. The confidence intervals for all other variables

were only marginally affected. Due to the small relative effect, the sensitivity analysis produces

identical results between the basecase and the scenario, i.e. the output depends on the

uncertainty of the input variables in the same fashion as in the baseline.

14

Estimated emission reductions in PM and NOx together with increased purchase, operation

costs and – marginally – increased fuel consumption values were considered for the

introduction of a heavy duty Euro VI emission standard in Scenario 3. The effect of the

introduction was only shown for PM and NOx and for no other output variable. The contribution

of input variables to the uncertainty of the scenario is identical to the basecase as the

coefficient of variance of the Euro VI emission factors has been assumed the same as Euro V.

Major conclusions and recommendations

The analysis conducted in this study demonstrated that a Monte Carlo analysis of TREMOVE is

a useful tool to characterise its uncertainty and sensitivity. Based on this analysis, a number of

conclusions may be drawn:

1. A limited number of input variables (14) seems to drive the total model uncertainty. In

addition, several output variables can be approximated as linear combinations of input

variables with a small loss in precision. This is probably due to the limited elasticity in

shifts between different modes of transport and vehicle types offered by the demand

module. If this limited flexibility is validated (see point 6 in this list), then it can be

suggested that several model operations can be simplified with a beneficial effects on

model transparency and processing time.

2. The fact that a limited number of variables is important for most of the model output

uncertainty means that better quality / more precision in the estimates of these

specific variables will reduce the uncertainty of the output. Of particular importance

appear to be the emission factors, the purchase cost of vehicles, the parameters

defining the scrappage probability, the parameters used to estimate the residual cost

when a vehicle is scrapped and cost-related parameters (maintenance, insurance,

ownership, labour).

3. This study was limited to one country only (UK). Given the linear behaviour of the

model and the limited sensitivity of the demand to the input variables uncertainty,

extending the analysis to other countries does not seem to offer new insights. This

might affect the numerical values of the uncertainty indicators produced but would not

change the conclusions of the study.To improve the model output priority should

rather be given in improving the quality of the major input variables identified in this

study.

However:

4. The total uncertainty of the projection, taking into account macroeconomic and

demographic factors may be realistically assessed only by introducing alternative

basecase projections in the model. This can be a useful future activity.

5. Our analysis only took into account the variables and parameters inclusive in the

model. Expanding the model to cover additional vehicle types, such as alternative fuel

vehicles, hybrids, plug-in hybrids, and electric vehicles may increase the uncertainty of

the estimates but is deemed necessary to cover future applications of the model.

Finally the following recommendations spur from the analysis carried out:

15

6. The currently (2010-2011) changing environment in Europe due to the financial and

credit crisis offers several opportunities for validation of key model elasticities. The

model could be applied to simulate effects of increasing fuel taxation, ownership

taxation, scrappage activities, etc., that take place today in several countries, and

compare with real-world trends.

7. A follow up activity could be to derive the linear functions between output data and

input variables and compare how much they deviate from TREMOVE output. This could

serve three purposes: (i) Quantify how much TREMOVE output deviates from linear

behaviour, (ii) have a simplified TREMOVE model to easily perform scenarios for which

maximum accuracy is not necessary, (iii) identify areas where TREMOVE structure

could be simplified without loss of precision.

17

1 Introduction

The European Commission awarded the contract for estimating the uncertainty of the

TREMOVE model output with respect to changes in model input data to EMISIA SA. This final

report summarizes all activities that was conducted in the framework of the project, presents

the results of the calculations and offers recommendations for further improvement of

TREMOVE in terms of uncertainty calculations.

1.1 Background

Road-transport is a significant source of air pollution and greenhouse gas emissions. According

to European Environment Agency data, road transport is responsible for 22.4%, 39.8%, 42.7%

and 16.2% of total CO2, NOx, CO and PM10 emitted in the EEA32 territory. Since many years

the EC has been defining and implementing policies which aim at organising and controlling

transport in such a manner that it serves its purpose with minimum impacts.

In parallel, the European Commission and the Council have set forward a Directive which sets

National Emission Ceilings (NECD – 2001/81/EC) to regulate the total amount of pollutants that

can be produced annually in each country, which targeted year 2010. The Commission is also

working on the revision of this directive, which would target year 2020 and the inclusion of

PM2.5 ceilings.

Key regulations in the European Commission need to be accompanied by detailed impact

assessments (http://ec.europa.eu/governance/impact/index_en.htm), i.e. studies which

provide an informative assessment of the impacts and the costs that the different policy

options are associated with. In the transport sector, the TREMOVE model [1] has been used to

facilitate impact assessments. TREMOVE is a policy assessment model developed by Transport

and Mobility Leuven (TML). The initial sources of TREMOVE are the models Trenen, Foremove

and COPERT. It covers EU27+CH, HR, NO, TR for the period 1995-2030.

TREMOVE is a model consisting of three main modules: a demand, a stock, and an emissions

module. These are accompanied by two additional modules, the well-to-tank and the welfare

modules. These two are add-ons on the main structure of the model, aimed at estimating the

upstream (fuel production) costs of transport and the benefit (in monetary terms) of emission

reduction to the society, respectively. Well-to-tank and the welfare modules are post-

calculations on the TREMOVE main output (activity, costs, emissions and consumption).

In TREMOVE, a baseline projection is the first development. This is inherent to the model

version and includes a number of exogenous parameters (demand and GDP projections, fuel

and operation costs, vehicle stock mix, etc.). Once the baseline has been built, the model is

calibrated around this baseline. This means that all model functions are calibrated so that the

model produces the desired output (baseline demand and emissions) when fed with the

assumed costs (baseline costs). After the baseline has been agreed and the model is

calibrated, it is then ready for scenario execution.

18

The basic idea in running a scenario is to feed in the marginal costs and the technological

impacts of each policy scenario considered. Costs may be associated with fuel price, purchase

expenses, taxation, operation and maintenance expenses, etc. Higher costs lead to a lower

demand than the baseline and vice-versa. Demand then drives vehicle sales and this gears the

vehicle stock replacement. After the technology mix has been determined, the emission module

will calculate emissions. The complete calculation is not executed in one loop. The new stock

and the, presumably, new fuel consumption calculated, after changing the cost assumptions,

also lead to differentiated total cost calculation. The initial cost assumptions and the costs

calculated by the model need therefore to be equilibrated by performing some internal software

loops.

A large number of modelling assumptions, parameters, and variables are required to simulate

the whole chain of events, from costs to demand, distribution of demand to vehicle classes and

then estimate of emissions and consumption of all individual vehicle classes. Currently, all

model outputs are produced in a deterministic manner, i.e. only one result is possible with a

given input. However, one may expect that the reality is more complicated than that. Due to

uncertainties in the estimation of input data and, naturally, all modelling parameters, the

model output is also bound to an uncertainty range. That is, there is a range of variation in the

modelling output induced by the uncertainty in the estimation of the input and modelling

approach. The first target of this study was to characterize the uncertainty of the main model

outputs (costs, activity, emissions and consumption). Such a characterization is important in

order to understand what level of variation is examined in a modelling output, i.e. how much

could reality deviate from the values calculated by the model. This information may then be

used to judge whether two different scenarios lead to statistically different results and/or if a

scenario (policy decision) will lead to a statistically different value than the baseline.

Once the uncertainty in the output has been established, one will wonder how this can be

reduced. This question can be answered by running a sensitivity analysis which identifies the

terms which induce most of the variance in the modelling output. A large variance of the

output may be induced either by a term which is largely unknown and hence associated with a

large uncertainty range or by a term which has a large impact on the modelling output. In the

latter case, even a small perturbation in its value is largely magnified in the modelling output

(non linear term). Finally, uncertainty may be induced by interrelations between the different

model terms, which designate a second-order effect. The identification of the terms which are

most important in the modelling output and better estimates of their values would then greatly

benefit the reliability of the model output. Such a sensitivity analysis was the second target of

this study.

Uncertainty and sensitivity analysis of the TREMOVE model is a complicated procedure.

TREMOVE is written in GAMS code and uses a number of different files for the execution of a

run. The main code is included in GAMS (.gms) files. All the calculations performed by the

model can be found in these files. The structure of the files is based on a “subroutine” format,

where the code in each file is read/called from another “parent” file. Input variables and

parameters can be found in ‘include’ (.inc) files. They are actually text files, using a different

file extension. To transfer data between the different modules GAMS uses .gdx files which can

be created and later read by the GAMS code more efficiently compared to MSAccess or MSExcel

files. Calculated data are then exported to MSAccess and MSExcel files since users are more

familiar with such data structure. Additional files, such as compressed (.zip) files and batch

19

(.bat) files are also used by the model to initialise the run and complete the calculations. For

the purposes of this project only the .gms and .inc files were studied and modified.

1.2 Objectives of the study

The main objectives of the study were to:

� Characterize the uncertainty associated with input data to the model and the modelling

terms. This required identification of the main input variables and key model

parameters and collection of literature data on the variance associated with them.

� Characterize the uncertainty of the model output, in terms of the activity, costs,

emissions and fuel consumption calculated.

� Examine the uncertainty of the baseline projection and three separate scenarios. The

scenarios should cover a large range of situations in order to examine what is the

uncertainty of the model in a range of situations.

� Run a sensitivity analysis to identify which are the main modelling terms and input

data that can be considered responsible for the uncertainty in the output. The analysis

should identify both linear effects, non-linear effects, and second-order effects.

1.3 Structure of this report

This is the final report of the project summarizing all activities and presenting the results of the

uncertainty and sensitivity analyses conducted on TREMOVE v 3.3.1 (June 2010). Further to

this introductory chapter, the report is structured as follows:

- Chapter 2 discusses the uncertainty range of the input variables and emission factors.

- Chapter 3 outlines the modelling theory and the data parameterisations.

- Chapter 4 summarizes the software modifications required to facilitate the calculations.

- Chapter 5 presents the results of the baseline uncertainty and sensitivity analyses.

- Chapter 6 describes the uncertainty and sensitivity analysis for three scenarios.

- Chapter 7 offers the conclusions and some recommendations derived from this work.

Finally, four Annexes present results in more detail.

20

2 Uncertainty ranges of input variables and modelling parameters

2.1 General

Following initial scoping and analysis work, Emisia supplied LWA with a selection of variables

for which “real world” values were required. Wherever possible, these values were to be

provided with accompanying upper and lower range limits to allow probability distribution

functions to be estimated. In keeping with LWA’s local knowledge it was agreed to supply data

for the UK.

Input variables are not single dimensional but multi dimensional vectors, since their value

depends on the type of vehicle, and year and emission factors also depend on the pollutant

considered. A thorough examination was made for all dimensions of the 33 variables studied in

this exercise.

Some discussions took place regarding the definition of the selected variables and their internal

relationships to the TREMOVE model. GAMS code was accessed regularly to help clarify the

variables’ specification.

Appropriate data sources were identified for as many variables as possible using the general

principle of official Government data as a preference, followed by trade and industry data,

followed by other public domain information sources.

Specific reference was made to data and publications from:

o UK Government Department for Transport,

o UK Government Office of National Statistics,

o UK Government Revenue and Customs,

o UK Driver and Vehicle Licensing Agency,

o UK Vehicle Certification Agency,

o UK Petrol Industry Association (2009 data)

o Society of Motor Manufacturers and Traders, (2009 data)

o Road Haulage Association, (2009 data)

o Automobile Association (AA) (2009 Data)

o Royal Automobile Club (2008 data)

o European Union Directorates

o European Central Bank

o Selected vehicle manufacturers

o Relevant research organisations

o Other public domain information sources such as press articles and expert internet forums

Where necessary the collected data was recalibrated and formatted to be compatible with the

units used by the TREMOVE variables. In cases where no external data source could be

identified for a particular variable then, where possible, expert judgment on likely ranges and

21

modelling approaches was made. In a small number of cases neither data nor expert judgment

was possible and an arbitrary variance range was selected to examine the sensitivity of the

model output to the variable variance.

Finally TREMOVE generated data for the UK was used for comparison with the LWA identified

data. In some cases where alternative data sources were not available TREMOVE UK data was

used with upper and lower limits based on experience with “real world” data.

2.2 Input variables description

Table 1 presents the input variables for which the uncertainty was studied. There are 4

categories of information for each parameter, the name used in the model, a short description

of the parameter, information on the way the parameter was varied (standard deviation used,

the assumption behind this decision or other relevant information) and if this parameter was

finally modified or not. Some of these variables either did not have an uncertainty range

associated with them or it was selected to modify them during scenario executions.

Table 1: Input variables studied to calculate the TREMOVE output uncertainty.

Name Description Justification Varied

FUEL_ENERGY_DENSITY Fuel energy density - GJ per kg

Y

FUELSPEC Fuel specification history Y

LTRIP Average estimated trip length – km

Y

paraB

b - parameter in TRENDS detailed report 1 : Road transport module page 15 - ie characteristic service life

Y

paraT

T - parameter in TRENDS detailed report 1 : Road transport module page 15 - ie failure steepness

Y

PUBLICCOSTCOV Public transport fare cost coverage

Y

RFACTORUNCONV

Ratio fuel consumption unconventional vs equivalent conventional vehicle - [(kg/km) / (kg/km)]

Y

RFC_REDUC_RESISTANCE

Real world fuel consumption reduction from utilisation of technologies to reduce vehicle and engine resistance factors – percentage

Y

RFCairco

Extra fuel consumption from use of air-conditioning equipment - litre per km

Y

RFUEL_COMPOSITION Average share of components in blended fuels - % in weight

Y

RHC Ratio of hydrogen to carbon atoms in fuels

Y

RINSCFRACTION

Insurance cost as percentage of vehicle purchase resource cost - %

Y

22


RLABOURC Labour cost - net wage - for truck drivers - EURO per hour

Y

RLABOURTX

Labour tax - bruto wage minus netto wage - for truck drivers - EURO per hour :

Y

RLOADCAP Average maximum loading capacity big truck - tonne

Y

RLOGITCNGAVAIL

Relative availability of CNG in fuel stations [# fuel stations with CNG / # cars in fleet]

Y

RLOGITPACC

Acceleration for big and medium car logit - seconds to 100 km per hour

Y

RLPG_FIT_COST Resource cost to retrofit LPG installation - EURO per vehicle

Y

RMILage Relative annual mileage as a function of vehicle age - %

Y

RMILnew

Average annual mileage of new cars in each year - exogenous estimate - vehicle kilometres per year

Y

ROWNTX Annual Ownership tax road vehicles - EURO 2005

Y

RPCS_BASE Road vehicle basic purchase resource cost - EURO 2005

Y

RPCS_INCREASE_2009 % Vehicle purchase cost increase to reach the 140g car CO2 target in 2009

Y

RPCS_INCREASE_2012

% Vehicle purchase cost increase to reach the car CO2 target in 2012 - on top of 140g costs

Y

RREPMAINTC_INCREASE_RTECH_RES

Increase in yearly maintenance cost for using technologies to reduce vehicle and engine resistance factors - EURO 2000

Y

RREPMAINTCFRACTION

Repair and Maintenance Cost excl. taxes as % of purchase resource cost (ex tax)

Y

RSHairco Share of new sold vehicles fitted with air-conditioning - %

Y

RSTNBY

Base year stock of road vehicles per road vehicle type and age - in thousands vehicles

Y

RVP Gasoline volatility (Reid Vapour Pressure) - kPa

Y

SRESIDUALparaA factor in the determination function for residual value

Y

SRESIDUALparaB factor in the determination function for residual value

Y

TMAX Maximum temperature per month - Celsius degrees

Y

TMIN Minimum temperature per month - Celsius degrees

Y

N

23


PUBLICVAT Public transport VAT rate - %

0 in UK

r Annuity interest rate No uncertainty addressed N

Rairco_maintenancefreq Interval between airco maintenance services - years

Expected to have minimal impact on COST calculation and cost is assumed to be implemented in total maintenance cost uncertainty.

N

REDUC_NEW_TECH

Emission reduction percentage for future emission standards relative to latest existing standard

Will be changed in scenario N

RFACTORACEA

COPERT III factor to include historic and projected decrease in car fuel cons following ACEA voluntary 140g agreement - 1.00 for 2002

Reduction factors calculated based on actual historic data (fixed values)

N

RFACTORDIE

COPERT III diesel car fuel consumption factor from ACEA agreement monitoring dB - litre per 100 km


N

RFACTORREAL

COPERT III factor to convert COPERT car fuel cons to ACEA monitoring dB value plus real world factor


N

RFC_ACEA_2002

2002 measured fuel consumption in ACEA agreement monitoring dB - l/100 km - dm³ for CNG


N

RFC_REDUC_GSI

Real world fuel consumption reduction from utilisation of Gear Shift Indicator - %

Can be changed in scenario N

RFCairco_REDUC_SCENARIO

Reduction in real-world airco fuel consumption for policy scenario - %


RFCOST_COMP

Road fuel component resource cost - euro 2000 per litre - except CNG in euro 2000 per m³


RFTAX_COMP

Road fuel component excise tax - euro 2000 per litre - except CNG in euro 2000 per m³


RFVAT Road fuel VAT rate - % No uncertainty addressed N

RINSTXfix Fix annual tax on insurance

0 in UK N

RINSTXrate % tax rate on insurance No uncertainty addressed N

RLOGITPGDP GDP per inhabitant - EURO 1995

GDP is fixed for historic years. GDP can have a big impact for projection years. However TREMOVE is known not to be able to model accurately changes in macro economic indicators. As a result this has been kept fixed and TREMOVE sensitivity in GDP values can be run as a scenario.

N

RMILinc Annual increase of mileage per year for road vehicles - %

0 in UK N

N

24


RPCS_INCREASE_AIRCO_SCENARIO

Absolute vehicle purchase cost increase for airco policy scenario - EURO - 0 in basecase

Will be addressed through the RPCS variable

RPCS_INCREASE_GSI Vehicle purchase cost increase for gear shift indicator - euro 2000


N

RPCS_INCREASE_TPMS

Vehicle purchase cost increase for tyre pressure monitoring system - euro 2000


N

RRegTX Registration Tax new purchased road vehicles - EURO 2005

No uncertainty addressed N

RTECH_GSI_SHARE Percentage of new sold road vehicles equipped with Gear Shift Indicator

Can be changed in scenario (minimal impact expected on uncertainty)

N

RTECH_RESISTANCE_MX

% Vehicles equipped with technologies to reduce vehicle and engine resistance


N

RVAT VAT percentages per road vehicle type in 2000 - %

No uncertainty addressed N

TECHMX Technology distribution matrix - share of new cars fitted with technology (%)

Will be changed in scenario N

A special reference should be made to the cost input variables of the model. TREMOVE

exchanges cost data between the vehicle stock and the demand module. To facilitate the

exchange of the information a model parameter was introduced; the COSTROAD parameter.

This parameter summirises the cost information calculated in the vehicle stock module. In this

study all input data resulting in the population of the COSTROAD parameter as well as all

components of this parameter were investigated. This includes the following variables: vehicle

purchase resource cost, vehicle purchase VAT, vehicle registration tax, vehicle ownership tax,

vehicle insurance cost, vehicle insurance tax, vehicle repair/maintenance cost, vehicle

repair/maintenance tax, fuel resource cost, fuel tax, fuel VAT, driver labour cost, driver labour

tax.

In order to collect all data related to the input variables it was decided to follow a predefined

format to include the necessary information for the uncertainty estimate. This would facilitate

the collection of information, by providing a complete overview of the available data. For this

reason a table (Table 2) was created which contained the following information:

• Name: the name of the variable

• Type: the type of the variable (input or parameter)

• Units: the units of the variable

• Description: a short description of the variable

• Sources: sources used for the uncertainty of the variable

• Comments: general comments on the variable

• Quantification of variability: the actual values used for the uncertainty characterisation

where possible, or alternatively, a short description of it. All actual values used can be

found in Annex III.

• Type of distribution: the shape of the distribution selected for the probability density

functions

25

• Reasoning: information taken into account in order to decide on the type of

distribution

Table 2: Template used to display variable related information.

UK Name: -

Type: - units: -

Description:

Sources:

Comments:

Quantification of

variability (UK):

Type of distribution:

Reasoning:-

-

-

-

-

-

The following tables contain the information on the input variables studied for their uncertainty.

FUEL_ENERGY_DENSITY

UK 5 Name: FUEL_ENERGY_DENSITYType: - units: GJ/kg

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: Equal propability within a small range

Uniform

Fuel energy density

Small variation from IEA 2004 report for Europe [18]

3s=0.03×µ

-

FUELSPEC

UK 6 Name: FUELSPECType: - units: -Description:

Sources:

Comments:

Quantification of

variability (UK):



Uniform

fuel specification history

Specifications are mandated by regulations. Small variability reflects typical refinery output range.

3s=0.02×µ

-

26

LTRIP

UK 7 Name: LTRIPType: - units: km

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: By definition

L-Normal

The average distance travelled by a trip of passenger cars in a country.

Duboudin C., Crozat C., T F., 2002. Analyse de la méthodologie COPERT III. Analyse d’incertitude et de sensibilité, Rapport d’activité remis à l’ADEME par la Société de Calcul Mathématique, SA. En application du contrat n° 01 03 021, Paris, France, p.262.

s = 0.2×µ, according to French COPERT Uncertainty report

In the absence of more detailed data for UK the French uncertainty range has been utilised.

The probability of vehicles to remain in the stock, as a function of their age, can be approached

by a Weibull distribution. In fact, the Weibull distribution provides the survival probability for

each vehicle category with age ϕi(age), and this can be used to calculate the age distribution of

the fleet. This probability is given by the following equation:

+−=

iparaB

i

ii paraT

paraBageage exp)(ϕ where φ(0) = 1 (Eq: 1)

The determination of the probability requires a pair two parameters, paraB and paraT. The two

parameters do not have an exact physical meaning. However, it can be considered that they

approximate the useful life of the vehicle (paraT) and a characteristic (paraB) of the rate by

which the probability decreases. By taking an initial age distribution at a historical year (in our

case: 1995) and by introducing the new registrations per year (vehicles of age 0) and the

Weibull scrappage probability, TREMOVE calculates the age distribution of the vehicles at any

given year.

A more detailed description of the methodology used can be found in the Detailed Report 1 for

the Road Transport Module of the Project “Development of a Database System for the

Calculation of Indicators of Environmental Pressure Caused by Transport” (Giannouli et al. [2])

and does not need to be repeated here.

At a second step, the technology split for each country is calculated by applying the technology

implementation matrix of the particular country to the age distribution. The technology

implementation matrix contains the distribution of new registrations of different years to the

various technologies. The central estimate for the age distribution of vehicles of UK was based

on the FLEETS data and the paraB and paraT parameters were calculated on this basis. Then,

an artificial uncertainty range was assigned to the probability function of UK. This artificial

uncertainty is schematically shown in Figure 1. It was in principle assumed that the survival

probability for vehicles with age of five and fifteen years ranges between +/- 10 and +/- 15

percentage units respectively from the central value. Figure 1 shows the original Weibull

27

distribution function for gasoline passenger cars <1.4 l, the range assumed for the uncertainty

of the survival probability, and three alternative curves which fulfil the selected uncertainty

range.

Gasoline PC <1,4 l

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1,0

0 5 10 15 20age

φ(a

ge

)G<1,4Alt1Alt2Alt3

Figure 1: Weibull distribution function of probability as a function of age. Three alternative solutions

that fulfill the artificial uncertainty introduced (example: Gasoline cars <1.4 l).

By using the above methodology a number of paraB and paraT pairs were calculated for each

vehicle category, that fulfilled the uncertainty range introduced. From these pairs, 100 were

finally selected by sampling percentiles from the joint probability distribution function of paraB

and paraT. They served as data pool providing each time the required couple of values used for

the calculations.

paraB

8 Name: paraB- units: Coefficient

See text

Uniform

b - parameter in TRENDS detailed report 1 : Road transport module page 15 -

i.e. characteristic service life

Neither data nor commentary on suggested values was possible.

It was in principle assumed that the survival probability for vehicles with age of

five and fifteen years ranges between +/- 10 and +/- 15 percentage units

respectively from the central value.

-

28

paraT

9 Name: paraT- units: Coefficient

See text

Uniform

T - parameter in TRENDS detailed report 1 : Road transport module page 15 -

i.e. failure steepness


It was in principle assumed that the survival probability for vehicles with age of

five and fifteen years ranges between +/- 10 and +/- 15 percentage units

respectively from the central value.

-

PUBLICCOSTCOV

UK 10 Name: PUBLICCOSTCOVType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: Typical uncertainty distribution

Normal

Cost coverage of public transport fares

Department of Transport UK - Annual Bus Statistics 2009/10 Department of Transport - options for reform March 2008.

3s=0.2×µ

-

RFACTORUNCONV

UK 18 Name: RFACTORUNCONVType: - units: (kg/km) / (kg/km)

Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Ratio fuel consumption unconventional vs equivalent conventional vehicle

-

For Buses the value will be between 0.75 and 0.9, for Passenger cars between 0.81 and 0.85

-

29


UK 21 Name: RFC_REDUC_RESISTANCEType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):



Uniform

Real world fuel consumption reduction from utilisation of technologies to reduce

vehicle and engine resistance factors

-

Range will be between µ+-0.01

-

RFCairco

UK 22 Name: RFCaircoType: - units: l/kmDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:

This will depend on combination of parameters (type of vehicle, ambient conditions, driving pattern etc) so normal distribution seems appropriate

Normal

Extra fuel consumption from use of airconditioning equipment

Large range reflecting the uncertainty in additional fuel consumption from A/C use, due to vehicle type, ambient conditions, driving conditions.

3s=0.5×µ

-

RFUEL_COMPOSITION

UK 26 Name: RFUEL_COMPOSITIONType: - units: % in weight

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:Composition limits defined by regulations. Normal distribution expresses typical

refinery output effect.

Normal

Average share of components in blended fuels

UKPIA (UK Petrol Industry Association - 2009) data supplied

Six "real world" values quoted for UK for the years 2008-9 through 2013-14.

There may be future changes dependent on the standards, suggested range

3s=0.3×µ. In total pure unblended diesel and pure unblended petrol will be

calculated as the remaining fuel in use.

The oil industry is adding biofuels to road fuels under the Renewable Transport

Fuel Obligation (RTFO), of 2.5% by volume in 2008/9, 3.25% in 2009/10, 3.5%

in 2010/11, 4% in 2011/12, 4.5% in 2012/13 and 5% in 2013/14

30

RHC

UK 28 Name: RHCType: - units: -

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:

Hydrocarbon species in fuel can vary within small range. Normal distribution expresses variation in chemical composition.

Normal

The ratio of atoms of hydrogen over carbon in the fuel molecule

Estimated range based on country submissions through COPERT inventories.

Similar uncertainty for both Gasoline and Diesel. The ratio is expected to vary from 1.8 to 2.1, therefore, 3s = 0.15

Typically 1.8-2.1

RINSCFRACTION

UK 29 Name: RINSCFRACTIONType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Insurance cost as percentage of vehicle purchase resource cost

Automobile Association and Road Hauliers Association

The HDV data was from the RHA. It was possible to prepare a selection of samples for passenger cars.

Notes: Assumptions are in the UK AA report that motorists will benefit from an average 60% discount on the full price of insurance. It is worth noting a considerable variation is possible with different underwriters/providers. One example in 2010 gave a range of between £470 and £750 with a mean of £636. This gave a range of ratio of between 0.016 - 0.0214. There are a considerable number of variables influencing the price of vehicle insurance: age and experience of driver, male/female, age and value of vehicle, model of vehicle (currently 20 categories), security of vehicle, address where it is kept, cost of parts (?imports).

RLABOURC

UK 32 Name: RLABOURCType: - units: Euro/hourDescription:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Labour cost - net wage - for truck drivers

Road Hauliers Association and Her Majesty's Revenue & Customs (HMRC)

3s=0.3×µ

The nett wage for truck drivers appears significantly less than shown in the model results.

31

RLABOURTX

UK 33 Name: RLABOURTXType: - units: Euro/hourDescription:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Labour tax - bruto wage minus netto wage - for truck drivers

Road Hauliers Association and Her Majesty's Revenue & Customs (HMRC)

3s=0.3×µ

This delta in the UK is much smaller than already shown in the mode results.

RLOADCAP

UK 34 Name: RLOADCAPType: - units: tonneDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:Average values of loading capacity are to be expected, normal distribution expresses typical uncertainty of standard value.

Normal

Average maximum loading capacity big truck

Typical uncertainty range for average payload of trucks in Europe.

3s=0.2×µ

-

RLOGITCNGAVAIL

UK 35 Name: RLOGITCNGAVAILType: - units: Coefficient

Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Relative availability of CNG in fuel stations [# fuel stations with CNG / # cars in fleet]

Note the distinction between CNG and LPG and the variable usage and supply situation across Europe. Neither data nor commentary on suggested values was possible.

3s=0.2×µ

-

32

RLOGITPACC

UK 36 Name: RLOGITPACCType: - units: seconds

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:Acceleration within vehicle class varies within a small normally distributed range.

Normal

Acceleration for big and medium car logit - seconds to 100km per hour

Data sourced from What Car publication July 2010.

UK "real world" values for parameters were provided, with accompanying upper and lower range limits to allow probability distribution functions to be estimated. Where "real world" values were not possible a pragmatic decision was made to utilise the existing TREMOVE output for the UK and apply upper and lower limits based on experience with "real world" data.

Representative vehicle data was combined and statistics applied according to TREMOVE categories small, medium and large diesel, and small medium and large gasoline. No data available for CNG vehicles, gasoline values used.

RLPG_FIT_COST

UK 38 Name: RLPG_FIT_COSTType: - units: EuroDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:

Expected minimun effect on calculations. Uniform distribution better expresses variation between vehicle types and retrofitting stations prices.

Uniform

Resource cost to retrofit LPG installation

Market information

Costs vary between 1800 and 2500 Euro.

-

RMILage

UK 39 Name: RMILageType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: See text

Uniform

The decrease of the annual mileage relative to vehicle age

Annual mileage per vehicle type has been found from national statistics on

mobility.

Data aquired from the FLEETS project

RMILage for a brand new vehicle (age=0) equals 1. For a vehicle of 40 years of

age this value could be as low as below 0.1.

33

RMILnew

UK 41 Name: RMILnewType: - units: km/year

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: See text

-

Average annual mileage of new cars in each year - exogeneous estimate

-

-

The effect of mileage on total uncertainty will be addressed through the

uncertainty of the RMILage variable.

The calculation of the average annual mileage driven per vehicle (RMIL) in a particular vehicle

technology is a function of the annual mileage of a new vehicle (RMILnew) and a correction

function for the effect of vehicle age (RMILage). The decrease of annual mileage with age has

been approached by a Weibull function. This reflects the fact that new cars are driven more

than old ones. The shape of the curve is considered to be a good approximation of the actual

shape of the mileage reduction with age. An example of actual mileage degradation with age,

which is based on recordings of Inspection and Maintenance data from the Italian passenger

car fleet is shown in Caserini et al. [16]. It is evident that the curves flat out after some years.

The equation of the Weibull function used is given in (Eq: 2). The modelling parameters (bm,

Tm) and RMILnew are specific to country and vehicle subsector considered.

Figure 2: Annual mileage as a function of vehicle age for the Italian passenger car fleet.

Source: (Caserini et al. [16]).

34

bm

Tm

bmageRMILage

+−= exp

(Eq: 2)

RMILnewRMILageRMIL ⋅= (Eq: 3)

The uncertainty in the calculation of the RMIL parameter originates from the uncertainty in bm,

Tm and RMILnew. This uncertainty will be addressed through the uncertainty of the dependency

of the mileage to the vehicle age (RMILage), thus the mileage of a new vehicle (RMILnew) will

not be explicitly modelled.

The RMILage is assumed to range between a minimum and a maximum. These boundaries are

defined from the extents of the functions of all countries that submitted such detailed data in

the framework of the FLEETS project (Ntziachristos et al. [3]). These extents, for the example

of gasoline passenger cars of <1.4 l are shown in Figure 3. It was therefore assumed in our

case, that RMILage can receive any value within these two boundaries. We then calculated all

(bm, Tm) pairs that satisfied this limitation. With this procedure, a large number of bm and Tm

couples were derived, different for each vehicle category. From these pairs, 100 were finally

selected by sampling percentiles from the joint probability distribution function of bm and Tm.

They served as data pool providing each time the required couple of bm and Tm used for the

calculations.

PC Gasoline <1,4l

0,0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1,0

0 10 20 30 40age

φ (

age

)

minmaxAlt1Alt2Alt3

Figure 3: Range for the RMILage variable (example of passenger cars <1.4l) and three examples of

bm and Tm functions (Alt1 through 3) fulfilling the selected criteria (min and max).

35

ROWNTX

UK 42 Name: ROWNTXType: - units: Euro 2005Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:Value differs normally according to vehicle and driver characteristics.

Normal

Annual Ownership tax road vehicles

Driver and Vehicle Licensing Authority (DVLA)

"Real world" data supplied for light duty trucks, motorcycles and mopeds for the year 2010.

The UK system for annual licensing (road tax) has operated a scale based on CO2 emissions and not engine size, making the correlation for passenger cars very difficult. For HGV's , the taxation system again is based on number of axles and weight, again not easy correlation for this exercise.

RPCS_BASE

UK 43 Name: RPCS_BASEType: - units: €Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Road vehicle basic purchase resource cost

Car data sourced from What Car - edition July 2010. HDT data from the Road Haulage Association and Motorcycle data from the Royal Automobile C lub - 2008


The passenger cars were categorised as per TREMOVE and average values obtained. In the absence of data for Buses, LTD's and VAN's, the TREMOVE UK output was quoted using the spread of values as for the 32t HDT. For motorcycles, (MC2-4), data was extracted from the RAC publication for 2010. For Mopeds and MC1, again TREMOVE Uk data was suggested using the same spread as for the MC2-4 group.

RPCS_INCREASE_2009

UK 44 Name: RPCS_INCREASE_2009Type: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Vehicle purchase cost increase to reach the 140g car CO2 target in 2009

Typical range assumed to account for manufacturer to manufacturer variability.

3s=0.2×µ

-

36

RPCS_INCREASE_2012

UK 45 Name: RPCS_INCREASE_2012Type: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Vehicle purchase cost increase to reach the car CO2 target in 2012 - on top of 140g costs

Typical range assumed to account for manufacturer to manufacturer variability.

3s=0.2×µ

-


UK 50 Name:RREPMAINTC_INCREASE_RTECH_RES

Type: - units: Euro 2000

Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Increase in yearly maintenance cost for using technologies to reduce vehicle and engine resistance factors

Assuming additional cost of low resistance tires of 50-100 euros per 4 tire set, and a lifetime of 3.5 years. Assuming 15 euros per liter of oil, consumption of 6.5 liters per year and additional cost of low friction oil of 10-30%.

For the LRRT the valua will range from 15-30 euros and for the LV from 10-30.

-

RREPMAINTCFRACTION

UK 51 Name: RREPMAINTCFRACTIONType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Repair and Maintenance Cost excl. taxes as % of purchase resource cost (ex tax)

Value is a fraction of purchase price, therefore takes into account general increase of maintenance cost with retail price. The range assumes typical variation of maintenance cost per manufacturer.

3s=0.3×µ

-

37

RSHairco

UK 52 Name: RSHaircoType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):



Normal

Share of new sold vehicles fitted with air-conditioning

Data was extracted from ABOUT publishing "Global Market for Automotive Aircon 2004".


Data from the ABOUT report was analysed to give values and ranges for vehicles in each TREMOVE sub category. The car groupings were not identical to TREMOVE and sensible assumptions were made to the groupings.

RSTNBY

UK 53 Name: RSTNBYType: - units: no vehicles

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Base year stock of road vehicles per road vehicle type and age

-

Uncertainty estimated through the uncertainty of the paraB and paraT variables.

Depending on the paraB and paraT value a different scrappage rate is calculated

and as a result the distribution of vehicles according to age is varied in order to

meet the total demand.

-

RVP

UK 57 Name: RVPType: - units: kPa

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning:RVP limits defined by regulations. Normal distribution expresses typical refinery

output effect.

Normal

The vapour pressure of gasoline (defined by a test at 38 oC). The vapour

pressure is a measure of the fuel volatility. The higher the vapour pressure, the

easier the fuel evaporates at a given temperature. The vapour pressure is

important to calculate NMVOC emissions due to evaporation losses. These are

only relevant for gasoline, due to the low volatility of the diesel fuel.

The maximum RVP is defined by the regulations. Some detailed data on RVP for

different countries and relevant information and sources may be found in Hill N.

[19].

Limited uncertainty expected, as fuels are centrally produced and the refineries

need to follow the regulations. Assumption 3s = 0,05×µ

The typical range is 70 kPa (summer grade) to 110 kPa (winter grade).

38

SRESIDUALparaA

UK 58 Name: SRESIDUALparaAType: - units: Coefficient

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: See text

Uniform

estimation residual value function as a percentage of purchase cost


It was in principle assumed that the residual value for vehicles with age of five and fifteen years ranges between +/- 10 and +/- 15 percentage units respectively from the central value.

-

SRESIDUALparaB

UK 59 Name: SRESIDUALparaBType: - units: Coefficient

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: See text

Uniform

estimation residual value function as a percentage of purchase cost


It was in principle assumed that the residual value for vehicles with age of five and fifteen years ranges between +/- 10 and +/- 15 percentage units respectively from the central value.

-

The calculation of the residual value (SRESIDUAL) for a particular vehicle is a function of the

purchase value of a new vehicle (RPCS_base) and a correction function for the effect of vehicle

age. The decrease of the vehicle value with age has been approached by an exponential

function. This reflects the fact that cars lose value as they are getting older.

This function is given by the following equation:

( )[ ]1exp)( −⋅⋅= agearaBSRESIDUALparaASRESIDUALpageRES i (Eq: 4)

The function uses two parameters, (SRESIDUALparaA and SRESIDUALparaB). The two

parameters do not have an exact physical meaning.

As a result the residual value (SRESIDUAL) can be calculated by the following equation:

baseRPCSageRESSRESIDUAL i _)( ⋅= (Eq: 5)

To calculate these two parameters (SRESIDUALparaA and SRESIDUALparaB) the central

estimate for the residual value of vehicles of UK was in TREMOVE was used. Then, an artificial

uncertainty range was assigned to the probability function of UK. This artificial uncertainty is

schematically shown in Figure 4. It was in principle assumed that the residual value for vehicles

with age of five and fifteen years ranges between ±10 and ±15 percentage units respectively

from the central value. Figure 4 shows the original function for gasoline passenger cars <1.4 l,

39

the range assumed for the uncertainty of the residual value, and three alternative curves which

fulfil the selected uncertainty range.

Residual value

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0 5 10 15 20 25 30 35 40

Vehicle age

%

Tremove values

Alt1

Alt2

Alt3

Figure 4: Residual value as a function of age. Three alternative solutions that fulfill the artificial

uncertainty introduced (example: Gasoline cars <1.4 l).

By using the above methodology a number of SRESIDUALparaA and SRESIDUALparaB pairs

were calculated for each vehicle category that fulfilled the uncertainty range introduced. From

these couples, 100 were finally selected by sampling. They served as data pool providing each

time the required couple of values used for the calculations.

TMAX

UK 61 Name: TMAXType: - units: oC

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: Effect of nature

Normal

The average of the maxima in daily temperature for a duration of a month. This maximum temperature is required as input to both evaporation and cold-start calculations. For countries with significant temperature differences over their area (e.g. south and north), the temperature should correspond to the average (possibly weighted average) of areas where most of the traffic is located.

National meteorological insitutes and internet databases (i.e. www.weatherbase.com).

An uncertainty range required to cover national differences between north and south. 3s=3oC

Month specific. Average min temperature ranges between 8 to +22.

40

TMIN

UK 62 Name: TMINType: - units: oC

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: Effect of nature

Normal

The average of the minima in daily temperature for a duration of a month. This minimum temperature is required as input to both evaporation and cold-start calculations. For countries with significant temperature differences over their area (e.g. south and north), the temperature should correspond to the average (possibly weighted average) of areas where most of the traffic is located.

National meteorological insitutes and internet databases (i.e. www.weatherbase.com).

An uncertainty range required to cover national differences between north and south. 3s=3oC

Month specific. Average min temperature ranges between 2 to +11.

PUBLICVAT

UK 11 Name: PUBLICVATType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Public Transport VAT rate

http://ec.europa.eu/taxation_customs

No uncertainty was estimated.

In UK there is no VAT on public transport in UK. There is considerable variation across the EU with some MS (e.g. Hungary) charging 25% VAT on Public transport.

r

UK 12 Name: rType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Annuity interest rate

-


Can be changed in scenario

41

Rairco_maintenancefreq

UK 13 Name: Rairco_maintenancefreqType: - units: YearsDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Interval between airco maintenance services - years

-

-

Expected to have minimal imapact on COST calculation and cost is assumed to be implemented in total maintenance cost uncertainty.

REDUC_NEW_TECH

UK 14 Name: REDUC_NEW_TECHType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Emission reduction percentage for future emission standards relative to latest existing standard

-



RFACTORACEA

UK 15 Name: RFACTORACEAType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

COPERT III factor to include historic and projected decrease in car fuel cons following ACEA voluntary 140g agreement - 1.00 for 2002

-

-


42

RFACTORDIE

UK 16 Name: RFACTORDIEType: - units: l/100 km

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

COPERT III diesel car fuel consumption factor from ACEA agreement monitoring dB

-

-


RFACTORREAL

UK 17 Name: RFACTORREALType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

COPERT III factor to convert copert car fuel cons to acea monitoring dB value plus real world factor

-

-


RFC_ACEA_2002

UK 19 Name: RFC_ACEA_2002Type: - units: l/100 km

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

2002 measured fuel consumption in ACEA agreement monitoring dB

-

-


RFC_REDUC_GSI

UK 20 Name: RFC_REDUC_GSIType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Real world fuel consumption reduction from utilisation of Gear Shift Indicator

-



43

RFCairco_REDUC_SCENARIO

UK 23 Name: RFCairco_REDUC_SCENARIOType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Reduction in real-world airco fuel consumption for policy scenario

-



RFCOST_COMP

UK 24 Name: RFCOST_COMPType: - units: Euro/litleDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Road fuel component resource cost

-



RFTAX_COMP

UK 25 Name: RFTAX_COMPType: - units: Euro/litleDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Road fuel component excise tax

-



RFVAT

UK 27 Name: RFVATType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Road fuel VAT

Her Majesty's Revenue & Customs (HMRC)

Fixed factor, uncertainty is 0.

VAT has been constant at 17.5% to 2008. It was reduced to 15% in 2009 and

rose to 17.5% in 2010 and is now 20% in 2011.

44

RINSTXfix

UK 30 Name: RINSTXfixType: - units: EuroDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Fix annual tax on insurance

-

-

0 in UK

RINSTXrate

UK 31 Name: RINSTXrateType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Tax rate on insurance

Her Majesty's Revenue & Customs (HMRC)


Taxation on insurance premiums commenced in 1995 and has shown increases over the years to 5% in 2009/10

RLOGITPGDP

UK 37 Name: RLOGITPGDPType: - units: EuroDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

GDP per inhabitant

-

-

GDP is fixed for historic years. GDP can have a big impact for projection years. However TREMOVE is known not to be able to model accurately changes in macro economic indicators. As a result this has been kept fixed and TREMOVE sensitivity in GDP values can be run as a scenario.

RMILinc

UK 40 Name: RMILincType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Annual increase of mileage per year for road vehicles

-

-

0 in UK

45

RPCS_INCREASE_AIRCO_SCENARIO

UK 46 Name:RPCS_INCREASE_AIRCO_SCENARIO

Type: - units: Euro

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Absolute vehicle purchase cost increase for airco policy scenario

-

-


RPCS_INCREASE_GSI

UK 47 Name: RPCS_INCREASE_GSIType: - units: EuroDescription:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Vehicle purchase cost increase for gear shift indicator

-

-


RPCS_INCREASE_TPMS

UK 48 Name: RPCS_INCREASE_TPMSType: - units: Euro

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Vehicle purchase cost increase for tyre pressure monitoring system

-

-


RRegTX

UK 49 Name: RRegTXType: - units: €Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Registration tax new vehicles

Driver and Vehicle Licensing Authority (DVLA)


There is one registration fee for all vehicles of £55 - based on the average € rate for 2009 of 1.1224, equates to €61.73

46

RTECH_GSI_SHARE

UK 54 Name: RTECH_GSI_SHAREType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

New sold road vehicles equipped with Gear Shift Indicator

-

-


RTECH_RESISTANCE_MX

UK 55 Name: RTECH_RESISTANCE_MXType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Vehicles equipped with technologies to reduce vehicle and engine resistance

-

-


RVAT

UK 56 Name: RVATType: - units: %Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

VAT percentages per road vehicle type in 2000

-

-

Fixed values

TECHMX

UK 60 Name: TECHMXType: - units: %

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: -

-

Technology distribution matrix - share of new cars fitted with technology

-

-


47

2.3 Emission factor modelling

The uncertainty of emission factors is a major part of the uncertainty in all transport emission

models, as they constitute the core of the emission calculation. The uncertainty of the emission

factors originates from the variability of the underlying experimental data, i.e. the variability in

the emission level of each individual vehicle which has been included in the sample of vehicles

used to derive the emission factors. A typical range of the variability of individual

measurements for emission factors is shown in Figure 5 for gasoline passenger cars of Euro 3

technology. In TREMOVE, there are two sets of emission factors, the hot ones and the cold-

start ones. The hot emission factors originate from individual measurements of

vehicles/engines mainly conducted in the Artemis project. Some older measurements were

based in previous projects, such as CORINAIR89, COST319, MEET, etc. The uncertainty of old

emission factors was taken from a previous Monte Carlo exercise (Kioutsioukis et al. [5])

conducted in COPERT III. However, emission factors for Euro 1 and later technologies are

solely based on the FP5 Artemis project. The uncertainty of cold-start emission factors was

more difficult to assess, as the values used in TREMOVE and have been transferred from

COPERT 4 are a hybrid of the Artemis and the older CORINAIR methodologies. In the absence

of detailed data and in order not to neglect the contribution of cold start variability, we

assumed that the ratio of standard deviation over mean for the cold emission factors is equal

to the hot ones. This is an approximation which was introduced in the absence of more detailed

data.

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0 10 20 30 40 50 60 70 80 90 100 110 120 130

Speed [km/h]

NO

x E

F[g

r/km

]

MeasurementsBest Fit Curve

Figure 5: Example of variability of individual measurements for the derivation of emission factors.

Gasoline Euro 3 passenger cars. Source: ARTEMIS database.

Table 3 lists the parameters (parameter name and a short description) that are used to

calculate the emission factors and fuel consumption factors in TREMOVE. However these were

not directly modified in the uncertainty calculations. Instead the emission factor and fuel

consumption values were first calculated by TREMOVE and then they were varied at a post-

48

processing procedure by means of an error correction. This modification was included in the

TREMOVE code. A detailed description of the methodology can be found in chapter 3.

Table 3: Emission and fuel consumption input parameters used in TREMOVE.

Name Description Appearance

A1 COPERT III coefficient for hot emission and fuel consumption factors

Emissions


Emissions


Emissions


Emissions


Emissions

A6 COPERT IV coefficient for hot emission and fuel consumption factors

Emissions


Emissions


Emissions

AA0 COPERT IV coefficient for hot emission and fuel consumption factors for hdv

Emissions


Emissions


Emissions


Emissions


Emissions

AMCEUDC COPERT III coefficient for mileage corrections (from LAT/Thessaloniki) - slope scaled by 100 000

Emissions

AMCUDC COPERT III coefficient for mileage corrections (from LAT/Thessaloniki) - slope scaled by 100 000

Emissions

AV_TEMP Average ambient temperature per month - Celsius decrees

Emissions

B0 COPERT III coefficient for hot emission and fuel consumption factors

Emissions


Emissions


Emissions


Emissions

BETAEST1 Cold mileage percentage parameter 1 Emissions




BETAREV COPERT III cold start quicker light-off revision for cold mileage percentage for gasoline EURO II-IV vehicles - 1 for other vehs and techs

Emissions

BMCEUDC COPERT III coefficient for mileage corrections (from LAT/Thessaloniki)

Emissions

BMCUDC COPERT III coefficient for mileage corrections (from LAT/Thessaloniki)

Emissions

C0 COPERT III coefficient for hot emission and fuel consumption factors

Emissions


Emissions


Emissions

CCID COPERT III coefficient for hot emission and fuel consumption factors

Emissions

CMCEUDC COPERT III coefficient for mileage corrections (from Emissions

49


LAT/Thessaloniki)

CMCUDC COPERT III coefficient for mileage corrections (from LAT/Thessaloniki)

Emissions

D1 COPERT III coefficient for hot emission and fuel consumption factors

Emissions


Emissions


Emissions

E0 COPERT III speed limit for hot emission and fuel consumption factors

Emissions

E0c COPERT IV coefficient for hot emission and fuel consumption factors cold E.F. equivalent to E0

Emissions

EFNXPM Non-exhaust PM emission factor for road vehicles Emissions

F0 COPERT IV coefficient for hot emission factors FOR N2O

Emissions

F0c COPERT IV coefficient for cold emission factors FOR N2O

Emissions


Emissions


Emissions


Emissions


Emissions

FF0 COPERT IV coefficient for hot emission factors FOR N2O

Emissions

FF0c COPERT IV coefficient for cold emission factors FOR N2O

Emissions


Emissions


Emissions


Emissions


Emissions

FFF0 COPERT IV coefficient for hot emission factors FOR N2O

Emissions

FFF0c COPERT IV coefficient for cold emission factors FOR N2O

Emissions


Emissions


Emissions


Emissions


Emissions

LC_EMI_FACTOR Non-road fuel and electricity production emission factor - tonne pollutant per tonne fuel or kWh electricity

Emissions

LC_EMI_FACTOR_RFUEL_COMP

Road fuel component production emission factor - tonnes pollutant per tonne fuel produced

Emissions

LIM1 COPERT III speed limit for hot emission and fuel consumption factors

Emissions


Emissions


Emissions


Emissions

LIMTRCH COPERT III temperature limit for cold hot ratio parameters - °C

Emissions

LIMVRCH COPERT III speed limit for cold hot ratio parameters - km per h

Emissions

50


M0 COPERT III coefficient for evaporative VOC emissions Emissions

M1 Driving mode share on evaporation Emissions

M2 COPERT III coefficient - fraction of benzene in NMVOC fraction of exhaust emissions - % weight

Emissions

M3 COPERT III coefficient - fraction of benzene in NMVOC fraction of evaporative emissions - % weight

Emissions

NVFUNC COPERT III coefficient - 1 if not speed-dependent function for hot emission and fuel consumption factors

Emissions

R0COLDFAST COPERT III coefficient for cold to hot ratio Emissions

R0COLDSLOW COPERT III coefficient for cold to hot ratio Emissions

R0WARMFAST COPERT III coefficient for cold to hot ratio Emissions

R0WARMSLOW COPERT III coefficient for cold to hot ratio Emissions









REDUC COPERT III coefficient - emission reduction percentage for future technology

Emissions

REDUC_UNCONV Emission reduction percentage for unconventional vehicle types

Emissions

RHFC134a_IRREGairco Irregular leakage of HFC134a - grammes per airconditioned vehicle year

Emissions

RHFC134a_REGairco Regular leakage of HFC134a - grammes per airconditioned vehicle-year

Emissions

RHFC134a_SALEairco HFC134a emissions of airco installation - grammes per new sold airconditioned vehicle

Emissions

RHFC134a_SCRAPairco HFC134a emissions of airco scrappage - grammes per airconditioned vehicle scrappage

Emissions

RHFC134a_SERVICEairco HFC134a emissions of airco maintenance - grammes per maintenance service

Emissions

SHC Fraction of vehicle categories equipped with emission control

Emissions

SHFI Fraction of vehicle categories equipped with fuel injection

Emissions

SULP_LIM1 COPERT IV coefficient for hot emission factors FOR N2O

Emissions

SULP_LIM1c COPERT IV coefficient for cold emission factors FOR N2O

Emissions

SULP_LIM2 COPERT IV coefficient for hot emission factors FOR N2O

Emissions

SULP_LIM2c COPERT IV coefficient for cold emission factors FOR N2O

Emissions

As described above, four TREMOVE intermediate parameters were used to model the emissions

factors calculation: the hot emission factor, the cold emission factor, the hot fuel consumption

factor and the cold fuel consumption factor. The following tables describe their sources of

uncertainty used to model the emissions module.

51

eEF

UK 1 Name: Hot emission factorType: - units: g/km

Description:

Sources:

Comments:

Quantification of

variability (UK):


Reasoning: Distribution of experimental data

L-Normal

The emission rate of vehicles of a specific technology in g/km, under thermally stabilised engine operation. In COPERT the emission factors are expressed as a function of mean travelling speed. In cases with limited information, emission factors are expressed as a function of the drving mode (urban, rural, highway).

Hot emission factors have been derived from measurements conducted in several research programmes. The most important ones include COST319, FP4 MEET, and FP6 ARTEMIS. Vehicles are driven over specific driving cycles, considered representative of actual driving conditions and the emission level is associated with the mean travelling speed over the cycle. A function is then drawn using regression analysis to associate emission level with travelling speed.

For all pollutants, the uncertainty range has been expressed as standard deviation of the experimental data per 10 km/h speed class intervals.

There is no typical range, as this depends on the uncertainty of the experimental data used to develop the emission factor.

eEF_FC

UK 2 Name: Hot fuel consumption factorType: - units: g/km

Description:

Sources:

Comments:

Quantification of

variability (UK):



L-Normal

The fuel consumption rate of vehicles of a specific technology in g/km, under thermally stabilised engine operation. In COPERT the fuel consumption factor is expressed as a function of mean travelling speed. In cases with limited information, fuel consumption factor is expressed as a function of the drving mode (urban, rural, highway).

Fuel consumption factors have been derived from measurements conducted in several research programmes. The most important ones include COST319, FP4 MEET, and FP6 ARTEMIS. Vehicles are driven over specific driving cycles, considered representative of actual driving conditions and the fuel consumption is associated with the mean travelling speed over the cycle. A function is then drawn using regression analysis to associate emission level with travelling speed.

For the fuel consumption, the uncertainty range has been expressed as standard deviation of the experimental data per 10 km/h speed class intervals.

There is no typical range, as this depends on the uncertainty of the experimental data used to develop the fuel consumption factor.

52

eEFratio

UK 3 Name: Cold-start emission factorType: - units: -

Description:

Sources:

Comments:

Quantification of

variability (UK):



L-Normal

The ratio expressing cold-start over hot emission. Cold-start emissions lead to higher emissions as both the engine and the emission control system have not reached their normal operation temperature.

The over-emission ratio in COPERT has been derived as computed value out of a detailed cold-start study conducted in the framework of FP4 MEET and further elaborated in FP6 ARTEMIS (Andre and Joumard, INRETS report LTE 0509). Since these are computed values, it is difficult to obtain independent (literature) sources to quantify it.

Cold emission factors in COPERT have been produced as a hybrid of the COPERT II, MEET and Artemis methodologies, using approximations to convert the MEET approach (as corrected in Artemis) to older CORINAIR cold-start approach. Cold-start modelling is one of least elaborate items of COPERT 4. As it was not possible to estimate the uncertainty of the emission factors from the uncertainty in the experimental data, we have assumed that the standsrd deviation over mean ecold/ehot is the same with the standard deviation over mean of the hot emission factor. In this way, the contribution of cold-start to uncertainty is estimated in a realistic way.


eEFratio_FC

UK 4 Name: Cold-start fuel consumtpion factor

Type: - units: -Description:

Sources:

Comments:

Quantification of

variability (UK):



L-Normal

The ratio expressing cold-start over hot fuel consumption.

The over-emission ratio in COPERT has been derived as computed value out of a detailed cold-start study conducted in the framework of FP4 MEET and further elaborated in FP6 ARTEMIS (Andre and Joumard, INRETS report LTE 0509). Since these are computed values, it is difficult to obtain independent (literature) sources to quantify it.

Cold emission factors in COPERT have been produced as a hybrid of the COPERT II, MEET and Artemis methodologies, using approximations to convert the MEET approach (as corrected in Artemis) to older CORINAIR cold-start approach. Cold-start modelling is one of least elaborate items of COPERT 4. As it was not possible to estimate the uncertainty of the emission factors from the uncertainty in the experimental data, we have assumed that the standsrd deviation over mean ecold/ehot is the same with the standard deviation over mean of the hot emission factor. In this way, the contribution of cold-start to uncertainty is estimated in a realistic way.


53

2.4 Input variables and parameters not varied

The inception and the interim reports of this study explain that the uncertainty will be studied

only for the stock and emission and consumption modules of the model. In summary, the

justification is as follows: The demand module is calibrated on the basis of an exogenous

baseline, therefore objective uncertainty characterisation is not possible without characterising

the uncertainty of this exogenous baseline. In addition, it is structured using calibrated price

elasticities derived from empirical elasticities of substitution. It is again not possible to

independently vary these price elasticities, as this would bring the model off-equilibrium.

Therefore, factors in the demand model were not modified.

In addition, no values of the welfare and life-cycle modules were modified. First, these modules

seem outdated and are not much used today which makes uncertainty estimates obsolete.

Second, they are used as post-processors, i.e. they do not take part in the equilibration model

loops. Therefore, one can characterise their uncertainty independently, without needing to run

the model. Third, these are modules utilising variables for which there is still large scientific

discussion related to their uncertainty. For example, the external cost of a tonne of pollutant is

a highly uncertain value, much beyond the range of the ‘more conservative’ values used in the

main TREMOVE module. Characterising the uncertainty of such values goes much beyond the

aims of this study. In principle, quantifying the uncertainty of the welfare module boils down in

quantifying the uncertainty of the external cost of pollution.

We also did not modify the values in the aviation, inland waterways, and rail modules. The first

reason was that the approach in these modules is much more simplified than the road

transport one. In aviation, much more elaborate models exist and have been used for aviation

scenarios, such as the Eurocontrol AEM model. Tremove is not a very elaborate model for

realistic policy development in aviation emissions. Hence characterising the uncertainty of the

aviation module would serve no real purpose. Inland waterways and rail modules on the other

hand have at least an order of magnitude less significance in total emissions than road

transport. For UK in particular inland waterways play an insignificant role (0,6% of total freight

activity in UK). In addition, they also use simplified approaches compared to the elaborate road

transport module. It was therefore decided that the uncertainty in the calculation of emissions

and activity in these modules will be only induced by the uncertainty of the road transport

module and no additional modifications were introduced.

Table 4: Input variables not related to Road Transport.


AIR_DETOUR % detour-deviation from straight OD distance Air

AIRCONSFfuelD Fuel consumption factor for aircrafts by distance class - g fuel per pkm

Air

AIRCONSFfuelDsplit_alt LTO and cruise in total aircraft fuel consumption Air

AIREMIFD Emission factor for aircrafts by distance class - g pollutant per pkm

Air

AIREMIFDsplit_alt LTO and cruise in total aircraft emissions Air

AIRPOLLCOSTUNIT External cost per unit of emitted pollutant for air - differentiated over distance classes in EURO per tonne

Air

PLANEVAT Plane transport VAT rate - % Air

NETWORKTAX Road tolls by category road type and period - EURO per vkm This is part of the COSTROAD parameter in the vehicle stock module but is actually calculated in the demand module

Demand

IWCONFIG Indicates for each configuration if and when it becomes available for each vessel type separately

IWW

54


IWCONSFfuel Inland Waterway Fuel consumption factor - g per vkm IWW

IWEMIF Inland Waterway Emission factor - g per vkm IWW

IWENGCOST Additional cost of the engine type compared to the basecase configuration - EURO per vkm

IWW

IWEQCOST Additional cost of the engine type compared to the basecase configuration - EURO per vkm

IWW

IWFCOST Inland waterway fuel resource cost - in EURO per litre IWW

IWFREDUC Reduction in fuel consumption of the configuration compared to the basecase configuration - %

IWW

IWFTAX Inland waterway fuel tax - in EURO per litre IWW

IWFUELdens Density of inland waterway fuels - grammes per litre IWW

Ktoe_per_Pjoule Converson factor from PJ to ktoe Report

Pjoule_per_GWh Converson factor from GWh to Pjoule Report

METRAMCONSFelec Metro-Tram electricity consumption factor - kWh per vkm Rail

SULPHUR_RDIESEL Sulphur content of diesel train fuel for years for years for which it is specified - ppm

Rail

TACTTREXalleng TRENDS/EX-TREMIS activity by train type - in million pkm or tonkm per year

Rail

TCONSFelec Train electricity consumption factor - kWh per vkm Rail

TEMIF Train direct emission factor - g per vkm Rail

TLOADfDif UIC % difference between load factor for diesel and electric - 0 for electric

Rail

TSTBY EX-TREMIS stock of train vehicles (1980-2030) per train vehicle type - in thousands vehicles

Rail

TSTNBY EX-TREMIS stock of train vehicles per train vehicle type and age 1995-2005 - in thousands vehicles

Rail

TVKMEXTREMIS EX-TREMIS vehicle-kilometres by train type : data and projections - million vkm

Rail

ACCIDENTCOST Average accident cost for non-road modes - EURO per 1000 pkm or 1000 tkm

Welfare

ACCIDENTMARGCOST Marginal external accident cost for road modes - EURO per vkm - by road type

Welfare

L Marginal cost of public funds Welfare

LC_POLLCOSTUNIT External cost per unit of emitted pollutant in EURO per tonne Welfare

NOISECOST Average noise cost - EURO per 1000 pkm or 1000 tkm Welfare

POLLCOSTUNIT External cost per unit of emitted pollutant for road in EURO per tonne

Welfare

WEARMARGCOST Marginal wear and tear cost - EURO per vkm Welfare

2.5 Changes over interim report

TREMOVE is a policy assessment model, designed to study the effects of different transport and

environment policies on the emissions of the transport sector. The model consists of three

main modules: a demand, a stock, and an emissions module. These are accompanied by two

additional modules, the well-to-tank and the welfare modules. The model structure also

includes a number of input variables, model variables and model parameters. The input

variables contain all the information coming from different sources necessary to calculate the

basecase. Model parameters contain the information necessary to perform the calculations

internally in the model. They are mainly equation parameters used to calculate eg emission

factors. Model variables store all intermediate information exchanged between the different

subroutines and modules. The sensitivity analysis aimed at quantifying the uncertainty of the

input variables. Due to the large number of input variables it was initially decided to model a

subset of the input variables and for the remaining to influence them via model parameters.

For this reason, and in order to avoid inconsistencies a comprehensive study of the model was

55

conducted. Key variables and parameters were identified and their interaction and relationship

was examined.

In the interim report, it was decided that one of the key model parameters that was going to

be independently varied was the COSTROAD parameter. This parameter summarizes the cost

components originating from the vehicle stock module for all non-bus road modes, in euro per

vkm. There were two main reasons why this parameter was initially selected to be varied

although this was not one of the input variable. First, to save precious calculation time, instead

of varying each individual variable that contributes to the cost items of the COSTROAD

variable. Second, this parameter summarised all the information related to costs in an orderly

fashion (€/vkm). Literature values on cost could be directly found in these units per cost item

of the COSTROAD variable and this transparently facilitated the modelling procedure as well as

the software modifications.

Following this logic the first test runs were performed and results were collected. After

examining the results it was apparent that they were not consistent with expected results. The

correlation between the modification of the input and the resulting output data could not be

explicitly explained. A more detailed examination showed the causes of this error. First of all

the level of aggregation of the input data and the model parameter COSTROAD was different.

Input data were aggregated according to the vehicle type (eg vehicle size) or even vehicle

technology, while the COSTROAD parameter was aggregated according to vehicle category.

Vehicle categories in the TREMOVE model are cars, buses, mopeds, motorcycles, light duty

trucks and vans, heavy duty vehicles in 4 weight categories. This lead to the second and most

important error inherent in the model calculations. The COSTROAD parameter exchanges

information between the vehicle stock and the demand module. Calculations in the demand

module are performed at the aggregation level of vehicle category. On the other hand

calculations in the vehicle stock module are performed at a vehicle type or even technology

level. The flow of information exchange can be seen in Figure 6.

Figure 6: Data flow between vehicle stock and demand module in TREMOVE.

The demand module determines the total vehicle kilometres per vehicle category for a specific

year. This data is loaded to the vehicle stock module where it is split into the different vehicle

types and technologies. Total cost in TREMOVE is then calculated based on this data and the

cost information given to the model as input data. This information follows the aggregation

56

level of the vehicle stock module, so in order for this to be sent back to the demand module,

data must be aggregated back to vehicle category level.

In the first approach the COSTROAD parameter was altered according to the specific scenario

right before being inserted back to the demand module. However by modifying the COSTROAD

parameter at this point, the vehicle split that was underlying to the total cost was not taken

into account, since it was not affected by this modification. For this approach to be consistent

with the model calculations one should modify the vehicle split accordingly, something that

would not be possible. In any case this would create a vicious circle where the aggregated cost

would be calculated based on the vehicle split that would be affected by the same total cost.

For these reasons, it was decided, in spite the extra calculation time that would be needed, to

vary independently only the t variables in order to ensure model calculation consistency.

The second modification to the interim report is that the final calculations do not include an

uncertainty variance of the logit module parameters. In the interim report, it was proposed

that surrogate values could be used for the logit parameter variance. These surrogates were

basically literature values of elasticity of demand equivalent to logit-functions relevant

parameters. That is, it was suggested that literature uncertainty ranges for elasticity of

demand of ownership costs, fuel costs, and running costs could be used as proxies for the

uncertainty range of the parameters RLOGITP_ACC, RLOGITP_FCOSTS, and RLOGITP_OCOSTS

respectively. However, in the meeting where the interim report was presented, the Commission

experts suggested that this is not a reasonable approach as the elasticity of demand is

potentially an equivalent of the logit function output and not the logit functions parameters.

Therefore, in the absence of any other alternative to characterise the variance of the

parameters, it was decided not to vary them independently. Therefore, the variance of the logit

functions output is only determined by the variance of the input variables to the logit functions

(section 2.2). Several other parameters in the logit functions are dummy parameters, i.e. they

have obtained fixed values so that the output of the logit functions agrees historical data on

vehicle choices. Naturally, no uncertainty range could be defined since these are fixed to

historical data.

Table 5: Input variables & parameters used in TREMOVE logit module.


RLDVskale Scale parameter in LDV logit model Logit

RMOCskale Scale parameter in MOC logit model Logit

RREPMAINTC_LINK_RPCS Switch to relate vehicle repair and maintenance cost linearly to purchase cost - 1 is linear - 0 is no link

Logit

Rrepmaintcevol % change in proportion of rep&maint cost RVEH vs PCDM for cars over time - ie vintage - 0 to 1

Logit

RINC_LFC_UNSOLD Enables setting LIFEC high for vehicles that are not sold in year T

Logit

RLDVdummy Dummy parameter in LDV logit model Logit

RLOGITCBUS Coefficient in the bus logit model Logit

RLOGITDUM_B Car logit model - dummy coefficient for big cars Logit

RLOGITDUM_DB Car logit model - dummy coefficient for big diesel cars

Logit

RLOGITDUM_DM Car logit model - dummy coefficient for medium diesel cars

Logit

RLOGITDUM_DS Car logit model - dummy coefficient for small diesel cars

Logit

RLOGITDUM_NO Car logit model - dummy for Norway Logit

57


RLOGITDUM_S Car logit model - dummy coefficient for small cars Logit

RLOGITDUM_UK Car logit model - dummy for United Kingdom Logit

RLOGITP_ACC Car logit model - acceleration coefficient Logit

RLOGITP_ACCNO Car logit model - additional acceleration coefficient for Norway

Logit

RLOGITP_ACCUK Car logit model - additional acceleration coefficient for United Kingdom

Logit

RLOGITP_DUMMYDB Car logit model - big diesel car dummy coefficient Logit

RLOGITP_FCOSTS Car logit model - fuel cost coefficient Logit

RLOGITP_INCLARGE Car logit model - parameter for income in case of big car choice

Logit

RLOGITP_INCSMALL Car logit model - parameter for income in case of small car choice

Logit

RLOGITP_IVLARGE Car logit model - inclusive value scaling parameter for big cars

Logit

RLOGITP_IVMEDIUM Car logit model - inclusive value scaling parameter for medium cars

Logit

RLOGITP_IVSMALL Car logit model - inclusive value scaling parameter for small cars

Logit

RLOGITP_OCOSTS Car logit model - other cost coefficient Logit

RMOCdummy Dummy parameter in motorcycle logit model Logit

SLOGITCONST car scrap logt: constant term Logit

SLOGITDBIG car scrap logt: dummy for big cars Logit

SLOGITDCOUNTRY car scrap logt: dummy for countries Logit

SLOGITDSMA car scrap logt: dummy for small cars Logit

SLOGITDUMMY car scrap logt: dummies Logit

SLOGITPROPACC car scrap logt: probability of scrap due to accident Logit

SLOGITPURCHASE car scrap logt: purchase cost coefficient Logit

SLOGITREPAIR car scrap logt: repair cost coefficient Logit

SLOGITREPxRES car scrap logt: repair x residual value coefficient Logit

SLOGITRESIDUAL car scrap logt: residual value coefficient Logit

58

3 Modelling theory / approach

3.1 General

Uncertainty analysis is the study of the variation in model output resulting from the collective

variation in the model inputs. The objective of uncertainty analysis is to:

1. quantify the uncertainty in the model output, given the uncertainty in the inputs,

2. develop confidence intervals about the mean or distribution function of the model

output.

Sensitivity analysis quantifies the relative contribution of the input factors in forming the

uncertainty in the model output. The output uncertainty is mapped back to the input factors to

identify the ones that are mainly responsible for that output uncertainty. The objectives of the

sensitivity analysis are:

1. identify input variables that have a large influence on the output uncertainty

(for subsequent calibration / optimisation tasks, or prioritisation of research);

2. identify non relevant variables (for model reduction purposes);

3. improve the understanding of the model structure (highlighting interactions among

variables, combinations of variables that result in high / low values for the model

output);

4. model verification and corroboration (to check whether the model behaviour is in line

with scientist expectations);

5. model quality assessment (to check whether the model output uncertainty depends on

hard science, eg lack of knowledge in data, or on soft-science, eg subjective

preferences and assumptions), etc.

A number of sensitivity analysis methods can be used to accomplish such task and,

consequently, many techniques have been proposed (e.g. linear regression or correlation

analysis, measures of importance, sensitivity indices, screening, etc.). A thorough description

of such techniques can be found in Saltelli et al. [7]. Here we will focus on two of them, which

have been extensively used in this project: the screening approach with quasi-random LpTau

sampling (Sobol et al. [9]) and the variance-based methods, these latter being implemented

via the Sobol method (Sobol [10]).

3.2 Methods

The TREMOVE model is a complex model that involves a large number of input factors. The

choice of a well-designed experiment is essential in order to identify the most important factors

among a large number and quantify their importance.

59

Global sensitivity analysis using variance-based methods considers the full range of variation of

the input parameters along their joint distribution. Variance-based methods seek to decompose

the total output variance into its contributions from each input factor. The importance of a

given input factor can be measured via the so-called sensitivity index, which is defined as the

fractional contribution to the model output variance due to the uncertainty in the input factor.

These methods involve Monte-Carlo (MC) sampling of the input factors according to specific

sampling strategies. Thus they reflect the full range of variation of the input factors. Because

the factors are varied simultaneously, this involves a multidimensional averaging. The apparent

drawback of variance based methods is the so-called curse of dimensionality, which is palpable

when the number of factors becomes large: the number of terms in the decomposition of the

output variance grows exponentially with the number of factors. In cases where the model

contains a large number of factors or/and it is computationally too expensive, the application of

variance based methods like FAST or Sobol’ is not possible.

Screening designs are a convenient choice when the objective is to identify the subset of input

factors that can be fixed at any given value over their range of uncertainty without reducing

significantly the output variance (i.e. identify non-influential factors). The screening methods

provide a list of factors ranked in order of decreasing importance allowing the modeller to

identify the subset of less influential ones. Screening designs like the Morris method are

computationally cheap and model free. As a drawback, these methods tend to provide

qualitative sensitivity measures, i.e. they rank the input factors in order of importance, but do

not quantify how much a given factor is more important than another. Nevertheless, it does

not supply the variance decomposition obtained with the variance-based measures.

For these reasons, the analysis has been performed in two steps. First, a screening analysis

based on quasi-random LpTau sequences identified the most influential input parameters.

Then, a variance based sensitivity analysis technique (Sobol) quantified the uncertainty of the

road transport emissions.

3.2.1 Variance-Based Methods

Let Y be an output value of the model and Xi the input factors or modelling terms this depends

on. In variance-based methods the output variance V(Y) can be decomposed in the sum of a

top marginal variance and a bottom marginal variance. Specifically,

( ) ( )[ ] ( )[ ]UYVEUYEVYV += (Eq: 6)

where U is a group of one or more Xi terms. The top marginal variance from U is the expected

reduction of the variance of Y in case U becomes fully known and is fixed at nominal values,

whereas other inputs remain variable as before. The bottom marginal variance from U is defined

as the expected value of the variance of Y in case all inputs but U become fully known, U

remaining as variable as before.

The main effect or “first order” sensitivity index Si, representing the sensitivity of Y to the factor

Xi, is defined as the top marginal variance divided by the total variance, where the subset U

reduces to the single factor Xi:

60

)Y(V

)]xXY(E[VS

*ii

i

==

(Eq: 7)

and represents the average output variance reduction that can be achieved when Xi becomes

fully known and is fixed. Estimation procedures for Si are the Fourier Amplitude Sensitivity Test,

FAST, the method of Sobol’ and others. Higher order sensitivity indices, which quantify the

sensitivity of the model output to interactions among subsets of factors, can be estimated

using similar formula. For instance, the second order sensitivity index Sij, representing the

sensitivity of Y to the interaction between Xi and Xj, is:

)Y(V

)]xXY(E[V)]xXY(E[V)]xX,xXY(E[VS

*jj

*ii

*jj

*ii

ij

=−=−===

(Eq: 8)

From the definitions in equations (Eq: 7) and (Eq: 8), a complete series development of the

output variance can be achieved:

k...12mji

ijmi ji

iji S...SSS1 ++++= ∑∑ ∑<<<

(Eq: 9)

where higher order terms are defined in a similar way to (Eq: 8).

Given that the estimation of each sensitivity index, be it Si, Sij or higher order, might require a

significant number of model executions, the analysis is rarely carried further after the

computation of second order indices (their number is k(k-1) where k is the number of input

variables), as the related computational load might be impracticable.

The investigation of higher order effects is computationally cheaper if total sensitivity indices

are employed. The total sensitivity index STi for the factor Xi collects in one single term all the

interactions involving Xi. It is defined as the average output variance that would remain as long

as Xi stays unknown (i.e. the bottom marginal variance with U grouping all factors but Xi):

)Y(V

)]xXY(V[ES

*ii

Ti

−− ==

(Eq: 10)

The term X-i indicates all the factors but Xi. The usefulness of the STi for each term is in that

they can be computed without necessarily evaluating the single indices Sijm…, thus making the

analysis affordable from a computational point of view.

Estimating the pair (Si, STi) is important to appreciate the difference between the impact on Y

of the factor Xi alone (the Si) and the overall impact on Y of factor Xi through interactions with

the others (the STi). Such property is particularly interesting in a calibration framework, where

high order interactions are usually encountered. Efficient estimators of the pair (Si, STi) are

provided by variance-based techniques such as the extension of the Fourier Amplitude

Sensitivity Test (xFAST) and the Sobol' method.

The extended FAST (Saltelli et al. [8]) and the Sobol methods (Sobol [9]) yields estimates of

the total sensitivity indices, STi defined as the sum of all the indices (Si and higher orders)

where the variable Xi is included. The STi concentrates in one single term all the effects of Xi on

Y. For additive models (no interactions), Si = STi for all the Xi. The estimation of the total

61

sensitivity indices STi makes the analysis affordable from a computational point of view, as only k

total indices are needed to account completely for the total output variance V. Furthermore, those

methods allow the simultaneous evaluation of the first and total effect indices. The estimation of

the pair (Si ,STi) is important to appreciate the difference between the impact of Xi alone on Y

(i.e. Si) and the overall impact of factor Xi through interactions with the other input variables

on Y (i.e. STi). Clearly the Si1, i2, …,is add up to one; this is not true for the STi’s.

3.2.2 Screening Methods

Screening methods are useful in the modelling practice to investigate which factors - among

the many potentially important factors - are really important. This could help in coming up with

a short list of influential factors.

Screening methods deal with models containing hundreds of input variables, or/and with very

computationally expensive models, such as TREMOVE. They are economical from a

computational point of view, but as a drawback, they provide qualitative sensitivity measures

(i.e. they rank the input variables in order of importance, but do not quantify how much a

given variable is more important than another). There is clearly a trade-off between

computational cost and information. Several approaches to the problem of screening have been

proposed in the literature.

Quasi-Random LpTau sampling generates quasi-random sequences. These are uniformly

distributed sets of points in the N-dimensional unit cube. Each point generated is then used as

input to the transformation that calculates the inverse cumulative function of each element of

the factor set. The sensitivity of model predictions to individual input variables can be then

determined by means of techniques based on regression analysis (PEAR and PCC, SRC) and

their rank transformation (SPEA, PRCC and SRRC). A brief description of the indices is given

hereafter.

Assume that after the Monte Carlo study is performed, the propagation of the sample through

the model creates a mapping of the form

[ ] m,...1i,x,...,x,x,y ik2i1ii = (Eq: 11)

where k is the number of independent variables and m is the sample size.

- PEAR: Pearson product moment correlation coefficient is the usual linear correlation

coefficient computed between yi and xij (i = 1, …, m). It provides a measure of the

linear relationship between Xj and Y.

- SPEA: Spearman coefficient is essentially the same as PEAR, but using the order ranks

instead of the raw values of both Y and Xj. The rank transformation is a simple

procedure which involves replacing the data with their corresponding ranks, i.e. assign

rank 1 to the smallest observation and continue to rank N for the largest observation.

For non-linear models, SPEA is preferred as a measure of correlation.

- PCC: Partial correlation coefficient between Y and Xj is defined as the correlation

coefficient between YY − and jj XX − . The partial correlation coefficients provide a

measure of the strength of the linear relationship between two variables after a

62

correction has been made for the linear effects of other variables in the analysis. In

other words, PCC gives the strength of the correlation between Y and a given input Xj

cleaned of any effect due to any correlation between Xj and any of the Xi,i≠j. In

particular PCC’s provide a measure of variable importance that tends to exclude the

effects of other variables.

- PRCC: Partial rank correlation coefficient is PCC computed on the ranks. For non-linear

models, PRCC is preferred as a measure of partial correlation.

- SRC: Standardised Regression Coefficients quantify the effect of varying each input

variable away from its mean by a fixed fraction of its variance while maintaining all

other variables at their expected values. They are computed through a regression

model, hence it is important to consider the model coefficient of determination (CMD)

obtained from the regression model. CMD provides a measure of how well the linear

regression model based on SRC’s can reproduce the actual output y. It represents the

fraction of the variance of the output explained by the regression. The closer CMD is to

unit, the better is the model performance. The validity of the SRC’s as a measure of

sensitivity is conditional on the degree to which the regression model fits the data, i.e.

to CMD.

- SRRC: Standardised Rank Regression Coefficients are SRCs computed on the ranks.

The difference between the CMDs computed on the raw values and on the ranks, is a

useful indicator of the non-linearity of the model.

3.3 Parameterisations of input data

TREMOVE estimates transport demand, modal shifts and vehicle stock renewal as well as

emissions of air pollutants and costs for policies such as road pricing, public transport pricing,

emission standards, subsidies for cleaner cars, and others. The model consists of three main

modules: a demand, a stock, and an emissions module. These are accompanied by two

additional modules, the well-to-tank and the welfare modules. These two are add-ons on the

main structure of the model, aiming at estimating the upstream (fuel production) costs of

transport and the benefit (in monetary terms) of emission reduction to the society,

respectively, in addition to the TREMOVE main output (cost, emissions, consumption).

Emission and fuel consumption estimates follow the COPERT4 methodology and are generally

distinguished in three sources: emissions produced during thermally stabilised engine operation

(hot emissions), emissions occurring during engine start from ambient temperature (cold-start

and warming-up effects) and NMVOC emissions due to fuel evaporation. The total emissions

are calculated as a product of activity data provided by the user and speed-dependent emission

factors calculated by the software. According to the recent COPERT4 uncertainty study

(Kouridis et al. [15]), the most important inputs to the emissions module are the emission

factors, the traffic data (like vehicle mileage) and the model parameters (like vehicle speed,

average trip length).

In addition, several uncertain variables are included in the demand and stock modules like the

acceleration, the vehicle registration tax, the vehicle repair cost and others. Many input

parameters are usually multi-dimensional arrays. For example, the emission factor is a 5-

dimensional variable, depending on the Vehicle Category (PC, LDV, etc), Technology (Euro-1,

Euro 2, etc), Engine Size (<1.4lt, >2.0lt, etc), pollutant and speed range considered.

63

The "total error", totalε , of the TREMOVE estimates results from an entire "chain of errors". This

total error consists of three error contributions:

(a) STOCKε denotes the error which comes along with the estimation of the total

amount of stock;

(b) COSTε represents the uncertainty in the parameters related to the cost

module;

(c) EMISSIONSε denotes the uncertainty associated with the road transport

emissions.

In the Monte Carlo version of TREMOVE, we wish to acknowledge the uncertainty in all these

inputs. The process of considering uncertainty in scalar (0-D) variables is straightforward

through their statistical distribution (although perhaps not easy to quantify), based on the

literature values collected in the previous chapter. The process is however not easy for multi-

dimensional input variables, for which we need to identify, via a statistical model, a suitably

small set of parameters that describe well the multi-dimensional system. By associating a

proper uncertainty to these model parameters, we can then represent and characterize the

uncertainty in the multi-dimensional system.

The parameterization of the contribution of all sources of uncertainty in TREMOVE resulted in a

significant reduction of the total uncertainty inputs to 33. Specifically:

(a) 6 parameters corresponding to STOCKε : uparaBT, erlogitpACC, eRSHairco,

rlogitCNGAVAIL, uBTmileage, RFACTORUNCONV.

(b) 12 parameters corresponding to COSTε : eRPCS_BASE, usresidualparaAB,

RLPG_FIT_COST, eRPCS_INCREASE_2009, eRPCS_INCREASE_2012, RINSCFRACTION,

RLABOURC, RLABOURTX, ROWNTX, RREPMAINTC_INCREASE_RTECH_RES,

eRREPMAINTCFRACTION, PUBLICCOSTCOV.

(c) 15 parameters corresponding to EMISSε : eEF, eEF_fc, eEFratio, eEFratio_fc, RHC,

eFUEL_ENERGY_DENSITY, eFUELSPEC, eRFC_REDUC_RESISTANCE, eRFCairco,

eRLOADCAP, RFUEL_COMPOSITION, RVP, TMAX, TMIN, ltrip.

The details of this process are given in the next section.

3.3.1 Parameterisation of STOCKε

The uncertainty of the log-normally distributed stock parameters rlogitpACC and RSHairco was

appointed to the stochastic variables erlogitpACC and eRSHairco and was quantified as

described hereafter.

(a) We have generated the statistical distributions based on the variable

uncertainty provided by expert opinion (LWA) as mean zijexpert

and standard deviation

sijexpert

for the year 2005 as well as the TREMOVE data for the period 1991-2030. This

yields zij(t)=(zijexpert

/ zijTREMOVE(2005))*zij

TREMOVE(t) for the variable mean and

sij(t)=(std(zijexpert)/mean(zij

expert))*zijTREMOVE(t) for the variable standard deviation.

64

(b) We fit a time dependent log-normal distribution to the TREMOVE data with

mean and standard deviation calculated in the previous step. The stochastic input Z (e.g.,

RSHairco) for the year t is based on the formula:

(Eq: 12)

(Eq: 13)

(Eq: 14)

e ~ N(0,1) (Eq: 15)

The parameterization of RFACTRUNCONV was based on the mean and standard deviation

provided by the expert group. A log-normal non-time dependent distribution was then fit to the

data utilizing the stochastic factor eRFACTRUNCONV.

Further, we evaluated the combination of values for the model parameters bm and Tm

corresponding to the characteristic service life that do not violate the constraints imposed

towards acceptable model response. This generated the 2-dimensional fitting function surface

in the parameter space. A stochastic variable (uparaBT) was employed to sample values from

the generated response surface, i.e. the permissible values of the joint probability distribution

function of bm and Tm. The same approach was also adopted for the average annual mileage of

new cars in each year (uBTmileage).

Finally, the variable rlogitCNGAVAIL was parameterized as N(0.29, 0.02) and was read directly

from the generated sample.

Table 6: List of the uncertain input variables, belonging to the stock error category, with their

statistical distributions

Normal distr µ σ

Uniform distr No Error Cat Input Variable Description Units Distribution

min max

1 eSTOCK uparaBT (B,T) - parameter in TRENDS: characteristic service life & failure steepness

- Uniform 0 1

2 eSTOCK eRLOGITPACC Acceleration for big and medium car logit

sec Normal 0 1

3 eSTOCK eRSHairco Share of new sold vehicles fitted with air-conditioning

% Normal 0 1

4 eSTOCK rlogitCNGAVAIL Relative availability of CNG in fuel stations

% Normal 0.29 0.02

5 eSTOCK uBTmileage Average annual mileage of new cars in each year - exogenous estimate - vehicle kilometres per year

# Uniform 0 1

6 eSTOCK eRFACTORUNCON

V

Ratio fuel consumption unconventional vs equivalent conventional vehicle - [(kg/km) / (kg/km)]

- Normal 0 1

65

The following figures display, for all related input variables, the probability density functions

(PDFs) that result from the modification of the basecase TREMOVE value, which was actually

used for the uncertainty calculations. The figure shows as an example the PDFs for correspond

to Passenger car Gasoline 1,4-2l for the year 2010. The values for the rest of the distributions

can be found in ANNEX III.

0

20

40

60

80

100

120

Fre

quency

seconds

RLOGITPACC

0

20

40

60

80

100

120

Fre

quency

-

RSHairco

0

20

40

60

80

100

120

Fre

quency

Coefficient

RLOGITCNGAVAIL

0

20

40

60

80

100

120

Fre

quency

(kg/km) / (kg/km)

RFACTORUNCONV

Figure 7: Probability density functions of the input variables used in TREMOVE (example gasoline PC

1.4-2.0 l).

3.3.2 Parameterisation of COSTε

The uncertainty of the log-normally distributed stock parameter RCPS_BASE was appointed to

the stochastic variable eRCPS_BASE and was quantified according to the equations (Eq: 12) to

(Eq: 15).

The parameterization of RREPMAINTC_INCREASE_RTECH_RES was based on the mean and

standard deviation provided by the expert group. A log-normal non-time dependent distribution

was then fit to the data utilizing the stochastic factor eRREPMAINTC_INCREASE_RTECH_RES.

The uncertainty for RREPMAINTCFRACTION, RPCS_INCREASE_2009, RPCS_INCREASE_2012,

RLABOURC, RLABOURTX and PUBLICCOSTCOV is given as a perturbation to the nominal values

in TREMOVE:

66

( ) ( ) )e*r1(*tZtZ TREMOVE += (Eq: 16)

The factor r has the following values (originating from the input data collected in the previous

chapter, i.e. CV of 20% or 30% respectively):

r = 0.1 RLABOURC, RLABOURTX, RREPMAINTCFRACTION

r = 0.07 PUBLICCOSTCOV, RPCS_INCREASE_2009, RPCS_INCREASE_2012

The uncertainty for RINSCFRACTION and ROWNTX is given as a perturbation to the statistical

distribution proposed by the expert group (LWA-EMISIA). The distribution for RINSCFRACTION

is non-time dependent while for ROWNTX two different distributions were provided

(before/after 2010):

( ) ( ) )t(sigma*etmeantZ += (Eq: 17)

For the remaining two variables, usresidualparaAB is modelled in a similar manner to uparaBT (but in a time-dependent mode) while RLPG_FIT_COST is read directly from the Monte Carlo sample.

Table 7: List of the uncertain input variables, belonging to the cost error category, with their statistical distributions

Normal distr µ σ


min max

2 eSTOCK eRPCS_BASE Road vehicle basic purchase resource cost

Euro Normal 0 1

3 eCOST RLPG_FIT_COST Resource cost to retrofit LPG installation

Euro Uniform 1800 2500

4 eCOST eRREPMAINTC_INCREASE

_RTECH_RES

Increase in yearly maintenance cost for using technologies to reduce vehicle and engine resistance factors

Euro Normal 0 1

5 eCOST eRREPMAINTCFRACTION Repair and Maintenance Cost excl. taxes as % of purchase resource cost (ex tax)

% Normal 0 1

6 eCOST eRPCS_INCREASE_2009 Vehicle purchase cost increase to reach the 140g car CO2 target in 2009

% Normal 0 1

7 eCOST eRPCS_INCREASE_2012 Vehicle purchase cost increase to reach the car CO2 target in 2012 - on top of 140g

% Normal 0 1

8 eCOST eRINSCFRACTION Insurance cost as percentage of vehicle purchase resource cost

% Normal 0 1

9 eCOST eRLABOURC Labour cost - net wage - for truck drivers

Euro Normal 0 1

10 eCOST eRLABOURTX Labour tax - bruto wage minus netto wage - for truck drivers

Euro/h Normal 0 1

11 eCOST eROWNTX Annual Ownership tax road vehicles

Euro Normal 0 1

12 eCOST ePUBLICCOSTCOV Public transport fare cost coverage

% Normal 0 1

The following two figures display, for cost relevant input variables, the distribution that results

from the modification of the basecase TREMOVE value, which was actually used for the

uncertainty calculations. Figure 8 presents examples of the distributions for gasoline passenger

cars 1.4-2l for year 2010; for all other vehicle categories and years data can be found in

ANNEX III.

67

0

20

40

60

80

100

120

Fre

quency

€

RPCS_BASE

0

20

40

60

80

100

120

Fre

quency

Euro 2005

ROWNTX

0

20

40

60

80

100

120

Fre

quency

Euro 2000


0

20

40

60

80

100

120

Fre

quency

-

RREPMAINTCFRACTION

0

20

40

60

80

100

120

Fre

quency

%

RPCS_INCREASE_2009

0

20

40

60

80

100

120

Fre

quency

%

RPCS_INCREASE_2012

0

20

40

60

80

100

120

Fre

quency

%

RINSCFRACTION

Figure 8: PDFs of the cost-related input variables in TREMOVE (examples for gasoline passenger

cars 1.4-2l for year 2010).

68

0

20

40

60

80

100

120

Fre

quency

Euro/hour

RLABOURTX

0

20

40

60

80

100

120

Fre

quency

Euro/hour

RLABOURC

0

20

40

60

80

100

120

Fre

quency

%

PUBLICCOSTCOV

Figure 9: PDFs of additional cost-related input variables in TREMOVE (examples for the year 2010).

3.3.3 Parameterisation of EMISSε

The uncertainty of the emission and consumption relevant parameters RHC, RVP, TMIN and

TMAX was quantified on the basis of the mean and standard deviation provided by the expert

group. A log-normal non-time dependent distribution was then fit to the data utilizing the

stochastic factors eRHC, eRVP, eTMAX and eTMIN.

The uncertainty for eFUEL_ENERGY_DENSITY, eFUELSPEC, eRFC_REDUC_RESISTANCE,

eRFCairco, eRLOADCAP and RFUEL_COMPOSITION was simulated according to equation (Eq:

16 (r=1). The perturbation size proposed by the expert group was 2% for FUELSPEC, 3% for

FUEL_ENERGY_DENSITY, 20% for RLOADCAP, 30% for RFC_REDUC_RESISTANCE and

RFUEL_COMPOSITION and 50% for RFCairco. The average road trip length (ltrip) is read

directly from the MC sample.

The data collected from laboratory measurements are usually processed by regression analysis

to provide a set of regression coefficients that are meant to explain the underlying

phenomenon through a polynomial curve that fits the observed data. Such regression

coefficients are subsequently stored in tables and employed during the execution of the

emission module (resp. COPERT 4). In order to estimate the uncertainty of the emission

factors, raw measured data were analyzed, as described by Kouridis et al. [15]. On the basis of

the statistical analysis of the vehicle measurements available, experimental errors for the

69

coefficients were estimated. Such experimental errors are in the form of stochastic variables

that, coupled with the polynomial regression curves (efCOPERTi), reproduce the experimental

pattern. The probability distribution functions for the stochastic emission factors (ef) are set up

utilising the following procedure:

HOT EMISSION FACTORS AND FUEL CONSUMPTION FACTORS

(a) The laboratory measurements have been clustered to 14 equally sized

velocity classes (1: 0-10 km/h, 2: 10-20 km/h, …, 14: 130-140 km/h); for each velocity

class (v1, v2,…, vK), k=1,2,…,14, we calculate its standard deviation (s1, s2,…, sK).

(b) We fit a speed dependent log-normal distribution to the laboratory

measurements with mean equal to the polynomial regression curve (efHOTCOPERT

i) and

standard deviation calculated in the previous step (s1, s2,…, sK). The hot emission factor

(efHOT) for the sampled velocity Vj is based on the formula:

( ) eEF*jHOT

jjeVefσ+µ

= (Eq: 18)

( )( ) ( )

+−=µ

2

jCOPERTHOT

ij

COPERTHOTj

Vef

s1ln5.0Vefln

(Eq: 19)

( )

+=σ

2

jCOPERTHOT

ij

Vef

s1ln

(Eq: 20)

eEF ~ N(0,1) (Eq: 21)

This procedure, which reproduces the experimental pattern of the hot emission factors, has

been repeated for all Vehicle Categories (PC, LDV, etc), Technologies (Euro-1, Euro 2, etc),

Engine Sizes (<1.4lt, >2.0lt, etc) and pollutants.

COLD EMISSION FACTORS AND FUEL CONSUMPTION FACTORS

(a) The cold emission factors have been split in fourteen speed classes, similar to hot ones (1:

0-10 km/h, 2: 10-20 km/h, …, 14: 130-140 km/h); for each velocity class (v1, v2,…, vK),

k=1,2,…,14, we calculate the standard deviation (s1, s2,…, sK) of (efCOLD/efHOT-

1)*efHOTCOPERT, assuming that the ratio of standard deviation over mean of the hot emission

factors is equal to the standard deviation over mean for the cold emission factor. The

highest speed classes are not relevant for the cold-start emission factor as the COPERT 4

cold-start functions are valid only up to 45 km/h and cold-start is allocated to urban

conditions only.

(b) No uncertainty of cold-start emission factors to temperature has been assumed, as no data

were available. Therefore, the uncertainty of cold-start on ambient conditions originates

only from the uncertainty in the temperature ranges, described earlier in this section.

(c) We fit a speed dependent log-normal distribution to the approximated variance of cold-

70

start emission factors with mean equal to the calculated value for the particular speed

(efCOPERTi) and standard deviation calculated in the previous step (s1, s2,…, sK). The cold

emission factor (efCOLD) for the sampled velocity Vj is based on the formula:

( ) ( )jCOPERTHOTjRATIOjCOLD Vef*)1)V(ef(Vef −= (Eq: 22)

( ) eEFratio*jRATIO

jje1Vefσ+µ

+= (Eq: 23)

( )( ) ( )

−+−−=µ

2

jCOPERTRATIO

ij

COPERTRATIOj

1Vef

s1ln5.01Vefln

(Eq: 24)

−+=σ

2

COPERTRATIO

ij

1ef

s1ln

(Eq: 25)

eEFratio~N(0,1) (Eq: 26)

The above procedure, which reproduces the experimental pattern of the cold-start emission

factors, has been repeated for all Vehicle Categories (PC, LDV, etc), Technologies (Euro-I, Euro

II, etc), Engine Sizes (<1.4lt, >2.0lt, etc) and pollutants.

71

Table 8: List of the uncertain input variables, belonging to the emissions error category, with

their statistical distributions.

Normal distr µ σ


min max

1 eEMISS eEF amplitude HOT Emission Factor

gr/km L-Normal 0 1

2 eEMISS eEFratio Cold-start emission factor - L-Normal 0 1

3 eEMISS eEFfc amplitude HOT Fuel consumption Factor

gr/km Normal 0 1

4 eEMISS eEFfcratio Cold-start emission factor - Normal 0 1

5 eEMISS ltrip Mean trip length km L-Normal 2.5 0.2

6 eEMISS eFUEL_ENERGY_DENSITY Fuel energy density GJ per

kg Uniform -0.03 0.03

7 eEMISS eFUELSPEC Fuel specification history - Uniform -0.02 0.02

8 eEMISS eRFC_REDUC_RESISTANCE

Real world fuel consumption reduction from utilisation of technologies to reduce vehicle and engine resistance factors

% Uniform -0.3 0.3

9 eEMISS eRFCairco Extra fuel consumption from use of air-conditioning equipment

l/km Normal 0 0.16

10 eEMISS RHC Ratio of hydrogen to carbon atoms in fuels

- Normal 0 1

11 eEMISS RVP Gasoline volatility (Reid Vapour Pressure)

kPa Normal 0 1

12 eEMISS TMAX Maximum temperature per month

oC Normal 0 1

13 eEMISS TMIN Minimum temperature per month

oC Normal 0 1

14 eEMISS eRLOADCAP Average maximum loading capacity big truck

t Normal 0 0.07

15 eEMISS RFUEL_COMPOSITION Average share of components in blended fuels

% Normal 0 0.1

The following figures display, for all related input variables, the distribution that results from

the modification of the basecase TREMOVE value, which was actually used for the uncertainty

calculations.

72

0

20

40

60

80

100

120

140Fre

quency

km

LTRIP

0

10

20

30

40

50

60

Fre

quency

GJ/kg

FUEL_ENERGY_DENSITY

0

10

20

30

40

50

60

Fre

quency

-

FUELSPEC

0

10

20

30

40

50

60

Fre

quency

%


0

20

40

60

80

100

120

Fre

quency

l/km

RFCairco

0

20

40

60

80

100

120

Fre

quency

-

RHC

0

20

40

60

80

100

120

Fre

quency

kPa

RVP

0

20

40

60

80

100

120

140

Fre

quency

oC

TMAX

73

0

20

40

60

80

100

120

140

160

180

Fre

quency

oC

TMIN

0

20

40

60

80

100

120

Fre

quency

tonne

RLOADCAP

0

20

40

60

80

100

120

Fre

quency

% in weight

RFUEL_COMPOSITION

Figure 10: PDFs of the emission and consumption related input variables in TREMOVE (examples).

74

4 TREMOVE software modification and update

4.1 General

To determine the uncertainty of the model a number of modifications needed to be made to the

model code, to the input files, to the way the model is being executed and to the way the

model performs the calculations. As mentioned earlier, these modifications did not affect the

TREMOVE basecase output but were only introduced to facilitate the execution of uncertainty

runs.

4.2 Software code modification

In order for the runs to be performed, the original TREMOVE values needed to be altered. Two

ways were used to facilitate this task. The first one was the direct replacement of the value of

a parameter (e.g. paraB, paraT). The second one was the implement in the code the different

equations used to calculate the PDFs . It was very important to make sure that the updates

were applied to the proper part of the code so that the correct values were used throughout

the calculation of the model. For this reason parts of the original model code were rendered

inactive and a new set of files were included in the code. This allowed the study team to have a

better overview of the modifications since all the additional code could be found in those files.

All of the modifications were included in the Vehicle Stock and emission and consumption

modules of TREMOVE, since no changes were made to the Demand module. The original files

modified in the code are the following:

o Vehicle Stock Module\Calculate_Transport_Demand.gms

o Vehicle Stock Module\Calculate_Transport_Demand_BY.gms

o Vehicle Stock Module\Calibrate_Base_Case.gms

o Vehicle Stock Module\Calibrate_CES_Tree.GMS

o Vehicle Stock Module\Calibrate_CES_Tree_BY.gms

o Vehicle Stock Module\Define_Parameters.gms

o Vehicle Stock Module\Define_Parameters_Emissions.gms

o Vehicle Stock Module\Define_Parameters_Road.gms

o Vehicle Stock Module\Define_Sets.gms

o Vehicle Stock Module\Define_Sets_Road.gms

o Vehicle Stock Module\Degradation_Mileage.gms

o Vehicle Stock Module\Emissions_Road.gms

o Vehicle Stock Module\External_Cost_Module.gms

75

o Vehicle Stock Module\External_Costs_Accidents.gms

o Vehicle Stock Module\Fuel_Consumption_Road.gms

o Vehicle Stock Module\Logit_Life_Time_Costs.gms

o Vehicle Stock Module\main.gms

o Vehicle Stock Module\Money_Costs_Road.gms

o Vehicle Stock Module\Money_Costs_Road_Private.gms

o Vehicle Stock Module\Purchase_Cost_Road.gms

o Vehicle Stock Module\Purchase_Cost_Road_residual.gms

o Vehicle Stock Module\Read_Demand_Module_Output.gms

o Vehicle Stock Module\Road_Scrap_Policy.gms

o Vehicle Stock Module\Run_Simulation.gms

o Vehicle Stock Module\Run_TREMOVE.gms

o Vehicle Stock Module\Sale_Shares_Road.gms

4.3 Software code added

Apart from the code modification a number of additional files were used during calculation.

These files include the necessary data and code fragments to be used by the model. Four

groups of files were added.

The first group are GAMS (gms) files which include the code for the different PDF equations for

the emissions module. They were included in the Vehicle Stock module, since they are the

same for all runs performed. The prefix “ZZZ_” was added for a better visualization of the

filenames in the folder where the files are located.

o ZZZ_cold_emissions.gms

o ZZZ_cold_FC.gms

o ZZZ_hot_emissions_for_HDV_&_Buses.gms

o ZZZ_hot_emissions_for_non_HDV_buses.gms

o ZZZ_hot_FC_for_HDV_&_Buses.gms

o ZZZ_hot_FC_for_non_HDV_buses.gms

o ZZZ_set_year.gms

The second group are GAMS (gms) files which include the updated code used to replace the

original TREMOVE input variables. These refer to all input variables other then the emissions

module. They are also common for all runs and thus included in the Vehicle Stock module. The

prefix “zz_” was added for a better visualization of the filenames in the folder where the files

are located.

76

o zz_FUEL_ENERGY_DENSITY.gms o zz_LTRIP.gms o zz_paraBT.gms o zz_RFACTORUNCONV.gms o zz_RFC_REDUC_RESISTANCE.gms o zz_RFCairco.gms o zz_RHC.gms o zz_RINSCFRACTION.gms o zz_RLOADCAP.gms o zz_RLPG_FIT_COST.gms o zz_RMILage.gms o zz_RREPMAINTC_INCREASE_RTECH_RES.gms o zz_RREPMAINTCFRACTION.gms o zz_RVP.gms o zz_SRESIDUALparaA.gms o zz_SRESIDUALparaB.gms o zz_TMAX.gms o zz_TMIN.gms o zzT_FUELSPEC.gms o zzT_PUBLICCOSTCOV.gms o zzT_RFUEL_COMPOSITION.gms o zzT_RLABOURC.gms o zzT_RLABOURTX.gms o zzT_RLOGITCNGAVAIL.gms o zzT_RLOGITPACC.gms o zzT_ROWNTX.gms o zzT_RPCS_BASE.gms o zzT_RPCS_INCREASE_2009.gms o zzT_RPCS_INCREASE_2012.gms o zzT_RSHairco.gms

The third group are TREMOVE (inc) files which include the parameters used in the above files.

They are also common for all runs and thus included in the Vehicle Stock module.

o m_RLOGITPACC o m_RPCS_BASE o m_RSHairco o m_ROWNTX o m_RVP o m_RFACTORUNCONV o m_RREPMAINTC_INCREASE_RES o m_RHC o m_TMAX o m_TMIN o m_RINSCFRACTION o s_RLOGITPACC o s_RPCS_BASE o s_RSHairco o s_ROWNTX o s_RVP o s_RFACTORUNCONV o s_RREPMAINTC_INCREASE_RES o s_RHC o s_TMAX o s_TMIN o s_RINSCFRACTION o std_V_hdv_hot.inc o std_V_non_hdv_cold.inc o std_V_non_hdv_hot.inc

The fourth group are the files for the individual runs. These are GAMS (gms) files which include

the parameters used in the different scenario runs. They are included in each ‘Scenario’ folder

which corresponds to each single individual run.

77

Table 9: Files which include all scenario related data

Filename Parameter included para_B.gms ParaB para_T.gms ParaT res_A.gms ResidualparameterA res_B.gms ResidualparameterB empty_values.inc RLOGITCNGAVAIL, LTRIP, RLPG_FIT_COST,

eEF, eEFratio, eEF_FC, eEFratio_FC

standard_values.inc RREPMAINTCFRACTION, RPCS_INCREASE_2009, RPCS_INCREASE_2012, FUEL_ENERGY_DENSITY, RFC_REDUC_RESISTANCE, RFCairco, RLOADCAP, RLABOURC, RLABOURTX, RFUEL_COMPOSITION, PUBLICCOSTCOV, FUELSPEC

Distributions_values.inc RLOGITPACC, RPCS_BASE, RSHairco, ROWNTX, RVP, RFACTORUNCONV, RREPMAINTC_INCREASE_RES, RHC, TMAX, TMIN, RINSCFRACTION

4.4 New features

Two new features were added to the software to facilitate the execution of the runs. The first

one aimed at reducing calculation time and the second one at the automatic execution of the

calculations.

A full TREMOVE run for one country on an average PC (basecase and scenario) takes about 40

minutes. The modifications included in the code by the project members increased this time by

30%. Taken into account the vast number of runs that had to be be performed (order of

several thousands) it was very important to reduce this time. TREMOVE calculates each time

the basecase and then the scenario, so a logical approach was to omit the basecase

calculations since they are not affected by the updated values. To do this an in depth analysis

of the model calculation process was performed. This analysis showed that it was possible to

save the intermediate files used to transfer the calculated data from the basecase to the

scenario calculations. These files include all the necessary data used by the scenario. They are

GAMS data files (gdx) used to transfer data between the Demand module and the Vehicle

Stock module and compressed files that contain the rest of the data. They were calculated

once and then used by the model every time a new run-scenario was performed. This new

feature decreased the total calculation time by 25%.

To facilitate the use of the saved basecase data, and the execution of the multiple scenarios a

second feature was introduced. A new Graphical Users Interface (Figure 11) was designed

which had 3 main features:

o the overview of the calculations;

o the use of the basecase data, thus eliminating the need to perform the basecase

calculations more than once;

o the preparation of the model to run each time a different scenario. This includes:

o preparing the input data for the each scenario,

78

o copying the results folder each time to a different location and naming it

according to the scenario number;

o executing the model.

Figure 11: Graphical User Interface of the TREMOVE uncertainty model.

4.5 Guidance to use the software

The study team prepared a new version of the TREMOVE model. This version was based on the

model version 3.3.1. The main functionalities and structure of the model were not altered. In

order for a run to be prepared the user has to provide all the necessary data for all runs. As

mentioned in the above paragraphs there are four groups of files which include the additional

code and data for the scenario execution. The first 3 groups along with the original TREMOVE

files form the new version of the model. The user will not need to modify these files. However

the fourth group, which includes the scenario data, must be created by the user. Since there is

a large amount of data required for the uncertainty calculations, an MsExcel file was created in

order for these files to be created automatically. The user has the option to fill all related data

in this file using a standard MsExcel sheet. An MsExcel Visual Basic code has been written in

order for the files to be created automatically.

As mentioned in the previous paragraph a simple graphical user interface has been designed to

facilitate the execution of the runs. A single button click will start the procedure. If the

computer uses a multi-core processor multiple instances of the software can be executed

simultaneously to reduce the required calculation time.

4.6 Differences between the two steps

The analysis has been performed in two steps. First, a screening analysis identified the most

influential input parameters (512 runs). Then, a variance based sensitivity analysis technique

quantified the uncertainty of the results (5950 runs). These runs were not identical, since the

79

second step took into account only the influential parameters. The new software however was

designed in such a way that some files and custom variables used in the first step could simply

be omitted in the second step in order for the calculations to be performed. Table 10 lists these

files, and indicates in what step they were used.

Table 10: Files used in the 512 screening and the 5950 sensitivity runs

Filename 512 runs

5950 runs

ZZZ_cold_emissions.gms YES YES

ZZZ_cold_FC.gms YES YES

ZZZ_hot_emissions_for_HDV_&_Buses.gms YES YES

ZZZ_hot_emissions_for_non_HDV_buses.gms YES YES

ZZZ_hot_FC_for_HDV_&_Buses.gms YES YES

ZZZ_hot_FC_for_non_HDV_buses.gms YES YES

ZZZ_set_year.gms YES YES

zz_FUEL_ENERGY_DENSITY.gms YES

zz_LTRIP.gms YES YES

zz_paraBT.gms YES YES

zz_RFACTORUNCONV.gms YES

zz_RFC_REDUC_RESISTANCE.gms YES

zz_RFCairco.gms YES

zz_RHC.gms YES

zz_RINSCFRACTION.gms YES YES

zz_RLOADCAP.gms YES

zz_RLPG_FIT_COST.gms YES

zz_RMILage.gms YES

zz_RREPMAINTC_INCREASE_RTECH_RES.gms YES

zz_RREPMAINTCFRACTION.gms YES YES

zz_RVP.gms YES

zz_SRESIDUALparaA.gms YES YES

zz_SRESIDUALparaB.gms YES YES

zz_TMAX.gms YES

zz_TMIN.gms YES

zzT_FUELSPEC.gms YES

zzT_PUBLICCOSTCOV.gms YES YES

zzT_RFUEL_COMPOSITION.gms YES

zzT_RLABOURC.gms YES YES

zzT_RLABOURTX.gms YES YES

zzT_RLOGITCNGAVAIL.gms YES

zzT_RLOGITPACC.gms YES

zzT_ROWNTX.gms YES YES

zzT_RPCS_BASE.gms YES YES

zzT_RPCS_INCREASE_2009.gms YES

zzT_RPCS_INCREASE_2012.gms YES

zzT_RSHairco.gms YES

m_RLOGITPACC YES

m_RPCS_BASE YES YES

m_RSHairco YES

m_ROWNTX YES YES

m_RVP YES

m_RFACTORUNCONV YES

m_RREPMAINTC_INCREASE_RES YES

m_RHC YES

m_TMAX YES

m_TMIN YES

m_RINSCFRACTION YES YES

s_RLOGITPACC YES

s_RPCS_BASE YES YES

s_RSHairco YES

s_ROWNTX YES YES

s_RVP YES

s_RFACTORUNCONV YES

s_RREPMAINTC_INCREASE_RES YES

s_RHC YES

s_TMAX YES

s_TMIN YES

s_RINSCFRACTION YES YES

std_V_hdv_hot.inc YES YES

std_V_non_hdv_cold.inc YES YES

std_V_non_hdv_hot.inc YES YES

Modified code files following the TREMOVE structure can be found in ANNEX III.

80

5 Variance of the baseline output

5.1 General

A realistic uncertainty calculation of the TREMOVE basecase is not possible without

characterising the uncertainty of the exogenous baseline projection used. This is because, the

exogenous baseline is the result of a forecast procedure, taking into account most probable

estimates for macroeconomic (GDP, energy costs) and demographic (population growth) data,

as well as energy and transport efficiency assumptions to estimate baseline demand. Change in

any of these values will affect the baseline demand – however in a fashion exogenous to

TREMOVE. It is obvious that the only way to realistically estimate the uncertainty of the

TREMOVE baseline is in fact to simulate alternative baselines. This can be made in two ways:

First, to compare the output of two different model versions as a measure of uncertainty, since

each individual model version is linked to a different baseline. However, this is also associated

to differences in the formulation of the different model versions (i.e. some parameterizations

will have changed between the two versions) and it seems that although this is straightforward

to conduct, it may lead to a higher uncertainty than should realistically be expected. The

second option, which is conceptually better, is to import alternative baseline estimates into

TREMOVE. For example, one could use the POLES, TRANS-TOOLS or PRIMES models to

estimate total demand based on different macroeconomic data and then introduce these into

TREMOVE. This would determine the uncertainty range of the TREMOVE baseline. However, this

is outside the scope of the study.

Instead, the target of this project was to characterize the variance in baseline output of a given

TREMOVE version, assuming that the exogenous data are fixed. This has a practical meaning:

The variance of the baseline output expresses the additional uncertainty introduced by the

model formulations, assuming that the macroeconomic and demographic data have been

agreed. This is useful in practical model applications. When Tremove is used, for example, to

assess the impact of the introduction of a new emission standard, one is not interested on the

total uncertainty including uncertainty of the macroeconomic data. Instead, one is interested in

identifying what is the uncertainty expected in the activity, emission and costs associated with

the introduction of the new standard, in comparison to the baseline. In order to identify this,

one needs to estimate the uncertainty of the baseline and the scenario induced by the model

formulations and not by external factors to the model. Therefore, this chapter quantifies the

variance of the baseline model output induced by the uncertainty in the model variables and

not by higher-order external factors to the model.

The results of the uncertainty and sensitivity analysis of the baseline are presented for the

United Kingdom. At a first stage, the rather large number of uncertain input variables (33) has

been filtered out from its non-influential inputs through a screening sensitivity analysis using

the LPτ sequence (Sobol at al. [9]). At the second stage, the set of influential inputs is

explored thoroughly by means of a quantitative sensitivity analysis (Cukier et al. [4], Saltelli et

al. [8]) to provide uncertainty and sensitivity estimates for the total cost, activity, consumption

and pollutant emissions for the years 2005-2030.

The sensitivity analysis has been performed through the following steps:

81

1. Prepare the Monte Carlo sample for the screening experiment using quasi-random LPτ

sequences.

2. Execute the Monte Carlo simulations and collect the results.

3. Compute the sensitivity measures corresponding to the raw and the rank data in order

to isolate the non-influential inputs.

4. Prepare the Monte Carlo sample for the variance-based sensitivity analysis, for the

influential variables identified important in the previous step.

5. Execute the Monte Carlo simulations and collect the results

6. Quantify the importance of the uncertain inputs, taken singularly as well as their

interactions.

The sensitivity indices have been calculated for 24 output variables listed in Table 11.

5.2 Screening uncertainty and sensitivity analysis

The relative importance of the 33 uncertain input factors is initially explored with the screening

design based on quasi-random LPτ sequence. A sample of 512 simulations was generated for

this screening test. The estimated sensitivity coefficients, calculated over the 24 output

variables listed in Table 11, are displayed in the 24 individual screens of Figure 12. In each

screen, the first column of graphs displays the sensitivity coefficients calculated on raw data

(PEAR, PCC, SRC), see section 3.2.2. The second column of graphs displays the sensitivity

coefficients calculated on ranks (SPEA, PRCC, SRRC). A value close to one (minus one) means

that the output variable is positively (negatively) correlated to the individual variable.

The linearity of the regression model (SRC, SRRC) is shown in the CMD graph (third column),

which gives the percentage of data variance explained by the regression model. The closer the

CMD value is to unit, the better the output coefficient can be modelled by a linear combination

of the input variables. Two lines are given in each graph. The line “data” corresponds to the

linearity of the output variables in respect to the input variables, while the line rank

corresponds to the linearity of the output variable ranks in respect to the input variable ranks.

According to the measures, the most influential input set, for all the output variables

considered, contains 14 entries:

� the average trip length (ltrip)

� the hot and cold emission factors (eEF, eEFratio, eEFfc, eEFfcratio)

� the (B,T) - parameter: characteristic service life & faillure steepness (paraB and paraT

pairs)

� the road vehicle basic purchase resource cost - EURO 2000 (eRPCSBASE)

� the estimated residual value function as a percentage of purchase cost

(usresidualparaAB)

� the repair and maintenance cost excluding taxes as % of purchase resource cost (ex

tax) (eRREPMAINTCFRACTION)

� the insurance cost as percentage of vehicle purchase resource cost (RINSCFRACTION)

� the labour cost - net wage - for truck drivers - EURO per hour (RLABOURC)

82

� the labour tax - bruto wage minus netto wage - for truck drivers - EURO per hour

(RLABOURTX)

� the annual ownership tax road vehicles - EURO 2005 (ROWNTX)

� the public transport fare cost coverage (PUBLICCOSTCOV)

It is evident, based on Figure 12 that the output variables belonging to the stock and cost

modules behave linearly in respect to the input variables uncertainty. TAXregistration,

COSTpurchase (2010 only) and VATpurchase (2010 only) are the least linear variables of those

subgroups with a minimum CMD of 0.85. The variables referring to the emission module have a

quasi-similar behaviour. The most non-linear variable is CO (CMD~0.79) followed by VOC

(CMD~0.85) while NOx and PM have CMD around 0.95.

Some first conclusions can already be obtained based on this screening test:

- The model output referring to activity and cost is mostly a linear combination of the

input variables. This means that second order effects, such as combined effects of two

or more variables or non-linear dependence of the model output to the input variables

are limited.

- A different manifestation of the previous point is that each output variable is mostly

determined by the corresponding input variable. Emissions are mostly correlated to

emission factors, costs are correlated with their individual cost input variables. Stock

and activity data are basically determined by purchase and fuel costs.

- Some interesting effects are observed, depending on the time horizon considered. For

example, the vehicle number in the short future is more correlated to the fuel price

than the purchase cost. However, fuel prices affect both new and existing vehicles,

therefore the correlation between number of vehicles and fuel costs becomes less

important into the future. On the other hand, purchase costs only affects new

registrations and not the existing stock. Therefore purchase cost becomes a much

more relevant parameter into the future. Such observations confirm both the correct

operation of the model and our screening analysis results.

- The linearity of the effects may at first seem striking for a complex model like

TREMOVE. On the other hand, one needs to take into consideration that the structure

and the values in the elasticities tree allows only limited flexibility to different choices.

- A relatively small number of input variables (14) seem to determine the model output.

The reduced set of uncertain inputs is further analyzed in the next section.

83

Table 11: List of the output variables analysed and their respective units

Module Output Variable Description Units

1 EMISS FC Fuel Consumption kTon

2 EMISS PM exhaust PM emissions kTon

3 EMISS CO exhaust CO emissions kTon

4 EMISS VOC exhaust VOC emissions kTon

5 EMISS NOx exhaust NOx emissions kTon

6 COST COSTpurchase Resource cost for purchase Euro

7 COST TAXregistration TAX for registration Euro

8 COST VATpurchase VAT for purchase Euro

9 COST TAXownership TAX for ownership Euro

10 COST COSTinsurance Resource cost for insurance Euro

11 COST TAXinsurance TAX cost for insurance Euro

12 COST COSTfuel Resource cost for fuel Euro

13 COST TAXfuel TAX cost for fuel Euro

14 COST VATfuel VAT cost for fuel Euro

15 COST COSTrepair Resource cost for repair and maintanance Euro

16 COST VATrepair VAT cost for repair and maintenance Euro

17 COST COSTlabour Resource cost for labour Euro

18 COST COSTlabourtax Labourtax is a resource cost Euro

19 COST COSTrest COSTrest contains monetary resource costs that can not

be allocated to other components Euro

20 COST TAXrest TAXrest contains monetary resource TAXes that can not be allocated to other components, eg public transport

subsidies are counted here as negative Euro

21 COST VATrest VATrest contains monetary resource VAT that can not be

allocated to other components Euro

22 COST Costs Total Cost Euro

23 STOCK Vehicles Number of vehicles #

24 STOCK VehKms Number of vehicle kilometres #

95

Figure 12: The sensitivity coefficients (PEAR, PCC, SRC, SPEA, PRCC, SRRC) calculated for each of the 24 output variables and the coefficient of model determination (CMD). The indices calculated on the raw (rank) data are shown in the first (second) column. The linearity of the regression model

(SRC, SRRC) is shown in the CMD graph, which gives the percentage of data variance explained by a linear regression model.

96

5.3 Variance-based uncertainty and sensitivity analysis

The 14 most influential parameters identified from the screening analysis were used next in a

quantitative sensitivity analysis. The target in this analysis is not only to qualitatively identify

which are the important input variables but also to quantitatively determine what is the total

uncertainty of the output and how much the uncertainty of each influential input variable

affects the variance of each output variable.

For this purpose, a sample was built by selecting 5950 design points over a particular space-

filling curve in the 14th dimensional input space so as to explore each factor with a different

frequency (Cukier et al. [4]). In modelling terms, this means that 5950 individual TREMOVE

runs were executed.

The results of this analysis, i.e. uncertainty in the calculation of the stock, activity and

emissions in the case UK is presented in Figure 13 while their descriptive statistics are given in

Table 12. The figures show the evolution of the medial of each output variable over time,

together with the 5th and the 95th probability percentiles. Such an uncertainty criterion was

selected because it can better facilitate the visualisation of the output also in case of non-

symmetrical distributions.

Observing the general trends, one observes that all output variables corresponding to the

demand or stock groups exhibit increase through time (with the exception of TAXrest), similar

to fuel consumption, while the opposite is true for the emission pollutants but not fuel

consumption. From a macroscopic point of view (i.e. real life experience) one may confirm that

the output of the 5950 runs is consistent to what one would expect.

The uncertainty of each output variable depends on the uncertainty of each individual input

variable (SI) individually but also combined. The first order sensitivity index shows how much

of the output variable uncertainty can be explained individually by the uncertainty of the

specific input variable. This is shown in Figure 14. A value of 1 would mean that the output

variance is fully explained by the variance of a single input variable. For several variables, the

output uncertainty is mostly determined by a single variable. For example, more than 80% of

the cost purchase is explained by the eRPCSBASE in the future.

The actual values of the sensitivity analysis are quoted in Table 13. The first order index (SI)

corresponds to the values also shown in Figure 14. Their summation (last raw per output

variable) shows how much of the total output variable uncertainty is explained by the additive

variance of all input variables. STI also shows higher order dependencies, i.e. how much of the

output variance is explained by each specific input variable in combination to the variance of

the other input variables. The difference between the total effect (STI) and the first order index

for an input variable (SI) indicates the fraction of the output variance that is accounted for by

interactions in which the specific input variable is involved. This means that the input variable

interacts with other input parameters but it does not indicate with which parameters this

interaction occurs.

97

Analytically:

Emiss Module:

FC: Fuel consumption is increasing in time but its uncertainty remains rather constant. 96% of

the FC variance in 2010 is explained by single contributions of the 14 variables, increasing

gradually to 97% in 2020 and 98% in 2030. Approximately 98% of the explained by single

contributions variance is due to only three variables: eEFfc (85% in 2010, 79% in 2020, 77%

in 2030), eEFfcratio (8% in 2010, 9% in 2020, 9% in 2030) and eRPCSBASE (1% in 2010, 7%

in 2020, 8% in 2030). The importance of eEFfc is always principal; however, it’s fraction of

explained variance exhibits a small decrease through time by the same amount that the first

order index of eRPCSBASE is increasing. Higher-order interactions are small and of the same

magnitude for all input factors. The behaviour of FC is identical with the variables belonging to

the stock module COSTfuel, TAXfuel and VATfuel.

PM: Particulate matter is decreasing through time while its uncertainty is relatively constant.

PM emissions variance explained by single contributions is due to the eEF (91% in 2010, 91%

in 2020, 93% in 2030). Other input factors identified are eEFratio and ltrip whose contribution

is less than 2% (taken singularly). Higher-order interactions are small.

CO: Carbon monoxide uncertainty is decreasing from 2010 to 2030 and so does its uncertainty.

The results are similar to those of PM. Uncertainty in the CO emissions is mostly influenced by

the hot emission factors, which taken singularly explain 87-90% of the variance. The decrease

in CO uncertainty with time is the expected gradual increase of diesel vehicles compared to

gasoline, as well as the better emission control expected from gasoline vehicles of the future.

The interaction effects of second and higher-order in the CO emissions are responsible for

about 9% of the total variance. The input factor exhibiting higher interactions is eEF, followed

by ROWNTX, RINSFRACTION and eEFratio.

VOC: The emissions and the uncertainty of the volatile organic compounds are decreasing. The

VOC uncertainty is roughly half compared to CO. 84% of the VOC emissions variance is

explained by the single contribution of the eEF in 2010 and drops to 84% in 2020-2030. Like

PM and CO, the fraction of the explained variance through interactions is decreasing in the

future, with a contribution from all variables but principally from eEF and eEFratio. The reasons

for this behaviour are similar to CO.

NOx: Nitrogen oxide emissions are decreasing with their uncertainty being relatively constant.

Like the other emissions, uncertainty in the NOx emissions is lumped principally to the

emission factors (eEF) whose self-contribution explain 94% of their variability in 2010,

dropping slightly to 91% and 92% in 2020 and 2030 respectively. Interactions are responsible

for less than 5% of the total variance.

98

Figure 13: Temporal evolution of the output uncertainty for UK. The bold line represents the

median while the dotted lines correspond to the 5th and 95th percentiles. Units for each variable are

given in Table 11. Negative TAXrest values indicate subsidies delivered to public transport.

99

Table 12: Descriptive statistics of the histograms presented in Figure 13 that belong to the cost,

emissions and vehicle module.

COSTpurchase [mil Euro] 2010 2020 2030

COSTrepair [mil Euro] 2010 2020 2030

Mean 84,320 99,596 114,760 mean 59,868 73,877 86,583

Median 84,178 99,323 114,467 median 59,798 73,706 86,483

st, deviation 3,628 8,375 10,187 st, deviation 1,827 7,361 8,801

TAXregistration

[mil Euro] 2010 2020 2030 VATrepair [mil Euro] 2010 2020 2030

mean 23 24 28 mean 7,302 9,110 10,699

median 23 24 27 median 7,293 9,090 10,691

st, deviation 2 2 2 st, deviation 218 896 1,073

VATpurchase

[mil Euro] 2010 2020 2030 COSTlabour [mil Euro] 2010 2020 2030

mean 10,861 11,599 13,137 mean 10,874 14,908 16,620

median 10,841 11,566 13,107 median 10,869 14,896 16,607

st, deviation 493 1,007 1,222 st, deviation 973 1,347 1,504

TAXownership

[mil Euro] 2010 2020 2030 COSTlabourtax

[mil Euro] 2010 2020 2030

mean 6,176 10,794 12,044 mean 11,629 15,941 17,771

median 6,166 10,810 12,070 median 11,634 15,944 17,773

st, deviation 494 1,161 1,403 st, deviation 1,023 1,394 1,554

COSTinsurance

[mil Euro] 2010 2020 2030 COSTrest [mil Euro] 2010 2020 2030

mean 24,974 38,372 44,742 mean 41,214 44,411 47,714

median 24,922 38,291 44,671 median 41,193 44,395 47,695

st, deviation 1,606 5,148 6,176 st, deviation 780 864 926

TAXinsurance

[mil Euro] 2010 2020 2030 TAXrest

[mil Euro] 2010 2020 2030

mean 1,263 1,938 2,259 mean -8,500 -8,818 -9,418

median 1,260 1,934 2,255 median -8,454 -8,768 -9,365

st, deviation 81 260 311 st, deviation 1,814 1,943 2,075

COSTfuel [mil Euro] 2010 2020 2030

VATrest [mil Euro] 2010 2020 2030

mean 30,963 32,309 40,455 mean 1,252 1,292 1,377

median 30,727 32,106 40,202 median 1,252 1,293 1,377

st, deviation 3,521 3,715 4,659 st, deviation 59 62 66

TAXfuel

[mil Euro] 2010 2020 2030 Costs

[mil Euro] 2010 2020 2030

mean 33,554 41,758 48,376 mean 323,890 396,704 458,755

median 33,301 41,490 48,064 median 323,333 396,036 458,228

st, deviation 3,866 4,841 5,619 st, deviation 7,930 16,826 19,578

VATfuel

[mil Euro] 2010 2020 2030

mean 8,119 9,490 11,497

median 8,041 9,418 11,412

st, deviation 1,082 1,261 1,519

FC

[ton] 2010 2020 2030 Nox

[kton] 2010 2020 2030

mean 46,962,996 55,921,530 60,405,978 mean 343,591 176,528 164,178

median 46,618,917 55,591,702 60,043,602 median 331,319 170,500 158,458

st, deviation 5,191,806 6,253,930 6,824,457 st, deviation 65,800 30,477 27,340

PM

[kton] 2010 2020 2030 VOC

[kton] 2010 2020 2030

mean 11,303 3,582 3,609 mean 52,523 32,702 33,974

median 10,750 3,400 3,421 median 46,927 30,279 31,559

st, deviation 2,881 966 964 st, deviation 19,662 8,358 8,258

100

CO [kton] 2010 2020 2030

mean 327,133 141,775 142,778

median 252,000 119,402 119,439

st, deviation 227,061 71,907 72,774

Vehicles

[#] 2010 2020 2030 VehKms [x10^6] 2010 2020 2030

mean 33,626,944 37,902,637 40,996,424 mean 585,216 665,651 720,489

median 33,652,081 37,918,723 40,997,888 median 585,653 665,914 720,553

st, deviation 576,196 1,154,149 1,288,394 st, deviation 9,886 19,654 21,921

� Stock Module:

o Vehicles & Veh-kms: Both variables increase in the future but their uncertainty

increases relatively more. 89% of the stock variance in 2010 is explained by

single contributions of the 14 variables; unlike the emission variables, the

linearity of the stock module is approximately constant (temporally) but at the

same time, the magnitude of the higher-order interactions is decreasing. Five

input factors control the variability, with the most important being eRPCSBASE

and eEFfc. The factor eRPCSBASE explains roughly 65% of the variance in the

future (28% in 2010) while the factor eEFfc exhibits a substantial drop down

to 12%, although in 2010 it is the principal uncertain factor responsible for

45% of the stock variability. Other contributing factors are usresidualparaAB,

eRREPMAINFRACTION and eEFfcratio.

� Cost Module:

o COSTpurchase - VATpurchase: Both variables increase in the future while their

uncertainty becomes double relative to its value in 2010.Approximately 97%

of their variance is explained by single contributions of the 14 variables;

among them, the most important single contribution is from the eRCPSBASE

which increases from 59% in 2010 to 85% in the future. The sum of all the

iS ’s is very close to 1 indicating that the model behaves almost additively

(with respect to the input parameters). Although the higher order interactions

are responsible for less than 5% of the total variance, they exhibit a

significant decrease in their absolute magnitude from 2010 to 2020 (it is about

half that of 2010). This should be explained by the dominant importance of

eRCPSBASE in the future.

o TAXregistration: TAXregistration increases in the future while keeping the

same level of uncertainty. Uncertainty in TAXregistration is principally lumped

to uparaBT that explain ~85-86% of its variability. The second important

factor is eRCPSBASE. The sum of all the iS ’s is 0.89 in 2010 and becomes

0.92 in the future. Between all examined output variables, TAXregistration

shows the highest magnitude of interactions that, in addition, are relatively

constant and arise from all input factors.

o TAXownership: TAXownership in 2030 is double relative to 2010 but its

uncertainty is increased to a lesser extend (factor of 1.5). Approximately 95%

of the variance is explained by single contributions of the 14 variables. This

fraction is explained principally by the single contribution of two variables:

ROWNTX (89% in 2010, 56% in 2020, 48% in 2030) and eRPCSBASE (4% in

2010, 34% in 2020, 41% in 2030). In 2010, the variability is controlled by the

101

ROWNTX while in 2030 both ROWNTX and eRPCSBASE are influencing

TAXownership. Interactions are unimportant.

o COSTinsurance - TAXinsurance: Insurance variables are doubled in 2030 in

comparison to 2010 and their uncertainty is increased by the same factor.

About 94% of their variance is explained by single contributions of the 14

variables; in 2010, this is attributed to three variables: eRPCSBASE (36%),

RINSCFRACTION (47%) and eEFfc (8%). In the future, eEFfc is unimportant

while the importance of the other two variables is increased; it reaches 42%

and 51% for eRPCSBASE and RINSCFRACTION respectively. Interactions are

negligible for all input factors except for RLABOURC.

o COSTfuel – TAXfuel - VATfuel: their behaviour is identical with FC.

o COSTrepair - VATrepair: Those variables demonstrate a 50% increase in 2030

over the 2010 values but their uncertainty shows an increase by a factor

higher than 3. The explained by single contributions variance is ~ 98% and is

due to three variables: eRPCSBASE, RINSCFRACTION and eEFfc with higher

contributions from eRPCSBASE and RINSCFRACTION. Analytically, the

explained variance is ~45% for eRPCSBASE, ~41-52% for RINSCFRACTION

and ~1-7% for eEFfc. RINSCFRACTION becomes the most important factor in

the future while at the same time, eEFfc turns into unimportant.

o COSTlabour: COSTlabour increases in the future while keeping the same level

of uncertainty. Uncertainty in COSTlabour is lumped solely to RLABOURC that

explain ~94% of their variability (the sum of all the iS ’s is 0.96).

o COSTlabourtax: Like COSTlabour, COSTlabourtax increases in the future while

keeping the same level of uncertainty. Uncertainty in COSTlabourtax is lumped

solely to RLABOURTX that explain ~94% of their variability (the sum of all the

iS ’s is 0.96).

o COSTrest: COSTrest increases in the future but maintains the same level of

uncertainty. The explained by single contributions variance is ~96% and is due

principally to PUBLICCOSTCOV (90% in 2010, 81% in 2020, 80% in 2030).

The other two factors of less importance are eEFfc (constantly 4%) and

eRCPSBASE (1% in 2010, 8% in 2020, 9% in 2030).

o TAXrest - VATrest: Both variables increase (in absolute values) in the future

while preserving their uncertainty magnitude. Uncertainty in TAXrest and

VATrest is lumped solely to PUBLICCOSTCOV that explain ~96% of their

variability, i.e. the sum of all the iS ’s.

o Costs: The variable increase in the future but its uncertainty increases

relatively more. It exhibit very high and significant correlation (-0.99) with the

stock variables. Five input factors control the variability, with the most

important being eRPCSBASE and eEFfc. The factor eRPCSBASE explains

roughly 59% of the variance in the future (30% in 2010) while the factor

eEFfc exhibits a substantial drop down to 14%, although in 2010 it is the

principal uncertain factor responsible for 41% of the stock variability. Other

contributing factors are usresidualparaAB, eRREPMAINFRACTION and

eEFfcratio.

102

Figure 14: Sensitivity coefficients of the model output based on the extended-FAST method.

103

Table 13: First and Total Order Sensitivity Indices (extended-FAST) for the output variables listed in

Table 11.

COSTpurchase SI 2010 SI

2020 SI 2030 STI

2010 STI 2020 STI

2030 uparaBT 0.05 0.01 0.01 0.17 0.05 0.05 eRPCSBASE 0.59 0.85 0.85 0.73 0.91 0.91 usresidualparaAB 0.16 0.04 0.04 0.28 0.09 0.09 eEF 0.01 0.00 0.00 0.13 0.04 0.04 eEFratio 0.01 0.00 0.00 0.11 0.03 0.03 ltrip 0.01 0.00 0.00 0.12 0.04 0.04 eRREPMAINTCFRACTION 0.01 0.00 0.00 0.11 0.04 0.04 RINSCFRACTION 0.01 0.00 0.00 0.13 0.04 0.04 RLABOURC 0.01 0.00 0.00 0.10 0.03 0.03 RLABOURTX 0.01 0.00 0.00 0.12 0.04 0.04 ROWNTX 0.01 0.00 0.00 0.13 0.04 0.04 PUBLICCOSTCOV 0.01 0.00 0.00 0.11 0.04 0.04 eEFfc 0.08 0.02 0.02 0.19 0.06 0.05 eEFfcratio 0.02 0.00 0.00 0.17 0.05 0.05 SUM 0.97 0.95 0.95 TAXregistration SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.85 0.85 0.86 0.98 0.96 0.96 eRPCSBASE 0.01 0.03 0.03 0.13 0.13 0.13 usresidualparaAB 0.00 0.00 0.00 0.12 0.10 0.10 eEF 0.00 0.00 0.00 0.13 0.11 0.10 eEFratio 0.00 0.00 0.00 0.13 0.11 0.10 ltrip 0.00 0.00 0.00 0.13 0.11 0.11 eRREPMAINTCFRACTION 0.00 0.00 0.00 0.12 0.11 0.10 RINSCFRACTION 0.00 0.00 0.00 0.12 0.10 0.09 RLABOURC 0.01 0.01 0.01 0.13 0.11 0.11 RLABOURTX 0.00 0.00 0.00 0.13 0.10 0.10 ROWNTX 0.00 0.00 0.00 0.12 0.10 0.10 PUBLICCOSTCOV 0.00 0.00 0.00 0.13 0.11 0.10 eEFfc 0.00 0.00 0.00 0.12 0.10 0.10 eEFfcratio 0.00 0.00 0.00 0.12 0.10 0.10 SUM 0.89 0.92 0.92 VATpurchase SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.01 0.00 0.00 0.15 0.04 0.04 eRPCSBASE 0.56 0.84 0.84 0.71 0.91 0.91 usresidualparaAB 0.19 0.05 0.05 0.32 0.09 0.10 eEF 0.01 0.00 0.00 0.15 0.04 0.04 eEFratio 0.01 0.00 0.00 0.12 0.04 0.04 ltrip 0.01 0.00 0.00 0.13 0.04 0.04 eRREPMAINTCFRACTION 0.01 0.00 0.00 0.12 0.04 0.04 RINSCFRACTION 0.01 0.01 0.01 0.15 0.05 0.05 RLABOURC 0.01 0.00 0.00 0.12 0.03 0.04 RLABOURTX 0.01 0.00 0.00 0.13 0.04 0.04 ROWNTX 0.01 0.00 0.00 0.14 0.04 0.04 PUBLICCOSTCOV 0.01 0.00 0.00 0.12 0.04 0.04 eEFfc 0.10 0.03 0.02 0.21 0.06 0.06 eEFfcratio 0.02 0.01 0.01 0.18 0.05 0.05 SUM 0.97 0.94 0.94 TAXownership SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.02 0.02 0.02 eRPCSBASE 0.04 0.34 0.41 0.07 0.37 0.45 usresidualparaAB 0.00 0.00 0.00 0.03 0.02 0.02 eEF 0.00 0.00 0.00 0.02 0.02 0.02 eEFratio 0.00 0.00 0.00 0.02 0.02 0.02 ltrip 0.00 0.00 0.00 0.02 0.02 0.02 eRREPMAINTCFRACTION 0.00 0.03 0.04 0.03 0.05 0.06 RINSCFRACTION 0.00 0.00 0.00 0.02 0.02 0.02 RLABOURC 0.00 0.00 0.00 0.02 0.02 0.02 RLABOURTX 0.00 0.00 0.00 0.02 0.02 0.02 ROWNTX 0.89 0.56 0.48 0.93 0.60 0.51 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.02 0.01 eEFfc 0.02 0.01 0.01 0.05 0.03 0.03 eEFfcratio 0.00 0.00 0.00 0.03 0.02 0.02 SUM 0.96 0.95 0.95

104

COSTinsurance SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.01 0.00 0.00 eRPCSBASE 0.36 0.42 0.42 0.41 0.46 0.45 usresidualparaAB 0.02 0.00 0.00 0.04 0.01 0.01 eEF 0.00 0.00 0.00 0.02 0.01 0.01 eEFratio 0.00 0.00 0.00 0.02 0.01 0.01 ltrip 0.00 0.00 0.00 0.02 0.01 0.01 eRREPMAINTCFRACTION 0.00 0.01 0.01 0.02 0.01 0.01 RINSCFRACTION 0.47 0.51 0.51 0.51 0.53 0.54 RLABOURC 0.01 0.00 0.00 0.15 0.13 0.12 RLABOURTX 0.00 0.00 0.00 0.01 0.00 0.00 ROWNTX 0.00 0.00 0.00 0.02 0.01 0.01 PUBLICCOSTCOV 0.00 0.00 0.00 0.01 0.00 0.00 eEFfc 0.08 0.01 0.01 0.10 0.02 0.02 eEFfcratio 0.01 0.00 0.00 0.02 0.01 0.01 SUM 0.94 0.94 0.95 TAXinsurance SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.01 0.00 0.00 eRPCSBASE 0.36 0.42 0.41 0.41 0.45 0.45 usresidualparaAB 0.02 0.00 0.00 0.04 0.01 0.01 eEF 0.00 0.00 0.00 0.02 0.01 0.01 eEFratio 0.00 0.00 0.00 0.02 0.01 0.01 ltrip 0.00 0.00 0.00 0.02 0.01 0.01 eRREPMAINTCFRACTION 0.00 0.01 0.01 0.02 0.01 0.01 RINSCFRACTION 0.47 0.51 0.51 0.51 0.54 0.54 RLABOURC 0.01 0.00 0.00 0.15 0.13 0.12 RLABOURTX 0.00 0.00 0.00 0.01 0.00 0.00 ROWNTX 0.00 0.00 0.00 0.02 0.01 0.01 PUBLICCOSTCOV 0.00 0.00 0.00 0.01 0.00 0.00 eEFfc 0.07 0.01 0.01 0.10 0.02 0.02 eEFfcratio 0.01 0.00 0.00 0.02 0.01 0.01 SUM 0.94 0.94 0.95 COSTfuel SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.02 0.02 0.02 eRPCSBASE 0.01 0.06 0.08 0.03 0.09 0.11 usresidualparaAB 0.00 0.00 0.00 0.02 0.02 0.02 eEF 0.00 0.00 0.00 0.02 0.02 0.02 eEFratio 0.00 0.00 0.00 0.02 0.02 0.02 ltrip 0.02 0.02 0.02 0.04 0.04 0.04 eRREPMAINTCFRACTION 0.00 0.01 0.01 0.02 0.03 0.03 RINSCFRACTION 0.00 0.00 0.00 0.02 0.03 0.03 RLABOURC 0.00 0.00 0.00 0.02 0.01 0.01 RLABOURTX 0.00 0.00 0.00 0.02 0.02 0.02 ROWNTX 0.00 0.00 0.00 0.02 0.02 0.02 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.02 0.02 eEFfc 0.85 0.79 0.77 0.88 0.82 0.80 eEFfcratio 0.08 0.09 0.09 0.09 0.10 0.10 SUM 0.96 0.97 0.97 TAXfuel SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030


105

VATfuel SI 2010 SI

2020 SI 2030 STI

2010 STI 2020 STI

2030 uparaBT 0.00 0.00 0.00 0.02 0.02 0.02 eRPCSBASE 0.01 0.06 0.09 0.03 0.09 0.11 usresidualparaAB 0.00 0.00 0.00 0.02 0.02 0.02 eEF 0.00 0.00 0.00 0.02 0.02 0.02 eEFratio 0.00 0.00 0.00 0.02 0.02 0.02 ltrip 0.02 0.02 0.02 0.04 0.04 0.04 eRREPMAINTCFRACTION 0.00 0.01 0.01 0.02 0.03 0.03 RINSCFRACTION 0.00 0.00 0.00 0.02 0.03 0.03 RLABOURC 0.00 0.00 0.00 0.02 0.01 0.01 RLABOURTX 0.00 0.00 0.00 0.02 0.02 0.02 ROWNTX 0.00 0.00 0.00 0.02 0.02 0.02 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.02 0.02 eEFfc 0.85 0.79 0.77 0.88 0.82 0.80 eEFfcratio 0.08 0.09 0.09 0.09 0.10 0.10 SUM 0.96 0.97 0.97 COSTrepair SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.01 0.00 0.00 0.03 0.02 0.02 eRPCSBASE 0.45 0.46 0.45 0.53 0.50 0.49 usresidualparaAB 0.02 0.00 0.00 0.05 0.02 0.02 eEF 0.00 0.00 0.00 0.04 0.02 0.02 eEFratio 0.00 0.00 0.00 0.07 0.04 0.04 ltrip 0.00 0.00 0.00 0.03 0.02 0.02 eRREPMAINTCFRACTION 0.41 0.51 0.52 0.45 0.54 0.55 RINSCFRACTION 0.00 0.00 0.00 0.02 0.01 0.01 RLABOURC 0.00 0.00 0.00 0.01 0.01 0.01 RLABOURTX 0.00 0.00 0.00 0.02 0.01 0.01 ROWNTX 0.00 0.00 0.00 0.02 0.01 0.01 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.01 0.01 eEFfc 0.07 0.01 0.01 0.09 0.01 0.02 eEFfcratio 0.01 0.00 0.00 0.02 0.01 0.01 SUM 0.97 0.98 0.99 VATrepair SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.03 0.02 0.02 eRPCSBASE 0.46 0.47 0.46 0.55 0.52 0.51 usresidualparaAB 0.02 0.00 0.00 0.05 0.02 0.02 eEF 0.00 0.00 0.00 0.05 0.02 0.02 eEFratio 0.00 0.00 0.00 0.08 0.04 0.04 ltrip 0.00 0.00 0.00 0.03 0.02 0.02 eRREPMAINTCFRACTION 0.38 0.50 0.51 0.43 0.53 0.54 RINSCFRACTION 0.00 0.00 0.00 0.03 0.01 0.01 RLABOURC 0.00 0.00 0.00 0.02 0.01 0.01 RLABOURTX 0.00 0.00 0.00 0.02 0.01 0.01 ROWNTX 0.00 0.00 0.00 0.02 0.01 0.01 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.01 0.01 eEFfc 0.10 0.01 0.01 0.11 0.02 0.02 eEFfcratio 0.01 0.00 0.00 0.03 0.01 0.01 SUM 0.98 0.99 0.99 COSTlabour SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030


106

COSTlabourtax SI 2010 SI

2020 SI 2030 STI

2010 STI 2020 STI

2030 uparaBT 0.00 0.00 0.00 0.02 0.02 0.02 eRPCSBASE 0.00 0.01 0.01 0.02 0.03 0.03 usresidualparaAB 0.00 0.00 0.00 0.02 0.02 0.02 eEF 0.00 0.00 0.00 0.02 0.02 0.02 eEFratio 0.00 0.00 0.00 0.02 0.02 0.02 ltrip 0.00 0.00 0.00 0.02 0.02 0.02 eRREPMAINTCFRACTION 0.00 0.00 0.00 0.02 0.02 0.02 RINSCFRACTION 0.00 0.00 0.00 0.02 0.02 0.02 RLABOURC 0.02 0.02 0.02 0.03 0.04 0.04 RLABOURTX 0.94 0.92 0.92 0.98 0.96 0.96 ROWNTX 0.00 0.00 0.00 0.02 0.02 0.02 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.02 0.02 eEFfc 0.00 0.00 0.00 0.03 0.02 0.02 eEFfcratio 0.00 0.00 0.00 0.02 0.02 0.02 SUM 0.96 0.96 0.96 COSTrest SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.02 0.02 0.02 eRPCSBASE 0.01 0.08 0.09 0.04 0.11 0.12 usresidualparaAB 0.00 0.00 0.00 0.04 0.04 0.04 eEF 0.00 0.00 0.00 0.02 0.02 0.02 eEFratio 0.00 0.00 0.00 0.03 0.03 0.03 ltrip 0.00 0.00 0.00 0.02 0.02 0.02 eRREPMAINTCFRACTION 0.00 0.01 0.01 0.03 0.03 0.03 RINSCFRACTION 0.00 0.00 0.00 0.03 0.02 0.02 RLABOURC 0.00 0.00 0.00 0.04 0.04 0.04 RLABOURTX 0.00 0.00 0.00 0.03 0.03 0.03 ROWNTX 0.00 0.00 0.00 0.02 0.02 0.02 PUBLICCOSTCOV 0.90 0.81 0.80 0.94 0.85 0.84 eEFfc 0.04 0.04 0.04 0.08 0.08 0.08 eEFfcratio 0.01 0.01 0.01 0.04 0.04 0.04 SUM 0.97 0.96 0.95 TAXrest SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.02 0.02 0.02 eRPCSBASE 0.00 0.00 0.00 0.03 0.03 0.03 usresidualparaAB 0.00 0.00 0.00 0.03 0.03 0.03 eEF 0.00 0.00 0.00 0.02 0.02 0.02 eEFratio 0.00 0.00 0.00 0.03 0.03 0.03 ltrip 0.00 0.00 0.00 0.02 0.02 0.02 eRREPMAINTCFRACTION 0.00 0.00 0.00 0.02 0.02 0.02 RINSCFRACTION 0.00 0.00 0.00 0.02 0.02 0.02 RLABOURC 0.00 0.00 0.00 0.03 0.03 0.03 RLABOURTX 0.00 0.00 0.00 0.03 0.03 0.03 ROWNTX 0.00 0.00 0.00 0.02 0.02 0.02 PUBLICCOSTCOV 0.96 0.96 0.96 1.00 1.00 1.00 eEFfc 0.00 0.00 0.00 0.04 0.04 0.04 eEFfcratio 0.00 0.00 0.00 0.04 0.04 0.04 SUM 0.96 0.96 0.96 VATrest SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030


107

Costs SI 2010 SI

2020 SI 2030 STI

2010 STI 2020 STI

2030 uparaBT 0.00 0.00 0.00 0.03 0.02 0.02 eRPCSBASE 0.30 0.59 0.59 0.33 0.63 0.62 usresidualparaAB 0.03 0.01 0.01 0.07 0.04 0.04 eEF 0.00 0.00 0.00 0.03 0.02 0.02 eEFratio 0.00 0.00 0.00 0.04 0.03 0.03 ltrip 0.01 0.00 0.00 0.04 0.02 0.02 eRREPMAINTCFRACTION 0.05 0.08 0.08 0.10 0.11 0.11 RINSCFRACTION 0.01 0.02 0.02 0.04 0.03 0.03 RLABOURC 0.01 0.01 0.01 0.06 0.03 0.03 RLABOURTX 0.01 0.00 0.00 0.03 0.02 0.02 ROWNTX 0.01 0.00 0.00 0.05 0.02 0.02 PUBLICCOSTCOV 0.02 0.01 0.00 0.05 0.02 0.02 eEFfc 0.41 0.14 0.14 0.45 0.16 0.16 eEFfcratio 0.03 0.01 0.01 0.05 0.02 0.02 SUM 0.88 0.88 0.88 FC SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.00 0.00 0.02 0.02 0.02 eRPCSBASE 0.01 0.07 0.08 0.03 0.09 0.11 usresidualparaAB 0.00 0.00 0.00 0.02 0.02 0.02 eEF 0.00 0.00 0.00 0.02 0.02 0.02 eEFratio 0.00 0.00 0.00 0.02 0.02 0.02 ltrip 0.02 0.02 0.02 0.04 0.04 0.04 eRREPMAINTCFRACTION 0.00 0.01 0.01 0.02 0.03 0.03 RINSCFRACTION 0.00 0.00 0.00 0.02 0.03 0.03 RLABOURC 0.00 0.00 0.00 0.02 0.01 0.01 RLABOURTX 0.00 0.00 0.00 0.02 0.02 0.02 ROWNTX 0.00 0.00 0.00 0.02 0.02 0.02 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.02 0.02 eEFfc 0.85 0.79 0.77 0.88 0.82 0.80 eEFfcratio 0.08 0.09 0.09 0.09 0.10 0.10 SUM 0.96 0.97 0.98 PM SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.01 0.01 0.00 0.03 0.02 0.01 eRPCSBASE 0.00 0.01 0.01 0.02 0.03 0.02 usresidualparaAB 0.00 0.00 0.00 0.01 0.01 0.01 eEF 0.91 0.91 0.93 0.95 0.95 0.96 eEFratio 0.02 0.01 0.01 0.05 0.04 0.03 ltrip 0.02 0.01 0.01 0.05 0.03 0.03 eRREPMAINTCFRACTION 0.00 0.00 0.00 0.02 0.02 0.02 RINSCFRACTION 0.00 0.00 0.00 0.02 0.02 0.02 RLABOURC 0.00 0.00 0.00 0.02 0.02 0.02 RLABOURTX 0.00 0.00 0.00 0.02 0.02 0.02 ROWNTX 0.00 0.00 0.00 0.02 0.02 0.02 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.02 0.02 eEFfc 0.00 0.00 0.00 0.02 0.02 0.02 eEFfcratio 0.00 0.00 0.00 0.02 0.02 0.02 SUM 0.97 0.96 0.96 CO SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030


108

VOC SI 2010 SI

2020 SI 2030 STI

2010 STI 2020 STI

2030 uparaBT 0.01 0.00 0.00 0.04 0.02 0.02 eRPCSBASE 0.00 0.02 0.02 0.03 0.05 0.04 usresidualparaAB 0.01 0.00 0.00 0.03 0.02 0.02 eEF 0.88 0.84 0.84 0.95 0.92 0.91 eEFratio 0.02 0.03 0.04 0.07 0.09 0.09 ltrip 0.01 0.01 0.01 0.06 0.05 0.05 eRREPMAINTCFRACTION 0.00 0.00 0.00 0.04 0.04 0.03 RINSCFRACTION 0.00 0.00 0.00 0.05 0.04 0.04 RLABOURC 0.00 0.00 0.00 0.04 0.04 0.03 RLABOURTX 0.00 0.00 0.00 0.04 0.03 0.03 ROWNTX 0.00 0.00 0.00 0.05 0.05 0.05 PUBLICCOSTCOV 0.00 0.00 0.00 0.04 0.03 0.03 eEFfc 0.00 0.00 0.00 0.04 0.04 0.03 eEFfcratio 0.00 0.00 0.00 0.04 0.04 0.03 SUM 0.93 0.91 0.91 NOx SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.00 0.01 0.00 0.02 0.02 0.02 eRPCSBASE 0.00 0.02 0.01 0.02 0.04 0.03 usresidualparaAB 0.00 0.00 0.00 0.02 0.02 0.02 eEF 0.94 0.91 0.92 0.98 0.96 0.97 eEFratio 0.01 0.01 0.01 0.04 0.04 0.03 ltrip 0.00 0.00 0.00 0.02 0.03 0.03 eRREPMAINTCFRACTION 0.00 0.00 0.00 0.02 0.03 0.03 RINSCFRACTION 0.00 0.00 0.00 0.03 0.03 0.03 RLABOURC 0.00 0.00 0.00 0.02 0.03 0.03 RLABOURTX 0.00 0.00 0.00 0.02 0.02 0.03 ROWNTX 0.00 0.00 0.00 0.03 0.03 0.03 PUBLICCOSTCOV 0.00 0.00 0.00 0.02 0.03 0.03 eEFfc 0.00 0.00 0.00 0.02 0.03 0.03 eEFfcratio 0.00 0.00 0.00 0.02 0.03 0.03 SUM 0.96 0.95 0.95 Vehicles SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030

uparaBT 0.01 0.00 0.00 0.04 0.02 0.02 eRPCSBASE 0.28 0.65 0.64 0.33 0.69 0.68 usresidualparaAB 0.05 0.02 0.02 0.11 0.04 0.04 eEF 0.00 0.00 0.00 0.03 0.02 0.02 eEFratio 0.00 0.00 0.00 0.05 0.02 0.02 ltrip 0.01 0.00 0.00 0.05 0.02 0.02 eRREPMAINTCFRACTION 0.04 0.07 0.07 0.10 0.09 0.09 RINSCFRACTION 0.01 0.01 0.01 0.04 0.03 0.03 RLABOURC 0.00 0.00 0.00 0.07 0.03 0.03 RLABOURTX 0.00 0.00 0.00 0.04 0.02 0.02 ROWNTX 0.00 0.00 0.00 0.05 0.03 0.03 PUBLICCOSTCOV 0.00 0.00 0.00 0.05 0.02 0.02 eEFfc 0.45 0.12 0.12 0.50 0.14 0.14 eEFfcratio 0.03 0.01 0.01 0.06 0.02 0.03 SUM 0.89 0.88 0.88 VehKms SI

2010 SI 2020 SI

2030 STI 2010 STI

2020 STI 2030


A more detailed representation of the temporal evolution of the output uncertainty, for all

vehicle types, compared against the TREMOVE basecase can be found in the ANNEX.

109

5.4 Discussion

A summary of the uncertainty analysis of TREMOVE application in UK is shown in Table 14. The

median value calculated together with the coefficient of variance (cov=standard deviation over

mean) is shown per output variable. The output variables are in a decreasing order of cov in the

year 2030. The results are shown in summary for all vehicle categories. The results of the

uncertainty analysis per vehicle category are shown in the Annex.

Table 14: Summary of Uncertainty Analysis results for UK

Output Variable Units Median 2010 Median 2020 Median 2030 cov 2010 cov 2020 cov 2030

CO Ton 252,000 119,402 119,439 69% 51% 51%

PM Ton 10,750 3,400 3,421 25% 27% 27%

VOC Ton 46,927 30,279 31,559 37% 26% 24%

TAXrest M€ -8,454 -8,768 -9,365 21% 22% 22%

NOx Ton 331,319 170,500 158,458 19% 17% 17%

COSTinsurance M€ 24,922 38,291 44,671 6% 13% 14%

TAXinsurance M€ 1,260 1,934 2,255 6% 13% 14%

VATfuel M€ 8,041 9,418 11,412 13% 13% 13%

TAXfuel M€ 33,301 41,490 48,064 12% 12% 12%

COSTfuel M€ 30,727 32,106 40,202 11% 11% 12%

TAXownership M€ 6,166 10,810 12,070 8% 11% 12%

FC Ton 46,618,917 55,591,702 60,043,602 11% 11% 11%

COSTrepair M€ 59,798 73,706 86,483 3% 10% 10%

VATrepair M€ 7,293 9,090 10,691 3% 10% 10%

COSTlabour M€ 10,869 14,896 16,607 9% 9% 9%

COSTlabourtax M€ 11,634 15,944 17,773 9% 9% 9%

VATpurchase M€ 10,841 11,566 13,107 5% 9% 9%

COSTpurchase M€ 84,178 99,323 114,467 4% 8% 9%


VATrest M€ 1,252 1,293 1,377 5% 5% 5%

Costs M€ 323,333 396,036 458,228 2% 4% 4%

Vehicles # 33,652,081 37,918,723 40,997,888 2% 3% 3%

Vehkms ×106 km 585,653 665,914 720,553 2% 3% 3%

COSTrest M€ 41,193 44,395 47,695 2% 2% 2%

The most important conclusions drawn from the uncertainty analysis include:

� The uncertainty is large for the emission of pollutants, mostly due to the uncertainty in the

emission factors. This is in the order of 20-30% but can reach up to 50% in the case of CO.

� Fuel dependent variables are second with regard to output uncertainty with cov values in

the order of 10-15%.

� Total cost figures exhibit uncertainty ranges in the order of 4-10%, i.e. they are rather

small depended on the variance of input data.

� Finally, population and activity data exhibit very little uncertainty, in the order of 2-3%. The

uncertainty per vehicle category (see Annex) is of the same magnitude. This shows that the

activity and population is little elastic to changes in costs and other variables in the model.

This is potentially very important. This means that changes in the activity and population

data are very limited despite the much wider uncertainty ranges assumed for the input

variables.

110

� The levels of all output variables corresponding to the demand or stock group increase

through time (with the exception of TAXrest). For the emissions group the opposite is true

except for the fuel consumption. This is expected assuming that road transport activity will

continue to increase in the future while emission standards will continue to effectively

control emissions. The increase in fuel consumption remains to be seen – it is reminded

that the TREMOVE version examined does not include hybrid vehicles, or electric vehicles.

� The variables corresponding to the demand or stock group that demonstrate an increase in

their uncertainty in the future are: COSTpurchase, VATpurchase, TAXownership,

COSTinsurance, TAXinsurance, COSTrepair, VATrepair, Costs, Vehicles and Vehkms. The

highest increase has been seen for COSTrepair and VATrepair. The variables corresponding

to the emissions group that demonstrate a decrease in their uncertainty in the future are

VOC and CO. The largest decrease has been seen for CO, however its uncertainty is always

the highest among the 24 considered output variables. All the other variables that are not

mentioned exhibit a very small variability in their temporal uncertainty. The large but

decreasing uncertainty of CO and VOC is caused by their large dependence on the emission

factors of gasoline cars. As gasoline cars are both expected to become less important in the

future (compared to diesel ones) but as also Euro 5 and 6 emission standards are expected

to lead to even more strict control over the lifetime of the vehicles, the observed trend is

explainable.

With regard to the sensitivity of the output to the input variables Table 15 shows a summary of

the first order and higher order contribution of each input variable to the output variables

uncertainty. The values have been ranked in a descending order according to the SI2030

value. Based on the sensitivity analysis, the following conclusions may be drawn:

Table 15: Summary of Sensitivity Analysis results for UK

Output Variable

Most Important Input Variable ΣSI

2010 ΣSI2020 ΣSI

2030 ΣSTI2010 ΣSTI

2020 ΣSTI2030

COSTrepair eRREPMAINTCFRACTION, eRPCSBASE 0.97 0.98 0.99 1.41 1.22 1.23

VATrepair eRREPMAINTCFRACTION, eRPCSBASE 0.98 0.99 0.99 1.48 1.23 1.24

FC eEFfc 0.96 0.97 0.98 1.26 1.27 1.27

COSTfuel eEFfc 0.96 0.97 0.97 1.26 1.27 1.27

TAXfuel eEFfc 0.96 0.97 0.97 1.26 1.27 1.27

VATfuel eEFfc 0.96 0.97 0.97 1.26 1.27 1.27

COSTlabour RLABOURC 0.96 0.96 0.96 1.16 1.16 1.16

COSTlabourtax RLABOURTX 0.96 0.96 0.96 1.25 1.24 1.24

TAXrest PUBLICCOSTCOV 0.96 0.96 0.96 1.36 1.36 1.36

VATrest PUBLICCOSTCOV 0.96 0.96 0.96 1.36 1.36 1.36

PM eEF 0.97 0.96 0.96 1.26 1.24 1.21

COSTpurchase eRPCSBASE 0.97 0.95 0.95 2.62 1.49 1.48

TAXownership ROWNTX 0.96 0.95 0.95 1.32 1.25 1.23

COSTinsurance RINSCFRACTION 0.94 0.94 0.95 1.36 1.21 1.20

TAXinsurance RINSCFRACTION 0.94 0.94 0.95 1.36 1.21 1.20

COSTrest PUBLICCOSTCOV 0.97 0.96 0.95 1.38 1.35 1.35

NOx eEF 0.96 0.95 0.95 1.27 1.31 1.31

VATpurchase eRPCSBASE 0.97 0.94 0.94 2.76 1.50 1.53

TAXregistration uparaBT 0.89 0.92 0.92 2.61 2.35 2.30

111

CO eEF 0.91 0.92 0.91 1.61 1.52 1.54

VOC eEF 0.93 0.91 0.91 1.50 1.45 1.42

Costs eRPCSBASE, eEFfc 0.88 0.88 0.88 1.37 1.16 1.16

Vehicles eRPCSBASE, eEFfc 0.89 0.88 0.88 1.51 1.19 1.19

Vehkms eRPCSBASE, eEFfc 0.89 0.88 0.88 1.51 1.18 1.18

� Ten out of the fourteen uncertain inputs are of principal importance for one or more output

variables.

� The hot emission factors (eEF) influences most the variability of the emissions (VOC, NOX,

PM, CO) while the basic road vehicle purchase resource cost (eRCPSBASE) controls the

variability of the stock and activity variables (vehicles and vehicle-kms). On the other hand,

many input factors are responsible for the variability of the cost related output.

� All model outputs exhibit high linearity. The least amount of explained-by-single

contributions variance estimated was 88% and corresponds to the output variables Costs,

Vehicles and Vehkms. The remaining fraction depends on higher order interdependencies

between the input variables.

� The linearity of the output variables is generally constant in time. On the other hand, the

total effects are either constant or decreasing in the future. This implies that the control of

the singular perturbations will be always effective.

� The highest amount of interaction effects has been seen for TAXregistration but it is

attributed to all input factors. The highest decrease in the effect of the second and higher

order terms has been observed for COSTpurchase and VATpurchase. This gives the

opportunity to work on the uncertainty of the RPCSBASE only in the future in order to

reduce the variability of COSTpurchase and VATpurchase.

It should be made clear that the conclusions reached from this analysis are specific to UK only.

Selecting a different country to perform the analysis would have an impact on both the total

uncertainty ranges estimated (Table 14) and the impact of each input variable uncertainty to the

output (Table 15). For example, repeating the analysis in the case of Greece would lead to

different cost and pollutants uncertainty as the Greek fleet does not include diesel passenger

cars except of taxis. This would have an impact on the contribution of uncertainty from gasoline

vs diesel passenger car emission factors. Similarly, uncertainty ranges for several values in

some of the late Member States (e.g. Bulgaria, Romania) would have been relatively wider than

in UK, mostly due to relatively poorer statistics available in these countries. The particular

numerical values that would have been produced would be interest in a national environment

where one needs to assess the uncertainty of the national emission calculations. However, they

would offer limited additional insights in our work. This is mainly because the uncertainty ranges

between the different output categories are clearly distinct, with activity data, cost items, and

total emissions to be on completely different levels (CV of ~2% for activity, ~10% for costs,

>20% for emissions). This is due to the structure of the TREMOVE model, i.e. its rather linear

behaviour of output to input values and the limited impact to total activity. Using a different

uncertainty range for some of the input variables would have a rather limited impact on activity

and a proportional impact to the affected cost category and a limited impact on emissions. Such

an effect does not change the main conclusions reached in this study concerning the treatment

of uncertainty and ways of improving it in future applications of Tremove.

112

6 Uncertainty and sensitivity analysis of scenarios

6.1 General

TREMOVE is by definition used to calculate the impact of different policy options, i.e. executing

scenarios to identify what is the impact of different policies on activity data, costs, and finally

emissions. It is therefore important to analyze the uncertainty of a number of different

scenarios in comparison to the baseline. In order to demonstrate this, we decided to run three

alternative scenarios that may activate different paths of uncertainty of the model. The three

scenarios considered were the following:

1. Increase the ownership tax of passenger cars to demonstrate shifts to other modes of

transport within the road sector.

2. Increase the fuel price to demonstrate general drop in transport activity

3. Introduce a new emission standard (Euro VI heavy duty vehicles) to demonstrate drop

in total emissions.

6.2 Methodology

Uncertainty and sensitivity analysis is applied in order to study the effect of selected

perturbations in the TREMOVE model. For each of the scenarios, two different sets of Monte

Carlo runs are computed: one corresponding to the control case and one to the perturbed case.

Each setting is related with a set of 512 simulations, based on quasi-random LPτ sequences.

The original set of 33 uncertain input factors was kept as the minimum input factor basis. A

Kolmogorov-Smirnov test is then applied on the results of the two different settings in order to

identify the variables where their output distribution has changed between the scenario and the

baseline. For those variables, sensitivity analysis investigates the changes in the output

distribution in terms of the changes in the input factors importance versus the new

parameterization changes.

6.3 Ownership tax increase

6.3.1 Description

The first scenario is a policy scenario which assumes that the ownership tax of passenger cars

is set to increase in the future. Such a policy could be followed for example as a means to

decrease passenger car use and promote other means of transport. Other reasons would be to

introduce a ‘green’ ownership tax, i.e. increase revenues that can be then used to fund

environment protection activities. By increasing the ownership tax of cars one would expect to

shift traffic to busses and power two wheelers, and also some traffic to aviation and railways.

In order to demonstrate the TREMOVE uncertainty behaviour to such a policy, it was decided to

increase the ownership tax of all passenger car categories. We assumed a rather aggressive

113

policy scenario, i.e. assuming that the ownership tax will linearly triple up to 2030. An example

of the values used for gasoline passenger cars in the category 1,4-2,0 l is shown in Figure 15.

It was considered to keep the uncertainty in the ownership tax of the scenario proportional to

the uncertainty in the baseline. Hence, the standard deviation used for the input uncertainty in

the scenario would be three times higher than the one used for the baseline uncertainty. The

mean and standard deviation values used in the Years 2011 and 2030 are summarized in Table

16.

Car 1.4-2.0l - Petrol

0

100

200

300

400

500

600

700

800

2000 2010 2020 2030

Year

Euro

/Year

ROWNTX - Scenario

ROWNTX - Baseline

Figure 15: Evolution of the ownership tax in the basecase of the TREMOVE model and the scenario

designed (Example gasoline cars 1.4 – 2.0 l).

Table 16: Ownership tax (€/year) in Scenario 1 for the boundary years 2011 and 2030. Values

have been linearly varied in the intermediate years.

Mean Stdev Vehicle Type

2011 2030 2011 2030

Car <1.4l - Petrol 126.0 378.0 0.0 0.0

Car 1.4-2.0l - Petrol 239.0 717.0 23.4 70.3

Car >2.0l - Petrol 769.0 2,307.0 66.2 198.7

Car <1.4l - Diesel 126.0 378.0 0.0 0.0

Car 1.4-2.0l - Diesel 239.0 717.0 23.4 70.3

Car >2.0l - Diesel 769.0 2,307.0 66.2 198.7

Car <1.4l - CNG 126.0 378.0 0.0 0.0

Car 1.4-2.0l - CNG 239.0 717.0 23.4 70.3

Car >2.0l - CNG 769.0 2,307.0 66.2 198.7

114

6.3.2 Results

A Kolmogorov-Smirnov test identified that different output distributions exist for fifteen of the

twenty-four variables. For all other output variables, the scenario and the baseline result to

almost identical output distributions. The variables for which statistically significant differences

in the output values exist are (in chronological order):

- TAXownership: all years after 2012

- Costs, Vehicles, VehKms: all years after 2015

- COSTrest: all years after 2022

- COSTpurchase, VATpurchase, COSTrepair, VATrepair: all years after 2023

- COSTinsurance, VATfuel: all years after 2027

- TAXinsurance, TAXfuel, FC: all years after 2028

- COSTfuel: all years after 2029

Figure 16 shows the uncertainty evolution for the abovementioned fifteen output variables in

the baseline (red) and the scenario (blue) run for UK. ‘TAXownership’ variable that, in terms of

its median, takes statistically significant higher values in the scenario run after 2012. This

would be expected given the large variation introduced in the ownership tax values.

In addition to the observed changes in ‘TAXownership’, increasing the ownership tax also shifts

overall costs higher and decreases the total number of vehicles, vehkms, fuel consumption and

the other cost related variables. Table 17 shows the difference in the medial values for the

output variables which are mostly affected. The differences are up to 3%.

115

Figure 16: Temporal evolution of the output uncertainty for UK, for the output variables that

demonstrate changes in their distribution between the scenario (blue) and the baseline (red) case.

The bold line represents the median while the dotted lines correspond to the 5th and 95th percentiles.

116

Table 17: Median values of the scenario 1 and the baseline in the years 2020 and 2030 and

percentage difference between the two.

Vehicle km [x10^6]

2020 2030

Vehicles [#]

2020 2030

Baseline 657,732 712,662 Baseline 37,434,661 40,543,655 Scenario 1 651,764 701,648 Scenario 1 37,091,918 39,884,879 % diff -0.9 -1.5 % diff -0.9 -1.6

NOx [Ton]

2020 2030

PM [Ton]

2020 2030

Baseline 204,279 179,722 Baseline 4,128 4,068 Scenario 1 203,329 178,311 Scenario 1 4,102 4,017 % diff -0.5 -0.8 % diff -0.6 -1.3

Costs [M€]

2020 2030

TAXownership [M€]

2020 2030

Baseline 375,041 434,464 Baseline 11,345 12,513 Scenario 1 380,952 446,045 Scenario 1 20,776 33,746 % diff 1.6 2.7 % diff 83.1 169.7

In order to further elaborate on the differences, Figure 17 shows examples of the scenario

effect in individual vehicle categories for the total activity and the vehicle population. Even at

this higher magnification level differences are only marginal. Total car activity in 2030 drops by

1% while power two wheeler activity (but also bus and light commercial vehicle activities) is

identical between the scenario and the baseline. Increase in the car ownership tax decreased

total road transport activity but does not seem to shift passenger transport to any other

modes.

400,000

450,000

500,000

550,000

600,000

650,000

700,000

2005 2010 2015 2020 2025 2030

VK

M [x1

0^

6]

Year

CAR

5,000

5,200

5,400

5,600

5,800

6,000

6,200

6,400

6,600

6,800

7,000

2005 2010 2015 2020 2025 2030

VK

M [x1

0^

6]

Year

2W

25,000,000

27,000,000

29,000,000

31,000,000

33,000,000

35,000,000

37,000,000

39,000,000

2005 2010 2015 2020 2025 2030

PO

P [#

]

Year

CAR

900,000

950,000

1,000,000

1,050,000

1,100,000

1,150,000

1,200,000

1,250,000

1,300,000

1,350,000

2005 2010 2015 2020 2025 2030

PO

P [#

]

Year

2W

Figure 17: Examples of ownership tax cost difference for selected output variables (vkm and population and selected vehicle categories). Continuous lines are baseline and dashed lines are

scenarios.

117

Sensitivity analysis applied on those variables is shown as pie charts (fraction of explained

variance calculated from standardized regression coefficients) in Figure 18. Although the

distribution of ‘TAXownership’ is statistically different between the baseline and the scenario,

rather small changes (<3%) occur in its input factors importance and their fraction of

explained variance. The sensitivity of the other fourteen variables remains unaltered in the

scenario compared to the baseline and for this reasons the pie charts are excluded.

Figure 18: Sensitivity analysis results: variation in input factor importance between the baseline

(left column) and the scenario (right column) run for UK for the output variable “Taxownership”.

Sensitivity for all other variables is identical between the scenario and the baseline.

6.3.3 Discussion

In Scenario 1, providing a much higher ownership tax estimate for passenger cars that reaches

three times higher than the baseline in 2030 only marginally affects all the output variables (of

course with the exception of the ownership cost). Median values between the baseline and the

scenario differ maximum by 1.5% in 2030 in the case of total road transport costs. All other

activity and emission data differ by less than 1%. The output variance is not significantly

affected compared to its baseline range. In view of the small changes, the sensitivity analysis

results to identical effect of each input variable uncertainty to the output variables. This

scenario shows that changes in the ownership tax have practically very limited effect on the

median and the uncertainty ranges of the TREMOVE activity and emission output.

Activity shifts within the road sector, i.e. shift of the passenger transport to either busses or

power two wheelers could not be observed. Overall, increasing the cost of passenger car

ownership leads only to a marginal (1%) decrease of activity in the road transport sector which

comes solely from an identical decrease in the passenger car activity.

118

6.4 Effect of fuel cost

6.4.1 Description

One major cost component in TREMOVE is the price of the different fuel categories. This cost is

directly connected to the fuel consumption and the total vehicle kilometres driven by each

vehicle. Detailed historical data is available for the fuel price. This means that no uncertainty

exists for past years, and the fuel price has no impact on the uncertainty of the model.

However the estimation of the future price can be difficult; taking also into consideration that

energy consumption affects the economy of the entire world. An example of this uncertainty is

the rise of the crude oil price in 2008 over 146$/barrel [13]. Another point is that even within a

small country like Greece, the price can vary between short periods of time. The pre-tax price

of the unleaded gasoline in Greece varied between 0.3 to 0.5 Euros per litre in a period of less

than 8 months (Oct 2009 to May 2010) [14]. It is evident that high uncertainty exists for the

estimation of the fuel price. The uncertainty does not come from some technical variance in the

input data; this is a very volatile value with its uncertainty depending on macroeconomic

factors and world events. This increased uncertainty could not be included in the baseline runs

due the fact that such large variance in the fuel price when combined with the rest of the input

data variability could have unpredictable effect in the final results. For this reason the second

scenario was designed to study the effect of the highly uncertain fuel price. Of course, it should

be repeated that the TREMOVE structure is not appropriate to quantify large differences in the

cost of fuel, such as fundamental differences in the projection of the price of the oil barrel. This

was first observed in the scientific review of TREMOVE by Annema et al. [19]. In their report,

they state: “The assumptions in TREMOVE restrict the range of policy assessments for which

the model is suitable to only those that are expected to have a limited impact on incomes or

production. TREMOVE assumes that incomes and production are unaffected by policy changes

and that income elasticity of any transport demand is equal to 1. Accordingly, TREMOVE is not

designed for the assessment of policies that are expected to have a significant impact on

incomes or production.” Obviously, fuel prices are one of the policies, or better, developments

that largely affect both income and production. Hence, TREMOVE is not well suited to reflect

the impacts of different fuel prices on demand and activity. However, it is still interesting to

observe how TREMOVE reacts to differences in the price of petroleum products used in road

transport, as behaviour of the model, even if this is not representative of realworld changes.

It was decided to use the variability of the unleaded Gasoline in Greece as a starting point and

calculate the fuel price based on that uncertainty. A price range from 0.3 to 0.5 Euro/litre gives

an uncertainty range of ±30%. This uncertainty was introduced in the model from year 2010

and onwards. The sensitivity analysis was performed for all output data. This means that the

median value of the scenario is identical to the baseline but the scenario introduced an

uncertainty range not existing in the baseline.

119

UK prices

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

1995 2005 2015 2025Year

Euro

/Litre

Gasoline

+3s Gas

-3s Gas

UK prices

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1995 2005 2015 2025Year

Euro

/Litre

Diesel

+3s Dies

-3s Dies

UK prices

0.0

0.1

0.1

0.2

0.2

0.3

1995 2005 2015 2025Year

Euro

/Litre

LPG

+3s LPG

-3s LPG

UK prices

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

1995 2005 2015 2025Year

Euro

/Litre

CNG

+3s CNG

-3s CNG

Figure 19: Scenario data introduced in the model.

6.4.2 Results

The second scenario deals with a different set-up of the FCOST parameter. FCOST in the

perturbed run imposes a 30% variability over the control run, such that the range in the

perturbed run is [0.7*FCOST(control), 1.3*FCOST(control)].

A Kolmogorov-Smirnov test identified that different distributions exist only for COSTfuel,

between the perturbed and the control case, for all years after 2010.

Figure 20 shows the temporal evolution of the output uncertainty for COSTfuel. The different

distributions produced by the scenario setting have the same median with the control case but

with both tails of the distribution expanded (i.e. small and large percentiles become more

extreme). This is particularly evident, as expected, for COSTfuel (and VATfuel and TAXfuel)

whose coefficient of variation (in 2030) jumps from 12% to 20%.

120


demonstrate changes in their distribution between the scenario (blue) and the baseline (red) case.


Sensitivity analysis applied on COSTfuel as pie charts (fraction of explained variance calculated

from standardized regression coefficients) is shown in Figure 21. The sensitivity does not

change for any other of the output variables. Specifically, FCOST becomes the principal

important factor for COSTfuel explaining more than 64% of its variance while the second

variable, eEFfc (the previously ranked one in importance), explains less than half of the

variance fraction attributed to FCOST (29-30%).

121

Figure 21: Sensitivity analysis results for the COSTfuel variable: variation in input factor

importance between the baseline (left column) and the scenario (right column) run for UK.

In this case, it is interesting to observe the change in the confidence interval of the output

variables as medians have not changed between the scenario and the baseline. This is shown

in Table 18 for selected output variables. The output is practically only wider for the COSTfuel

variable and it is only little varied for the other output variables (total cost being the second

important one). The tax fuel is not affected as its value is independent of the fuel cost.

Interestingly, the fuel VAT confidence interval is also much less affected that the fuel cost. This

is because the VAT is applied on the sum (fuel cost + fuel tax) and the constant fuel tax value

attenuates large variations in its output confidence interval. If one looks on effects on a per

vehicle category level (Annex 2), one can see that confidence intervals of VAT fuel for light

duty vehicles and two-wheelers increase somehow more as a combined effect of direct fuel

increase cost and cross-effects of the high variations of costs in the passenger car sector.

However, the induced increase in the confidence intervals for these vehicle categories is too

small to be shown in the final result, which is dominated by passenger cars.

122

Table 18: 95% confidence intervals for selected output variables between scenario 2 and the

baseline in the years 2020 and 2030.

Vehicle km [x10^6]

2020 2030

Vehicles [#]

2020 2030

Baseline 61,184 68,251 Baseline 3,581,258 3,989,701

Scenario 2 62,117 69,647 Scenario 2 3,642,697 4,077,402

NOx [Ton]

2020 2030

PM [Ton]

2020 2030

Baseline 120,352 93,729 Baseline 4,265 4,175

Scenario 2 118,574 94,837 Scenario 2 4,260 4,176

Costs [M€]

2020 2030

COSTfuel [M€]

2020 2030



TAXfuel [M€]

2020 2030

VATfuel [M€]

2020 2030



It is interesting to examine in this case how the change in the uncertainty of the fuel cost in

road transport affects the activity in other modes and in particular in aviation and railways. It

would be expected that increasing the road fuel price would shift some activity to non road

modes and vice versa. Therefore, by modifying the range of costs of road transport, one should

expect to observe an inversely proportional shift of non road transport demand. The change in

total demand for the aviation and railway sectors as an effect of the modified road transport

fuel prices is shown in Figure 21. Despite expectations, the fuel cost does not seem to affect

the median or the uncertainty of the activity in passenger transport of other modes, as the

median, the 5th percentile, and the 95th percentile are identical between the basecase and the

scenario.. This is an indication of very small elasticity between different passenger transport

options. It mostly shows that changing the road fuel price will get more people on or off the

road, depending of whether the fuel price has been reduced or increased but will not shift more

people to other modes. This result should be validated with real-world examples.

AIR

100 000

105 000

110 000

115 000

120 000

125 000

130 000

135 000

140 000

2005 2010 2015 2020 2025 2030Year

Million p

km

MED_BC

PERC _05

PERC _95

MED_S2

PERC _05

PERC _95

TRAIN

60 000

65 000

70 000

75 000

80 000

85 000

90 000

2005 2010 2015 2020 2025 2030Year

Million p

km

Figure 22: Effect of the road fuel prices uncertainty on the uncertainty of pkm conducted by

aviation and railways. The MED_PC and MED_S2 lines correspond to the medial values of the

baseline and the scenario, while the corresponding 05 and 95 lines correspond to the 5th and 95th

percentiles.

123

6.4.3 Discussion

In this scenario, the base price of fuels was varied within ±30% of its mean value in the

scenario while no uncertainty was considered in the baseline. This variation only affected the

relevant cost output variable. Confidence intervals for the output significantly increase only for

the fuel components and for total cost. The confidence intervals for all other variables were

only marginally affected. Due to the small relative effect, the sensitivity analysis produces

identical results between the baseline and the scenario, i.e. the output depends on the

uncertainty of the input variables in the same fashion as in the baseline.

Interestingly, introducing such a large uncertainty for the road fuel cost, does not seem to

drive more activity to other modes of transport. The variance in total pkm in air and rail

transport remains practically identical between the baseline and the scenarios. The total

variance of pkm increased by no more than 1.5% in both aviation and rail transport. This

dictates limited intermodal shifts.

6.5 HDV Euro VI

6.5.1 Description

As a third scenario for demonstrating TREMOVE’s application we chose one of the cases where

one of the TREMOVE versions (originally v2.43b) has also been used in the past: the heavy-

duty Euro VI emission standard introduction. Our aim is to demonstrate the expected

uncertainty in TREMOVE calculations when a new emission standard is simulated. It should be

made clear that the TREMOVE response depends on the vehicle type on which the new

emission standard is introduced. For example, the uncertainty would probably be different if a

new emission standard was introduced for passenger cars, just because of the different

modelling approach of passenger cars compared to heavy duty vehicles. However, we decided

to demonstrate the uncertainty in a scenario were TREMOVE has already been applied on and

for which realistic information of input data uncertainty can be found.

Euro VI vehicles will be introduced in 2013 in the fleet. The emission limits values for Euro VI

heavy duty trucks have been set by Regulation 595/2009. Table 19 presents the limit values

for Euro VI and Euro V as well as the percentage reduction of Euro VI over Euro V. The

emission factors of Euro VI in TREMOVE were simulated by introducing the relative emission

limit reduction of Table 19 directly on the Euro V emission factor. One may question whether

there is no uncertainty in this reduction, i.e. will the Euro VI emission factors perfectly comply

with the emission reductions over Euro V imposed by the regulations? The answer is probably

no because history has shown that real-world emissions usually exceed the emission limit

requirements. This is for various reasons, including the engine calibration and the

aftertreatment efficiency. However, as in several instances in this report, our intention is not to

make a prediction of what the real-world will bring but what is the inherent uncertainty in the

modelling approach of TREMOVE. The real-world vs. emission limit disparity is actually a

compromise of the emission regulation and not an effect that has to be simulated with

TREMOVE. When TREMOVE was used to simulate the Euro VI standards, fixed values for the

emission reductions were used per scenario. One would therefore need to calculate what is the

124

uncertainty associated with these fixed values and not to introduce yet another variance to

these values. To calculate the uncertainty in the emission factor, the cov of Euro V and Euro VI

emission factors has been assumed the same.

Table 19: Euro V and Euro VI emission limit values for heavy duty vehicles (g/kWh).

CO HC NOx PM

Euro V 1.5 0.46 2 0.02

Euro VI 1.5 0.13 0.4 0.01

% Reduction 0 72 80 50

The uncertainty associated with the introduction of a new emission standard has mainly to do

with the additional costs that this new technology is associated with. Costs are difficult to

estimate, let alone to predict beforehand. Therefore, an uncertainty range for the assumed

costs has to be introduced. Input data for the uncertainty in the cost calculation were obtained

by the report of Gense et al. [11]. The Euro VI emission standards finally decided, correspond

to Scenario 5 of that report (with a small variation concerning THC). Therefore, the cost values

linked to Scenario 5 will be used as an input. The report of Gense et al. [11] came up with two

sets of cost figures: One assuming that all cost of new technology was attributable to the need

to meet the Euro VI limits (100% allocation) and the second set assuming that 50% of the cost

of some new components would reflect new technology that would have been introduced in any

case to improve vehicle performance. With these assumptions, only 50% of the total cost of

some components should be attributed to Euro VI. These values are shown in Table 20.

Table 20: Costs (€/veh.) allocated to Euro VI technology in the report of Gense et al. [11].

100% allocation 50% allocation Engine Size [l]

Min Max Min Max

6 3,355 3,553 2,855 3,053

9 4,318 4,615 3,718 4,015

13 5,351 5,780 4,651 5,080

In order to allocate the cost figures from that report onto the four weight classes of heavy duty

trucks and one class of diesel busses in TREMOVE, a number of assumptions will have to be

made, which are largely based on the analysis of Zierock et al. (2007):

1. It seems reasonable to assume that since the exact percentage of cost allocation to

Euro VI cannot be predicted, a uniform distribution between the 50% and the 100%

allocation is the best approach to simulate uncertainty.

2. The report of Gense et al. [11] came up with cost figures per 3 engine capacity class

and not the 4 gross vehicle weight class (plus one for busses), as TREMOVE uses.

Zierock et al. [12] used market information to link the two size figures. However, there

is an uncertainty in the estimation because different GVW classes may use engines of

different capacity. In order to simulate this uncertainty, we have used the following

assumptions: Vehicles <7.5 t use only engines of 6l, Vehicles of >32 t use engines of

13 l, Vehicles of 7.5-16 t, use a mix of 6 l and 9 l engines, and vehicles 16t-32 t use a

mix of 9 l and 13 l engines. Busses use 9l-13 l engines. A 50-50 allocation has been

assumed for these intermediate classes Therefore the min-max range for the

intermediate GVW and bus classes are wider than the individual engine classes. Based

on these assumptions, Table 21 shows the final cost allocation per TREMOVE vehicle

category.

125

Table 21: Costs (€/veh.) allocated to Euro VI vehicle classes in TREMOVE.

Vehicle class Min Max

<7,5 t 2,855 3,553

7,5 – 16 t 3,287 4,084

16 – 32 t 4,185 5,198

>32 t 4,651 5,780

Diesel busses 4,185 5,198

The last component of cost calculation is attributed to the cost of urea. The report of Zierock et

al. [12] provides a range of values for the consumption rate (3-6% of diesel fuel) and cost

(0.27-0.67 Euro/lt) of urea. This can be converted to a range of Euro/km values (to be added

in the maintenance cost in TREMOVE) by considering a mean fuel consumption rate per vehicle

class. We have assumed that the mean fuel consumption of Euro V trucks also holds true for

Euro VI as well. This is not entirely true as we will later demonstrate but it is a very good

approximation for the needs of the uncertainty calculation we are attempting here. Finally,

Zierock et al. [12] assumed that only 30% of the urea cost should be allocated to Euro VI,

assuming that 70% of the Euro V vehicles already consume urea. In fact, we have updated

market figures which show that 74% to 76% of the Euro V market are SCR equipped vehicles.

For consistency with the earlier report we have therefore assumed that only 30% of the total

urea costs should be allocated to Euro VI. Based on this approach, the range of costs for urea

are presented in Table 22.

Finally, we have assumed that the fuel consumption of Euro VI vehicles will be higher than

Euro 5 uniformly by 0.5-2%. This is due to the much lower NOx limits which lead to lower

efficiency compared to Euro V.

Table 22: Calculation of urea cost per Euro VI kilometre (only 30% of urea cost allocated to Euro VI

as incremental difference over Euro V).

Vehicle class Fuel

Consumption (g/km)

Rate of urea cons. vs fuel consumption

(l/g fuel)

Cost of Urea (€/lt)

Min Cost of Urea (€/km)

Max Cost of Urea (€/km

<7,5 t 104 0.0003 0.0016

7,5 – 16 t 206 0.0006 0.0031

16 – 32 t 329 0.0010 0.0050

>32 t 387 0.0012 0.0058

Diesel busses 260

0.0000375- 0.000075

0.27-0.67

0.0008 0.0039

6.5.2 Results

The third scenario deals with a different set-up of three variables controlling the “extra car

cost”, the “extra fuel cost” and the “consumption increase”. A uniform distribution is attributed

to all of them.

A Kolmogorov-Smirnov test identified that different distributions exist for PM and NOx after

2016, between the perturbed and the control case. Interestingly, no cost related variables were

affected, despite the change in the cost items describing the Euro VI technology.

126

Figure 23 shows the uncertainty evolution for the abovementioned output variables in the

baseline (red) and the scenario (blue) cases for UK. Both variables demonstrate a decrease in

the future. NOx demonstrates a shift of the whole distribution towards lower values while PM is

characterized by a shrink that keeps the same lower tail. The change occurs after 2016. This is

expected as Euro VI introduced decreased emission limits compared to the baseline. Euro VI

was assumed to have been introduced in 2013. After about 2016, the NOx median of the

scenario becomes lower than the 5th percentile of the baseline.

The individual differences in the median values of the baseline and the scenario for selected

output variables are shown in Table 22. The table confirms that the only differences are

observed for NOx and PM and for no other output variable.

Sensitivity analysis applied on those variables is shown as pie charts (fraction of explained

variance calculated from standardized regression coefficients) in Figure 24. It is apparent that

actually there exist no changes in the input factors importance. In the perturbed setting, eEF

continue to drive the PM and NOx uncertainty, with a small reduction (~1-4%) in the fraction

of the explained variance. This is because the cov in the Euro V and Euro VI remained the

same, which means that the induced uncertainty is less in the Euro VI case, as a result of a

drop in the mean emission factor. The drop in the uncertainty is realistic as the aftertreatment

used at a Euro VI level (SCR and DPF) is expected to effectively control emissions within their

limits.


demonstrate changes in their distribution between the perturbed (blue) and the control (red) case.


127

Table 23: Median values of the scenario 3 and the baseline in the years 2020 and 2030 and

percentage difference between the two for selected output variables.

Vehicle km [x10^6]

2020 2030

Vehicles [#]

2020 2030

Baseline 657,732 712,662 Baseline 37,434,661 40,543,655

Scenario 3 657,673 712,541 Scenario 3 37,433,657 40,541,438

% diff 0.0 0.0 % diff 0.0 0.0

NOx [Ton]

2020 2030

PM [Ton]

2020 2030



% diff -29.1 -41.9 % diff -12.9 -17.3

Fuel cons. [Ton]

2020 2030

VOC [Ton]

2020 2030



% diff 0.1 0.0 % diff -0.6 -0.4

Costs [milEuro]

2020 2030

COSTpurchase [milEuro]

2020 2030



% diff 0.0 0.1 % diff 0.1 0.2

COSTrepair [milEuro]

2020 2030

Baseline 77,359 90,137

Scenario 3 77,449 90,317

% diff 0.1 0.2

6.5.3 Discussion

A Euro VI scenario was introduced in this case for HD vehicles, including reductions in emission

limits and increased purchase and operation costs. The effect of the introduction was only

shown for PM and NOx and for no other output variable. The contribution of input variables to

the uncertainty of the scenario closely matches to the baseline as the relative variance of the

Euro VI emission factors has been assumed the same as Euro V.

128

(a)

(b)

Figure 24: Sensitivity analysis results: variation in input factor importance between the control (left

column) and the perturbed (right column) run for UK, for the output variables (a) PM and (b) NOx

identified from the Kolmogorov-Smirnov test.

129

7 Conclusions and recommendations

Objective of this study was the uncertainty and sensitivity analysis of TREMOVE both with

regard to the baseline and three indicative scenarios. This was made possible by first

identifying a variance range for the input variables of the model. Out of the several variables

33 were the ones for which an uncertainty range was identified. These variables are contained

in the stock and emission and fuel consumption modules. For reasons that are explained

thoroughly in this final report of the study variables in the demand, welfare, well to tank and

non road modules were not varied. For the variables varied uncertainty ranges were

characterized on the basis of literature data or experimental data or in few cases based on

empirical assumptions. The output of TREMOVE was observed with a selection of 24 variables

including cost, activity data, stock, fuel consumption and pollutant variables.

Important variables

A screening test was first conducted which identified the influential input variables. 14 variables

were deemed important. These were:

- the average trip length (ltrip)

- the hot and cold emission factors (eEF, eEFratio, eEFfc, eEFfcratio)

- the (B,T) - parameter: characteristic service life & faillure steepness (paraB, paraT

pair)

- the road vehicle basic purchase resource cost - EURO 2000 (eRPCSBASE)

- the estimated residual value function as a percentage of purchase cost

(usresidualparaAB)

- the repair and maintenance cost excluding taxes as % of purchase resource cost (ex

tax) (eRREPMAINTCFRACTION)

- the insurance cost as percentage of vehicle purchase resource cost (RINSCFRACTION)

- the labour cost - net wage - for truck drivers - EURO per hour (RLABOURC)

- the labour tax - bruto wage minus netto wage - for truck drivers - EURO per hour

(RLABOURTX)

- the annual ownership tax road vehicles - EURO 2005 (ROWNTX)

- the public transport fare cost coverage (PUBLICCOSTCOV)

Further to the identification of the important variables the screening test also leads to the

following conclusions:

- the model output referring to activity and cost is mostly a linear combination of the

input variables

- differently expressed, the previous point means that each output variable is mostly

determined by its corresponding input variable. For example the total purchase cost is

determined by the uncertainty in the purchase cost of single vehicles

- the linear behaviour is mostly an effect of the structure and the values in the elasticity

trees of the model

130

Total uncertainty baseline

The uncertainty of the baseline expresses the variance which is induced to the basecase by the

model variables. The total baseline uncertainty is determined by exogenous factors to

TREMOVE such as macroeconomic and demographic data which can be assumed when

producing this basecase by higher level models (eg TRANSTOOLS, PRIMES, Poles).

Characterizing this overall uncertainty is beyond the objective of this study. The uncertainty of

the baseline for each of the years 2010, 2020, 2030 is given as the coefficient of variance

(cov=standard deviation over mean) for each output variable in the following table. The output

variables in the table are presented in a descending order according to the cov for the year

2030. Based on the values in the table the following conclusions can be drawn.

Output Variable

Units Median 2010 Median 2020 Median 2030 cov 2010

cov 2020

cov 2030

CO Ton 252,000 119,402 119,439 69% 51% 51%

PM Ton 10,750 3,400 3,421 25% 27% 27%

VOC Ton 46,927 30,279 31,559 37% 26% 24%

TAXrest M€ -8,454 -8,768 -9,365 21% 22% 22%

NOx Ton 331,319 170,500 158,458 19% 17% 17%

COSTinsurance M€ 24,922 38,291 44,671 6% 13% 14%

TAXinsurance M€ 1,260 1,934 2,255 6% 13% 14%

VATfuel M€ 8,041 9,418 11,412 13% 13% 13%

TAXfuel M€ 33,301 41,490 48,064 12% 12% 12%

COSTfuel M€ 30,727 32,106 40,202 11% 11% 12%

TAXownership M€ 6,166 10,810 12,070 8% 11% 12%

FC Ton 46,618,917 55,591,702 60,043,602 11% 11% 11%

COSTrepair M€ 59,798 73,706 86,483 3% 10% 10%

VATrepair M€ 7,293 9,090 10,691 3% 10% 10%

COSTlabour M€ 10,869 14,896 16,607 9% 9% 9%

COSTlabourtax M€ 11,634 15,944 17,773 9% 9% 9%

VATpurchase M€ 10,841 11,566 13,107 5% 9% 9%

COSTpurchase M€ 84,178 99,323 114,467 4% 8% 9%


VATrest M€ 1,252 1,293 1,377 5% 5% 5%

Costs M€ 323,333 396,036 458,228 2% 4% 4%

Vehicles # 33,652,081 37,918,723 40,997,888 2% 3% 3%

Vehkms ×106 585,653 665,914 720,553 2% 3% 3%

COSTrest M€ 41,193 44,395 47,695 2% 2% 2%

- The uncertainty is large for the emission of pollutants, mostly due to the uncertainty in

the emission factors. This is in the order of 20-30% but can reach up to 50% in the

case of CO.

- Fuel dependent variables are second with regard to output uncertainty with cov values

in the order of 10-15%.

- Total cost figures exhibit uncertainty ranges in the order of 4-10%, i.e. they are rather

small depended on the variance of input data.

- Finally, population and activity data exhibit very little uncertainty, in the order of 2-

3%. The uncertainty per vehicle category (see Annex) is of the same magnitude. This

131

shows that the activity and population is little elastic to changes in costs and other

variables in the model. This is potentially very important. This means that changes in

the activity and population data are very limited despite the much wider uncertainty

ranges assumed for the input variables.

- The levels of all output variables corresponding to the demand or stock group increase

through time (with the exception of TAXrest). For the emissions group the opposite is

true except for the fuel consumption. This is expected assuming that road transport

activity will continue to increase in the future while emission standards will continue to

effectively control emissions. The increase in fuel consumption remains to be seen – it

is reminded that the TREMOVE version examined does not include hybrid vehicles, or

electric vehicles.

- The variables corresponding to the demand or stock group that demonstrate an

increase in their uncertainty in the future are: COSTpurchase, VATpurchase,

TAXownership, COSTinsurance, TAXinsurance, COSTrepair, VATrepair, Costs, Vehicles

and Vehkms. The highest increase has been seen for COSTrepair and VATrepair. The

variables corresponding to the emissions group that demonstrate a decrease in their

uncertainty in the future are VOC and CO. The largest decrease has been seen for CO,

however its uncertainty is always the highest among the 24 considered output

variables. All the other variables that are not mentioned exhibit a very small variability

in their temporal uncertainty. The large but decreasing uncertainty of CO and VOC is

caused by their large dependence on the emission factors of gasoline cars. As gasoline

cars are both expected to become less important in the future (compared to diesel

ones) but as also Euro 5 and 6 emission standards are expected to lead to even more

strict control over the lifetime of the vehicles, the observed trend is explainable.

Sensitivity analysis

The sensitivity analysis quantifies how much the uncertainty of each input variable contributes

to the uncertainty if the output variables. The sensitivity analysis conducted identified both first

order dependencies and higher order ones. The following table summarizes the first order

interactions between input and output variables in a descending order of interdependency for

the year 2030. The table shows that first order interactions explain from 99% (COSTrepair) to,

worst case, 88% (vehkms) of the output uncertainty.

- Ten out of the fourteen uncertain inputs are of principal importance for one or more

output variables.

- The hot emission factors (eEF) influences most the variability of the emissions (VOC,

NOX, PM, CO) while the basic road vehicle purchase resource cost (eRCPSBASE)

controls the variability of the stock and activity variables (vehicles and vehicle-kms).

On the other hand, many input factors are responsible for the variability of the cost

related output.

- All model outputs exhibit high linearity. The least amount of explained-by-single

contributions variance estimated was 88% and corresponds to the output variables

Costs, Vehicles and Vehkms. The remaining fraction depends on higher order

interdependencies between the input variables.

132

- The linearity of the output variables is generally constant in time. On the other hand,

the total effects are either constant or decreasing in the future. This implies that the

control of the singular perturbations will be always effective.

- The highest amount of interaction effects has been seen for TAXregistration but it is

attributed to all input factors. The highest decrease in the effect of the second and

higher order terms has been observed for COSTpurchase and VATpurchase. This gives

the opportunity to work on the uncertainty of the RPCSBASE only in the future in order

to reduce the variability of COSTpurchase and VATpurchase.

Output Variable ΣSI

2010 ΣSI2020 ΣSI

2030

COSTrepair 0.97 0.98 0.99

VATrepair 0.98 0.99 0.99

FC 0.96 0.97 0.98

COSTfuel 0.96 0.97 0.97

TAXfuel 0.96 0.97 0.97

VATfuel 0.96 0.97 0.97

COSTlabour 0.96 0.96 0.96

COSTlabourtax 0.96 0.96 0.96

TAXrest 0.96 0.96 0.96

VATrest 0.96 0.96 0.96

PM 0.97 0.96 0.96

COSTpurchase 0.97 0.95 0.95

TAXownership 0.96 0.95 0.95

COSTinsurance 0.94 0.94 0.95

TAXinsurance 0.94 0.94 0.95

COSTrest 0.97 0.96 0.95

NOx 0.96 0.95 0.95

VATpurchase 0.97 0.94 0.94

TAXregistration 0.89 0.92 0.92

CO 0.91 0.92 0.91

VOC 0.93 0.91 0.91

Costs 0.88 0.88 0.88

Vehicles 0.89 0.88 0.88

Vehkms 0.89 0.88 0.88

Scenarios

Three scenarios were executed to quantify the uncertainty of TREMOVE when used for policy

impact assessment. The scenarios were selected in a way as to examine different instances of

the model operation. Specifically the three scenarios were simulating:

1. Scenario 1: increase in the ownership tax of passenger cars, ie a policy aiming at

transferring activity from the passenger car sector to other transport means.

133

2. Scenario 2: introduction of uncertainty range for the cost of road fuels. Despite that

TREMOVE is generally considered not appropriate to simulate large effects of fuel

prices, such a scenario may stimulate some shift of activity to non road modes.

3. Scenario 3: introduction of a new emission standard. EURO VI heavy duty was

selected as this has been real scenario simulated by a previous version of TREMOVE.

In scenario 1 the ownership cost for passenger cars was increased linearly from 2011 to 2030

to reach in the scenario three times as high as the baseline in 2030. This variation was

statistically observed to affect fifteen out of the 24 output variables, in principle the ones

related to the cost. Despite this tremendous assumed increase in ownership costs, the number

of vehicles and vehicle-km is only marginally (~1.5%) lower than the baseline, while total fuel

consumption is affected by 2.2%. Increase in the ownership cost of passenger cars led to an

almost proportional drop in their activity but did not lead to substantial modal shifts to other

road vehicles or non-road modes. Also, due to the small difference the impact of input variable

uncertainty to output variance (sensitivity) was identical between the scenario and the

baseline.

In scenario 2, a ±30% variance over the mean value was introduced for all road fuel

components. This is expected not to affect the median of the output variables but only their

confidence intervals. Despite the relatively large variance induced the only output variables

significantly affected are cost figures and in particular fuel cost figures. The confidence interval

of stock, activity and emission output was not affected by more than 2%. The activity of non

road modes was not at all influenced.

Finally, in scenario 3, emission factors (and their associated uncertainty) for heavy duty trucks

were reduced according to the EURO VI over EURO V limit values while additional costs were

introduced to account for the cost of deNOx aftertreatment and the operation cost increase due

to NOx reduction agent. Fuel consumption was also marginally increased. The scenario leads to

significantly different values only for PM and NOx, as expected. All other activity, stock and

cost figures remained practically unchanged.

Recommendations

The analysis conducted in this study made possible to derive some recommendations on how

TREMOVE behaves and how its estimates can be potentially improved:

1. The total uncertainty of the projection, taking into account macroeconomic and

demographic factors may be realistically assessed only by introducing alternative

basecase projections in the model. This can be a useful future activity.

2. A limited number of input variables (14) seems to drive the total model uncertainty. In

addition, several output variables can be approximated as linear combinations of input

variables with a small loss in precision. These effects are induced by the rather limited

elasticity in shifts between different modes of transport and vehicle types offered by

the demand module. If this limited flexibility is validated (see point 6 in this list), then

it can be suggested that several model operations can be simplified with a beneficial

effects on model transparency and processing time.

134

3. Our analysis only took into account the variables and parameters inclusive in the

model. Expanding the model to cover additional vehicle types, such as alternative fuel

vehicles, hybrids, plug-in hybrids, and electric vehicles may increase the uncertainty of

the estimates but is deemed necessary to cover future applications of the model.

4. The fact that a limited number of variables is important for most of the model output

uncertainty means that better quality / more precision in the estimates of these

specific variables will reduce the uncertainty of the output. Of particular importance

appear to be the emission factors, the purchase cost of vehicles, the parameters

defining the scrappage probability, the parameters used to estimate the residual cost

when a vehicle is scrapped and cost-related parameters (maintenance, insurance,

ownership, labour).

5. This study was limited to one country only (UK). Given the linear behaviour of the

model and the limited sensitivity of the demand to the input variables uncertainty,

extending the analysis to other countries does not seem to offer new insights. This

might affect the numerical values of the uncertainty indicators produced but would not

change the conclusions of the study. To improve the model output priority should

rather be given in improving the quality of the major input variables identified in this

study and in exploring how much the model elasticities reflect reality.

6. Validation of key model elasticities would be an interesting exercise, in the light of the

limited elasticity identified. The currently (2010-2011) changing environment in Europe

due to the financial and credit crisis offers several opportunities for validation. The

model could be applied to simulate effects of increasing fuel taxation, ownership

taxation, scrappage activities, etc., that take place today in several countries, and

compare with real-world trends.

7. A follow up activity could be to derive the linear functions between output data and

input variables and compare how much they deviate from TREMOVE output. This could

serve three purposes: (i) Quantify how much TREMOVE output deviates from linear

behaviour, (ii) Have a simplified TREMOVE model to easily perform scenarios for which

maximum accuracy is not necessary, (iii) Identify areas where TREMOVE structure

could be simplified without loss of precision.

135

References

[1] http://ec.europa.eu/environment/air/pollutants/models/tremove.htm

[2] Giannouli M., Samaras Z., Keller M., deHaan P., Kallivoda M., Sorenson S., Georgakaki A.,

2006. Development of a database system for the calculation of indicators of environmental

pressure caused by transport. Science of The Total Environment 357, 247-270.

[3] Ntziachristos L., Mellios G., Kouridis C., Papageorgiou T., Theodosopoulou M., Samaras Z.,

Zierock K.-H., Kouvaritakis N., Panos E., Karkatsoulis P., Schilling S., Merétei T., Aladár Bodor

P., Damjanovic S., Petit A., 2008. European Database of Vehicle Stock for the Calculation and

Forecast of Pollutant and Greenhouse Gases Emissions with TREMOVE and COPERT. Laboratory

of Applied Thermodynamics, 08.RE.0009.V2, Thessaloniki, Greece, p.260.

[4] Cukier R., Levine H., Shuler K., 1978. Nonlinear sensitivity analysis of multiparameter

model systems. Journal of Computational Physics 26, 1-42.

[5] Kioutsioukis I., Tarantola S., Saltelli A., Gatelli D., 2004. Uncertainty and global sensitivity

analysis of road transport emission estimates. Atmospheric Environment 38, 6609-6620.

[6] Ntziachristos L., Mellios G., Kouridis C., Papageorgiou T., Theodosopoulou M., Samaras

Z., Zierock K.-H., Kouvaritakis N., Panos E., Karkatsoulis P., Schilling S., Merétei T., Aladár

Bodor P., Damjanovic S., Petit A., 2008. European Database of Vehicle Stock for the

Calculation and Forecast of Pollutant and Greenhouse Gases Emissions with TREMOVE and

COPERT. Laboratory of Applied Thermodynamics, 08.RE.0009.V2, Thessaloniki, Greece, p.260.

[7] Saltelli A., Chan K., Scott M., Eds. (2000). Sensitivity analysis. New York, John Wiley &

Sons Ltd.

[8] Saltelli A., Tarantola S., Chan K.P.S., 1999. A quantitative model-independent method for

global sensitivity analysis of model output. Technometrics 41, 39-56.

[9] Sobol I.M., Turchaninov V.I., Levitan Y.L., and Shukhman B.V. (1992), Quasirandom

sequence generators. Keldysh Institute of Applied Mathematics, Russian Academic of Sciences,

Moscow

[10] Sobol I.M. (1993) Sensitivity estimates for Non-linear Mathematical Models, Mathematical

Modelling and Computational Experiment (1) 4 407-414

[11] Gense, N.L.J., Riemersma, Such, C., Ntziachristos, L. 2006. Euro VI technologies and

costs for heavy duty vehicles. The expert panels summary of stakeholders responses. TNO

Science and Industry report 06.OR.PT.023.2./NG, the Netherlands, p.56.

136

[12] Zierock, K.-H., Ntziachristos, L., Kouridis, Ch., Gkatzoflias, D., Samaras, Z. 2007.

TREMOVE Model scenario runs related to the impact assessment of EURO VI emission limit

values for Heavy Duty Vehicles. LAT Final Report, Thessaloniki, Greece, p.52.

[13] http://www.oilprices.com/history

[14] http://www.ypeka.gr/LinkClick.aspx?fileticket=luGs6ntUqHs%3d&tabid=478

[15] Kouridis, C., Gkatzoflias, D., Kioutsioukis, I., Ntziachristos, L., 2009. Uncertainty

estimates and guidance for road transport emission calculations. EMISIA Final Report,

Thessaloniki, Greece.

[16] Caserini, S., Pastorello, C., Tugnoli, S. (2007). Relationship between car mileage and

length of service: influence on atmospheric emission assessment. TFEIP Expert Panel on

Transport. Milan, Italy.

[17] Treanton, K. (2004). Special Issue Paper 8 Net calorific values. IEA. Paris, France.

[18] Hill, N. (2005). EU Fuel Quality Monitoring –2004 Summary Report. AEA Technology

Environment, UK.

[19]. Annema, J.A., Hoen, A., Turton, H., Schrattenholzer, L., Mudgal, S., Fergusson, M.,

2006. Scientific review of TREMOVE – a European transport policy assessment model, MNP

Report.

137

ANNEX I: uncertainty estimates of the baseline per vehicle category

138

COSTfuel

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

2005 2010 2015 2020 2025 2030

CO

STfu

el (

Mil E

uro

)

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

CO

STfu

el (

Mil E

uro

)

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

CO

STfu

el (

Mil E

uro

)

Year

LDV

0

20

40

60

80

100

120

140

2005 2010 2015 2020 2025 2030

CO

STfu

el (

Mil E

uro

)

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

CO

STfu

el (

Mil E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

139

COSTinsurance

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030CO

STin

sura

nce

(M

il E

uro

)

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030CO

STin

sura

nce

(M

il E

uro

)

Year

HDV

0

500

1,000

1,500

2,000

2,500

2005 2010 2015 2020 2025 2030CO

STin

sura

nce

(M

il E

uro

)

Year

LDV

0

50

100

150

200

250

300

350

400

450

2005 2010 2015 2020 2025 2030CO

STin

sura

nce

(M

il E

uro

)

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030CO

STin

sura

nce

(M

il E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

140

COSTlabour

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STla

bour (M

il Euro

)

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

CO

STla

bour (M

il Euro

)

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STla

bour (M

il Euro

)

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STla

bour (M

il Euro

)

Year

2W

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

CO

STla

bour (M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

141

COSTlabourtax

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STla

bourt

ax

(Mil E

uro

)

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

CO

STla

bourt

ax

(Mil E

uro

)

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STla

bourt

ax

(Mil E

uro

)

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STla

bourt

ax

(Mil E

uro

)

Year

2W

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

CO

STla

bourt

ax

(Mil E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

142

COSTpurchase

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

CO

STpurc

hase

(M

il Euro

)

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

10,000

2005 2010 2015 2020 2025 2030

CO

STpurc

hase

(M

il Euro

)

Year

HDV

0

2,000

4,000

6,000

8,000

10,000

12,000

2005 2010 2015 2020 2025 2030

CO

STpurc

hase

(M

il Euro

)

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

1,600

2005 2010 2015 2020 2025 2030

CO

STpurc

hase

(M

il Euro

)

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

2005 2010 2015 2020 2025 2030

CO

STpurc

hase

(M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

143

COSTrepair

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

90,000

100,000

2005 2010 2015 2020 2025 2030

CO

STre

pair (M

il Euro

)

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

2005 2010 2015 2020 2025 2030

CO

STre

pair (M

il Euro

)

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

CO

STre

pair (M

il Euro

)

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

2005 2010 2015 2020 2025 2030

CO

STre

pair (M

il Euro

)

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

CO

STre

pair (M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

144

COSTrest

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STre

st (M

il Euro

)

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

CO

STre

st (M

il Euro

)

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STre

st (M

il Euro

)

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

CO

STre

st (M

il Euro

)

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

CO

STre

st (M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

145

TAXfuel

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

TAXfu

el (

Mil E

uro

)

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

TAXfu

el (

Mil E

uro

)

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

TAXfu

el (

Mil E

uro

)

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

TAXfu

el (

Mil E

uro

)

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

2005 2010 2015 2020 2025 2030

TAXfu

el (

Mil E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

146

TAXinsurance

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

TAXin

sura

nce

(M

il Euro

)

Year

CAR

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

TAXin

sura

nce

(M

il Euro

)

Year

HDV

0

20

40

60

80

100

120

140

2005 2010 2015 2020 2025 2030

TAXin

sura

nce

(M

il Euro

)

Year

LDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

TAXin

sura

nce

(M

il Euro

)

Year

2W

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

TAXin

sura

nce

(M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

147

TAXownership

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

TA

Xow

ners

hip

(M

il Euro

)

Year

CAR

0

100

200

300

400

500

600

2005 2010 2015 2020 2025 2030

TA

Xow

ners

hip

(M

il Euro

)

Year

HDV

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

TA

Xow

ners

hip

(M

il Euro

)

Year

LDV

0

10

20

30

40

50

60

70

2005 2010 2015 2020 2025 2030

TA

Xow

ners

hip

(M

il Euro

)

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

TA

Xow

ners

hip

(M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

148

TAXregistration

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030TAXre

gis

trat

ion (M

il E

uro

)

Year

CAR

0

1

2

3

4

5

6

2005 2010 2015 2020 2025 2030TAXre

gis

trat

ion (M

il E

uro

)

Year

HDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030TAXre

gis

trat

ion (M

il E

uro

)

Year

LDV

0

1

2

3

4

5

6

7

8

2005 2010 2015 2020 2025 2030TAXre

gis

trat

ion (M

il E

uro

)

Year

2W

0

5

10

15

20

25

30

35

2005 2010 2015 2020 2025 2030TAXre

gis

trat

ion (M

il E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

149

VATfuel

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

VATfu

el (

Mil E

uro

)

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

VATfu

el (

Mil E

uro

)

Year

HDV

0

50

100

150

200

250

300

350

400

450

2005 2010 2015 2020 2025 2030

VATfu

el (

Mil E

uro

)

Year

LDV

0

5

10

15

20

25

30

35

40

45

50

2005 2010 2015 2020 2025 2030

VATfu

el (

Mil E

uro

)

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

VATfu

el (

Mil E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

150

VATpurchase

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

VA

Tp

urc

hase

(M

il Euro

)

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

VA

Tp

urc

hase

(M

il Euro

)

Year

HDV

0

50

100

150

200

250

300

2005 2010 2015 2020 2025 2030

VA

Tp

urc

hase

(M

il Euro

)

Year

LDV

0

50

100

150

200

250

2005 2010 2015 2020 2025 2030

VA

Tp

urc

hase

(M

il Euro

)

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

VA

Tp

urc

hase

(M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

151

VATrepair

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

VA

Tre

pair (M

il E

uro

)

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

VA

Tre

pair (M

il E

uro

)

Year

HDV

0

20

40

60

80

100

120

2005 2010 2015 2020 2025 2030

VA

Tre

pair (M

il E

uro

)

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

VA

Tre

pair (M

il E

uro

)

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

VA

Tre

pair (M

il E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

152

VATrest

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

VATre

st (M

il Euro

)

Year

CAR

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

VATre

st (M

il Euro

)

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

VATre

st (M

il Euro

)

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

VATre

st (M

il Euro

)

Year

2W

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

VATre

st (M

il Euro

)

Year

ALL

BC

MED

PERC_05

PERC_95

153

SumCosts

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

2005 2010 2015 2020 2025 2030

Sum

Co

sts

(M

il E

uro

)

Year

CAR

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

Sum

Co

sts

(M

il E

uro

)

Year

HDV

0

5,000

10,000

15,000

20,000

25,000

30,000

2005 2010 2015 2020 2025 2030

Sum

Co

sts

(M

il E

uro

)

Year

LDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

2005 2010 2015 2020 2025 2030

Sum

Co

sts

(M

il E

uro

)

Year

2W

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

450,000

500,000

2005 2010 2015 2020 2025 2030

Sum

Co

sts

(M

il E

uro

)

Year

ALL

BC

MED

PERC_05

PERC_95

154

FC

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

2005 2010 2015 2020 2025 2030

FC (Thousand T

onnes

)

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

2005 2010 2015 2020 2025 2030

FC (Thousand T

onnes

)

Year

HDV

0

1,000

2,000

3,000

4,000

5,000

6,000

2005 2010 2015 2020 2025 2030

FC (Thousand T

onnes

)

Year

LDV

0

20

40

60

80

100

120

140

160

180

200

2005 2010 2015 2020 2025 2030

FC (Thousand T

onnes

)

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

FC (Thousand T

onnes

)

Year

ALL

BC

MED

PERC_05

PERC_95

155

NOx

0

50

100

150

200

250

300

350

400

2005 2010 2015 2020 2025 2030

NO

x (T

housand T

onnes

)

Year

CAR

0

50

100

150

200

250

2005 2010 2015 2020 2025 2030

NO

x (T

housand T

onnes

)

Year

HDV

0

10

20

30

40

50

60

70

2005 2010 2015 2020 2025 2030

NO

x (T

housand T

onnes

)

Year

LDV

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2005 2010 2015 2020 2025 2030

NO

x (T

housand T

onnes

)

Year

2W

0

100

200

300

400

500

600

700

2005 2010 2015 2020 2025 2030

NO

x (T

housand T

onnes

)

Year

ALL

BC

MED

PERC_05

PERC_95

156

CO

0

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2005 2010 2015 2020 2025 2030

CO

(Thousand T

onnes

)

Year

CAR

0

10

20

30

40

50

60

70

80

2005 2010 2015 2020 2025 2030

CO

(Thousand T

onnes

)

Year

HDV

0

10

20

30

40

50

60

2005 2010 2015 2020 2025 2030

CO

(Thousand T

onnes

)

Year

LDV

0

20

40

60

80

100

120

2005 2010 2015 2020 2025 2030

CO

(Thousand T

onnes

)

Year

2W

0

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2,000

2005 2010 2015 2020 2025 2030

CO

(Thousand T

onnes

)

Year

ALL

BC

MED

PERC_05

PERC_95

157

PM

0

2

4

6

8

10

12

14

2005 2010 2015 2020 2025 2030

PM

(Thousand T

onnes

)

Year

CAR

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

2005 2010 2015 2020 2025 2030

PM

(Thousand T

onnes

)

Year

HDV

0.0

1.0

2.0

3.0

4.0

5.0

6.0

2005 2010 2015 2020 2025 2030

PM

(Thousand T

onnes

)

Year

LDV

0.0

0.1

0.1

0.2

0.2

0.3

2005 2010 2015 2020 2025 2030

PM

(Thousand T

onnes

)

Year

2W

0

5

10

15

20

25

30

2005 2010 2015 2020 2025 2030

PM

(Thousand T

onnes

)

Year

ALL

BC

MED

PERC_05

PERC_95

158

VOC

0

20

40

60

80

100

120

140

160

2005 2010 2015 2020 2025 2030

VO

C (Thousand T

onnes

)

Year

CAR

0

2

4

6

8

10

12

14

16

18

2005 2010 2015 2020 2025 2030

VO

C (Thousand T

onnes

)

Year

HDV

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8.0

9.0

10.0

2005 2010 2015 2020 2025 2030

VO

C (Thousand T

onnes

)

Year

LDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

VO

C (Thousand T

onnes

)

Year

2W

0

20

40

60

80

100

120

140

160

180

200

2005 2010 2015 2020 2025 2030

VO

C (Thousand T

onnes

)

Year

ALL

BC

MED

PERC_05

PERC_95

159

POP

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

2005 2010 2015 2020 2025 2030

PO

P (Thousand T

onnes

)

Year

CAR

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

PO

P (Thousand T

onnes

)

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

2005 2010 2015 2020 2025 2030

PO

P (Thousand T

onnes

)

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

2005 2010 2015 2020 2025 2030

PO

P (Thousand T

onnes

)

Year

2W

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

50,000

2005 2010 2015 2020 2025 2030

PO

P (Thousand T

onnes

)

Year

ALL

BC

MED

PERC_05

PERC_95

160

VKM

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

2005 2010 2015 2020 2025 2030

VKM

(M

illio

n)

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

2005 2010 2015 2020 2025 2030

VKM

(M

illio

n)

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

VKM

(M

illio

n)

Year

LDV

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

2005 2010 2015 2020 2025 2030

VKM

(M

illio

n)

Year

2W

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

800,000

2005 2010 2015 2020 2025 2030

VKM

(M

illio

n)

Year

ALL

BC

MED

PERC_05

PERC_95

161

ANNEX II : uncertainty estimates of the scenarios per vehicle category

162

Scenario 1

VKM

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

2005 2010 2015 2020 2025 2030[x

10^

6]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

LDV

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

2W

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

800,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

163

POP

0

5,000,000

10,000,000

15,000,000

20,000,000

25,000,000

30,000,000

35,000,000

40,000,000

2005 2010 2015 2020 2025 2030

[#]

Year

CAR

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

800,000

900,000

2005 2010 2015 2020 2025 2030

[#]

Year

HDV

0

500,000

1,000,000

1,500,000

2,000,000

2,500,000

3,000,000

3,500,000

2005 2010 2015 2020 2025 2030

[#]

Year

LDV

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

2005 2010 2015 2020 2025 2030

[#]

Year

2W

0

5,000,000

10,000,000

15,000,000

20,000,000

25,000,000

30,000,000

35,000,000

40,000,000

45,000,000

2005 2010 2015 2020 2025 2030

[#]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

164

FC

0

10,000,000

20,000,000

30,000,000

40,000,000

50,000,000

60,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

7,000,000

8,000,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

180,000

200,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

10,000,000

20,000,000

30,000,000

40,000,000

50,000,000

60,000,000

70,000,000

80,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

165

NOx

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

50,000

100,000

150,000

200,000

250,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

166

PM

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

1,000

2,000

3,000

4,000

5,000

6,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

20

40

60

80

100

120

140

160

180

200

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

30,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

167

CO

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

90,000

100,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

2,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

168

VOC

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

2,000

4,000

6,000

8,000

10,000

12,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

50,000

100,000

150,000

200,000

250,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

169

SumCosts

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

450,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

5,000

10,000

15,000

20,000

25,000

30,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

450,000

500,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

170

COSTfuel

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

50,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

171

COSTinsurance

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

50

100

150

200

250

300

350

400

450

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

172

COSTlabour

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

173

COSTlabourtax

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

174

COSTpurchase

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

10,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

2,000

4,000

6,000

8,000

10,000

12,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

1,600

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

175

COSTrepair

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

176

COSTrest

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

177

TAXfuel

05,000

10,00015,00020,00025,00030,00035,00040,00045,00050,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

178

TAXinsurance

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

20

40

60

80

100

120

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

179

TAXownership

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

100

200

300

400

500

600

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

10

20

30

40

50

60

70

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

180

TAXregistration

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

1

2

3

4

5

6

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

1

2

3

4

5

6

7

8

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5

10

15

20

25

30

35

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

181

VATfuel

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

50

100

150

200

250

300

350

400

450

500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

5

10

15

20

25

30

35

40

45

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

182

VATpurchase

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

50

100

150

200

250

300

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

200

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

183

VATrepair

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

20

40

60

80

100

120

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

184

VATrest

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN1

P05_SCEN1

P95_SCEN1

185

Scenario 2

VKM

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

2005 2010 2015 2020 2025 2030[x

10^

6]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

LDV

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

2W

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

800,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

186

POP

0

5,000,000

10,000,000

15,000,000

20,000,000

25,000,000

30,000,000

35,000,000

40,000,000

2005 2010 2015 2020 2025 2030

[#]

Year

CAR

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

800,000

900,000

2005 2010 2015 2020 2025 2030

[#]

Year

HDV

0

500,000

1,000,000

1,500,000

2,000,000

2,500,000

3,000,000

3,500,000

2005 2010 2015 2020 2025 2030

[#]

Year

LDV

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

2005 2010 2015 2020 2025 2030

[#]

Year

2W

0

5,000,000

10,000,000

15,000,000

20,000,000

25,000,000

30,000,000

35,000,000

40,000,000

45,000,000

2005 2010 2015 2020 2025 2030

[#]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

187

CO

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

90,000

100,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

2,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

188

FC

0

10,000,000

20,000,000

30,000,000

40,000,000

50,000,000

60,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

7,000,000

8,000,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

180,000

200,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

10,000,000

20,000,000

30,000,000

40,000,000

50,000,000

60,000,000

70,000,000

80,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

189

NOx

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

50,000

100,000

150,000

200,000

250,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

190

PM

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

1,000

2,000

3,000

4,000

5,000

6,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

20

40

60

80

100

120

140

160

180

200

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

30,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

191

VOC

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

2,000

4,000

6,000

8,000

10,000

12,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

50,000

100,000

150,000

200,000

250,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

192

SumCosts

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

5,000

10,000

15,000

20,000

25,000

30,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

450,000

500,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

193

COSTfuel

05,000

10,00015,00020,00025,00030,00035,00040,00045,00050,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

1,000

2,000

3,000

4,000

5,000

6,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

194

COSTinsurance

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

50

100

150

200

250

300

350

400

450

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

195

COSTlabour

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

196

COSTlabourtax

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

197

COSTpurchase

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

10,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

2,000

4,000

6,000

8,000

10,000

12,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

1,600

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

198

COSTrepair

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

199

COSTrest

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

200

TAXfuel

05,000

10,00015,00020,00025,00030,00035,00040,00045,00050,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

201

TAXinsurance

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

20

40

60

80

100

120

140

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

202

TAXownership

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

100

200

300

400

500

600

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

10

20

30

40

50

60

70

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

203

TAXregistration

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

1

2

3

4

5

6

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

1

2

3

4

5

6

7

8

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5

10

15

20

25

30

35

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

204

VATfuel

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

50

100

150

200

250

300

350

400

450

500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

5

10

15

20

25

30

35

40

45

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

205

VATpurchase

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

50

100

150

200

250

300

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

50

100

150

200

250

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

206

VATrepair

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

20

40

60

80

100

120

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

207

VATrest

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN2

P05_SCEN2

P95_SCEN2

208

Scenario 3

VKM

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

2005 2010 2015 2020 2025 2030[x

10^

6]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

LDV

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

2W

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

800,000

2005 2010 2015 2020 2025 2030

[x10^

6]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

209

POP

0

5,000,000

10,000,000

15,000,000

20,000,000

25,000,000

30,000,000

35,000,000

40,000,000

2005 2010 2015 2020 2025 2030

[#]

Year

CAR

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

800,000

900,000

2005 2010 2015 2020 2025 2030

[#]

Year

HDV

0

500,000

1,000,000

1,500,000

2,000,000

2,500,000

3,000,000

3,500,000

2005 2010 2015 2020 2025 2030

[#]

Year

LDV

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

2005 2010 2015 2020 2025 2030

[#]

Year

2W

0

5,000,000

10,000,000

15,000,000

20,000,000

25,000,000

30,000,000

35,000,000

40,000,000

45,000,000

2005 2010 2015 2020 2025 2030

[#]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

210

CO

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

90,000

100,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

200,000

400,000

600,000

800,000

1,000,000

1,200,000

1,400,000

1,600,000

1,800,000

2,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

211

FC

0

10,000,000

20,000,000

30,000,000

40,000,000

50,000,000

60,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

7,000,000

8,000,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

6,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

180,000

200,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

10,000,000

20,000,000

30,000,000

40,000,000

50,000,000

60,000,000

70,000,000

80,000,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

212

NOx

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

50,000

100,000

150,000

200,000

250,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

1,600

1,800

2,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

100,000

200,000

300,000

400,000

500,000

600,000

700,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

213

PM

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

1,000

2,000

3,000

4,000

5,000

6,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

20

40

60

80

100

120

140

160

180

200

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

30,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

214

VOC

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[ton]

Year

HDV

0

2,000

4,000

6,000

8,000

10,000

12,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

LDV

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

2W

0

50,000

100,000

150,000

200,000

250,000

2005 2010 2015 2020 2025 2030

[to

n]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

215

SumCosts

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

80,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

5,000

10,000

15,000

20,000

25,000

30,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

50,000

100,000

150,000

200,000

250,000

300,000

350,000

400,000

450,000

500,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

216

COSTfuel

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

30,000

35,000

40,000

45,000

50,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

217

COSTinsurance

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

50

100

150

200

250

300

350

400

450

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

218

COSTlabour

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

20,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

219

COSTlabourtax

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5,000

10,000

15,000

20,000

25,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

220

COSTpurchase

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

1,000

2,000

3,000

4,000

5,000

6,000

7,000

8,000

9,000

10,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

2,000

4,000

6,000

8,000

10,000

12,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

1,600

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

140,000

160,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

221

COSTrepair

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

200

400

600

800

1,000

1,200

1,400

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

20,000

40,000

60,000

80,000

100,000

120,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

222

COSTrest

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

223

TAXfuel

05,000

10,00015,00020,00025,00030,00035,00040,00045,00050,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

10,000

20,000

30,000

40,000

50,000

60,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

224

TAXinsurance

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

20

40

60

80

100

120

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

500

1,000

1,500

2,000

2,500

3,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

225

TAXownership

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

100

200

300

400

500

600

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

10

20

30

40

50

60

70

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

226

TAXregistration

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

1

2

3

4

5

6

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

5

10

15

20

25

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

1

2

3

4

5

6

7

8

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

5

10

15

20

25

30

35

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

227

VATfuel

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

50

100

150

200

250

300

350

400

450

500

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

5

10

15

20

25

30

35

40

45

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

228

VATpurchase

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

50

100

150

200

250

300

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

200

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

16,000

18,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

229

VATrepair

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

20

40

60

80

100

120

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

20

40

60

80

100

120

140

160

180

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

2,000

4,000

6,000

8,000

10,000

12,000

14,000

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

230

VATrest

00000111111

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

CAR

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

HDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

LDV

0

0

0

0

0

1

1

1

1

1

1

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

2W

0

100

200

300

400

500

600

700

800

900

2005 2010 2015 2020 2025 2030

[mil E

uro

]

Year

ALL

MED_BC

P05_BC

P95_BC

MED_SCEN3

P05_SCEN3

P95_SCEN3

231

ANNEX III: Description of the DVD contents

Related information has been included in an accompanying DVD. In detail this DVD contains:

• The presentation of the final meeting

• The results in an aggregated format

• The modified model code

• The setup files for the GUI

• Distribution parameters used for the uncertainty estimate

A copy of the DVD can be downloaded from the following address:

http://www.emisia.com/gui/unc.php

The copy of this DVD can be also obtained from the European Commission:

European Commission – DG Climate Action

Marek Sturc (unit CLIMA.A.4)

1049 Bruxelles

Belgium

email: [email protected]

233

ANNEX IV: Description of the full dataset

The full set of the output data as well as the scenario files used for the calculations have been

included in an accompanying HDD. Due to the large number of files and the total size of the

data more than 40 DVDs would be required to include this information. It was decided to use

the external drive instead. In detail this HDD contains 5 folders:

5950 – the 5950 runs executed to perform the uncertainty analysis

BASELINE – the 512 runs executed to perform the sensitivity analysis

SCEN_1 – the 512 runs executed to perform the ownership tax increase uncertainty-sensitivity

analysis

SCEN_2 – the 512 runs executed to perform the fuel cost estimation uncertainty-sensitivity

analysis

SCEN_3 – the 512 runs executed to perform the HDV Euro VI uncertainty-sensitivity analysis

DVD – the contents of the DVD

Uncertainty/Sensitivity analysis of the transport model ... · The uncertainty in the calculations...

Documents

Transcript of Uncertainty/Sensitivity analysis of the transport model ... · The uncertainty in the calculations...