1. Final Presentation of the Project, 21 Jan 2010 1 Uncertainty estimates and guidance for road transport emission calculations A JRC/IES project performed by EMISIA SA Leon Ntziachristos Laboratory of Applied Thermodynamics, Aristotle University Thessaloniki Charis Kouridis, Dimitrios Gktazoflias, Ioannis Kioutsioukis EMISIA SA, Thessaloniki Penny Dilara JRC, Transport and Air Quality Unit http://ies.jrc.ec.europa.eu/ http://www.jrc.ec.europa.eu/ Panagiota.dilara@jrc.ec.europa.eu

2. Final Presentation of the Project, 21 Jan 2010 2 Project ID Project was initiated Dec. 17, 2008 and lasted for about one year Objectives: –Evaluate the uncertainty linked with the various input parameters of the COPERT 4 model, –Assess the uncertainty of road transport emissions in two test cases, at national level, –Make possible to simulate uncertainties of COPERT 4 for any country (software development) –Prepare guidance on the assessment of uncertainty for the Tier 3 methods (COPERT 4).

3. Final Presentation of the Project, 21 Jan 2010 3 Operational Definitions Item: Any value required by the software to calculate the final output Input Variable: Any item for which actual values are not included in the software (stock size, mileage, speeds, temperatures, …) Internal Parameter: An item included for which actual values are included in the software and have been derived from experiments (emission factors, cold-trip distance, …) Uncertainty: Variance of final output (pollutant emission) due to the non exact knowledge of input variables and experimental variability of internal parameters Sensitivity: Part of the output variance explained by the variance of individual variables and parameters

4. Final Presentation of the Project, 21 Jan 2010 4 Approach Select two countries to simulate different cases –Italy: South, new vehicles, good stock description –Poland: North, old vehicles, poor stock description Quantify uncertainty range of variables and parameters Perform screening test to identify influential items (~500 runs) Perform uncertainty simulations to characterise total uncertainty, including only influential items (~6000 runs) Limit output according to statistical fuel consumption

5. Final Presentation of the Project, 21 Jan 2010 5 Items for which uncertainty has been assessed ItemDescriptionItemDescription NcatVehicle population at category levelLFHDVLoad Factor Nsub Vehicle population at sub-category level tmin Average min monthly temperature Ntech Vehicle population at technology level tmax Average max monthly temperature MtechAnnual mileageMm,techMean fleet mileage UStechUrban shareRVPFuel reid vapour pressure HstechHighway shareH:CHydrogen-to-carbon ratio RStechRural shareO:COxygen-to-carbon ratio USPtechUrban speedSSulfur level in fuel HSPtechHighway speedehot,techHot emission factor RSPtechRural speedecold/ehot,techCold-start emission factor LtripMean trip lengthbCold-trip distance

6. Final Presentation of the Project, 21 Jan 2010 6 Variance of the total stock ITALY ACEAACEMACIEurostat μσ 2005 Passenger Cars34 667 485 34 636 40034 657 12317 947 Light Duty Vehicles3 257 525 3 633 9003 445 713266 137 Heavy Duty Vehicles1 070 308 958 4001 014 35479 131 Buses94 437 94 10094 325195 Mopeds 5 325 0004 560 907 4 942 954540 295 Motorcycles 4 938 359 4 933 6004 936 7732 748 POLAND ACEAACEMPoland StatEurostat μσ 2005 Passenger Cars12 339 353 12 339 000 12 339 118204 Light Duty Vehicles1 717 435 2 304 5002 178 0002 066 645308 968 Heavy Duty Vehicles587 070 737 000662 035106 017 Buses79 567 79 60080 00079 722241 Mopeds 337 511 0 Motorcycles 753 648 754 000753 824249

7. Final Presentation of the Project, 21 Jan 2010 7 Subsector variance Italy Standard deviation is produced by allocating the unknown values to the smaller class, the larger class and uniformly between classes

8. Final Presentation of the Project, 21 Jan 2010 8 Subsector variance Poland Passenger cars: standard deviation calculated as one third of the difference between national statistics and FLEETS Light Duty Vehicles: uncertainty of stock proportionally allocated to stock of diesel and gasoline trucks. Other vehicle categories: standard deviation was estimated as 7% of the average (assumption).

9. Final Presentation of the Project, 21 Jan 2010 9 Technology classification variance – 1(2) Italy: Exact technology classification Poland: Technology classification varying, depended on variable scrappage rate Boundaries Introduced: Age of five years: ±5% Age of fifteen years: ±10% All scrappage rates respecting boundaries are accepted → these induce uncertainty 100 pairs finally selected by selecting percentiles

10. Final Presentation of the Project, 21 Jan 2010 10 Fleet Breakdown Model The stock at technology level is calculated top-down by a fleet breakdown model (FBM), in order to respect total uncertainty at sector, subsector and technology level. That is, the final stock variance should be such as not to violate any of the given uncertainties at any stock level. The FBM operates on the basis of dimensionless parameters to steer the stock distribution to the different levels. Details in the report, p.44.

11. Final Presentation of the Project, 21 Jan 2010 11 Example of technology classification variance Example for GPC<1.4 l Poland Standard deviation: 3.7%, i.e. 95% confidence interval is ±11%

12. Final Presentation of the Project, 21 Jan 2010 12 Emission Factor Uncertainty Emission factor functions are derived from several experimental measurements over speed

13. Final Presentation of the Project, 21 Jan 2010 13 Emission Factor Uncertainty Fourteen speed classes distinguished from 0 km/h to 140 km/h

14. Final Presentation of the Project, 21 Jan 2010 14 Emission Factor Uncertainty A lognormal distribution is fit per speed class, derived by the experimental data.

15. Final Presentation of the Project, 21 Jan 2010 15 Mileage Uncertainty – M0 Mileage is a function of vehicle age and is calculated as the product of mileage at age 0 (M0) and a decreasing function of age: M0 was fixed for Italy based on experimental data M0 was variable for Poland (s=0.1*M0) due to no experimental data available

16. Final Presentation of the Project, 21 Jan 2010 16 Mileage Uncertainty – Age The uncertainty in the decreasing mileage function with age was assessed by utilizing data from all countries (8 countries of EU15) The boundaries are the extents from the countries that submitted data Bm and Tm samples were selected for all curves that respected the boundaries

17. Final Presentation of the Project, 21 Jan 2010 17 Other variables - temperature Uncertainty of other variables was quantified based on literature data where available or best guess assumptions, when no data were available. Models were built for the temperature distribution over the months for the two countries.

18. Final Presentation of the Project, 21 Jan 2010 18 Statistical approach 1.Prepare the Monte Carlo sample for the screening experiment using the Morris design. 2.Execute the Monte Carlo simulations and collect the results. 3.Compute the sensitivity measures corresponding to the elementary effects 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. 7.Determine the input factors that are most responsible for producing model outputs within the targeted bounds of fuel consumption.

19. Final Presentation of the Project, 21 Jan 2010 19 Results – Screening test Italy

20. Final Presentation of the Project, 21 Jan 2010 20 Results – Screening test Poland

21. Final Presentation of the Project, 21 Jan 2010 21 Results – Influential Variables VariableSignificant for ItalySignificant for Poland Hot emission factor  Cold overemission  Mean trip distance  Oxygen to carbon ratio in the fuel  Population of passenger cars  - Population of light duty vehicles  Population of heavy duty vehicles  Population of mopeds  - Annual mileage of passenger cars  Annual mileage of light duty vehicles  Annual mileage of heavy duty vehicles  Annual mileage of urban busses-  Annual mileage of mopeds/motorcycles  - Urban passenger car speed  Highway passenger car speed  - Rural passenger car speed  - Urban speed of light duty vehicles  - Urban share of passenger cars  - Urban speed of light duty vehicles-  Urban speed of busses-  Annual mileage of vehicles at the year of their registration-  The split between diesel and gasoline cars-  Split of vehicles to capacity and weight classes-  Allocation to different technology classes- 

22. Final Presentation of the Project, 21 Jan 2010 22 Results – total uncertainty Italy w/o fuel correction

23. Final Presentation of the Project, 21 Jan 2010 23 Results – Descriptive statistics of Italy w/o fuel correction COVOCCH 4 NO X N2ON2OPM 2.5 PM 1 0 PM ex h FCCO 2 CO 2e Mean (t)1,215335216133.232362736,885110,570111,999 Median1,150329196032.932362636,828110,357111,751 St. Dev.371609921.14542,4847,5967,902 Coef. Var. (%) 301844153313 14777

24. Final Presentation of the Project, 21 Jan 2010 24 Results – Necessary fuel correction for Italy Unfiltered dataset: Std Dev = 7% of meanFiltered dataset: 3 Std Dev = 7% of mean

25. Final Presentation of the Project, 21 Jan 2010 25 Correction of sample required Cumulative distributions of unfiltered (red) and of filtered (blue) datasets eEF, milHDV and milLDV are not equivalent A corrected dataset was built to respect the fuel consumption induced limitations

26. Final Presentation of the Project, 21 Jan 2010 26 Results – total uncertainty Italy with corrected sample

27. Final Presentation of the Project, 21 Jan 2010 27 Confirmation of corrected sample

28. Final Presentation of the Project, 21 Jan 2010 28 Results – Descriptive statistics of Italy with corrected sample COVOCCH 4 NO x N2ON2OPM 2.5 PM 10 PM exh FCCO 2 CO 2e Mean1,134325196143.132372736,945110,735112,094 Median1,118324186082.932362736,901110,622111,941 St. Dev.218387590.83331,2414,0794,203 Variation (%) 1912341026989344

29. Final Presentation of the Project, 21 Jan 2010 29 Italy – Contribution of items to total uncertainty 1(2) VOCSISI S TI NO X SISI S TI PM2.5SISI S TI PM 10 SISI S TI PMexhSISI S TI eEF0.630.78eEF0.760.85eEF0.720.86eEF0.720.84eEF0.720.86 ltrip0.080.22milHDV0.120.22milHDV0.080.22milHDV0.080.21milHDV0.090.23 eEFratio0.050.15HDV0.010.08ltrip0.010.13ltrip0.010.13ltrip0.010.14 milMO0.050.17PC00.08HDV0.010.12HDV0.010.11HDV0.010.14 VUPC0.020.16ltrip00.08eEFratio0.010.13milPC0.010.12eEFratio0.010.14 O2C0.020.15LDV00.08LDV0.010.12LDV0.010.11LDV0.010.12 HDV0.010.13VHPC00.1milPC0.010.13eEFratio0.010.13milPC0.010.14 MOP0.010.14VUPC00.08PC00.13PC0.010.13milMO0.010.12 milHDV0.010.14O2C00.08milMO00.11milMO0.000.10PC0.000.14 LDV0.010.12milPC00.08milLDV00.12milLDV0.000.12milLDV0.000.13 PC00.15UPC00.08VHPC00.13VHPC0.000.12VHPC0.000.14 VRPC00.15MOP00.09MOP0.000.13MOP0.000.12MOP0.000.15 milPC00.14eEFratio00.1O2C00.11O2C00.1O2C00.12 VHPC00.13milMO00.07UPC00.12UPC00.11UPC00.13 milLDV00.14VRPC00.1VUPC00.12VRPC00.12VRPC00.13 UPC00.14milLDV00.08VRPC00.12VUPC00.11VUPC00.13 ΣS i 0.913.030.912.270.872.780.872.690.882.96

30. Final Presentation of the Project, 21 Jan 2010 30 Italy – Contribution of items to total uncertainty 2 COSISI S TI N2ON2OSISI CH 4 SISI S TI CO 2 SISI S TI FCSISI S TI eEF0.440.56eEFratio0.590.76eEFratio0.610.76eEF0.400.51eEF0.430.54 eEFratio0.190.29ltrip0.060.37eEF0.130.29eEFratio0.100.22eEFratio0.110.24 ltrip0.050.21VUPC0.060.23ltrip0.030.26milHDV0.090.2milHDV0.090.21 O2C0.030.16eEF0.040.16VUPC0.010.19milPC0.050.17milPC0.050.17 VUPC0.030.17milHDV0.010.14HDV00.16ltrip0.040.21ltrip0.040.21 milMO0.010.13milPC0.010.13milMO00.13O2C0.040.16HDV0.020.13 HDV0.010.15HDV00.13LDV00.16HDV0.020.13VUPC0.010.11 LDV00.12MOP00.18MOP00.18VUPC0.010.11PC0.010.12 VHPC00.15LDV00.13VHPC00.21PC0.010.12LDV0.010.13 VRPC00.17milLDV00.11milHDV00.16LDV0.010.12UPC0.010.14 MOP00.17milMO00.11VRPC00.2UPC0.010.14MOP0.000.12 UPC00.15VRPC0.000.18UPC00.16MOP0.000.12milLDV0.000.12 PC00.14UPC00.13PC00.2milLDV00.12VHPC00.12 milHDV00.12VHPC00.25milLDV00.13VHPC00.11O2C00.12 milPC00.15O2C00.24milPC00.16milMO00.14milMO00.14 milLDV00.1PC00.17O2C00.21VRPC00.12VRPC00.12 ΣS i 0.792.940.793.440.803.580.782.680.792.72

31. Final Presentation of the Project, 21 Jan 2010 31 Results – Italy/Poland Comparison CaseCOVOCCH 4 NO x N2ON2O PM 2.5 PM 10 PM exh FCCO 2 CO 2e Italy w/o FC301844153313 14777 Italy w. FC1912341026989344 Poland w/o FC201857172818171911 12 Poland w. FC1715541224131214888 Table shows Coefficient of Variance = Std. Deviation / Mean

32. Final Presentation of the Project, 21 Jan 2010 32 Comparison with Earlier Work The improvements of the current study, in comparison to the previous one (Kioutsioukis et al., 2004) for Italy, include: use of the updated version of the COPERT model (version 4) incorporation of emission factors uncertainty for all sectors (not only PC & LDV) and all vehicle technologies through Euro 4 (Euro V for trucks) application of a more realistic fleet breakdown model due to the detailed fleet inventory available application of a detailed and more realistic mileage module based on the age distribution of the fleet (decomposition down to the technology level) inclusion of more uncertain inputs: cold emission factors, hydrogen-to- carbon ratio, oxygen-to-carbon ratio, sulphur level in fuel, RVP. validation of the output and input uncertainty

33. Final Presentation of the Project, 21 Jan 2010 33 Conclusions – 1(3) The most uncertain emissions calculations are CH4 and N2O followed by CO. CO2 is calculated with the least uncertainty, as it directly depends on fuel consumption, followed by NOx and PM2.5 because diesel are less variable than gasoline emissions. The correction for fuel consumption is very critical as it significantly reduces the uncertainty of the calculation in all pollutants. Despite the relatively larger uncertainty in CH4 and N2O emissions, the uncertainty in total GHG emissions is dominated by CO2

34. Final Presentation of the Project, 21 Jan 2010 34 Conclusions – 2(3) The Italian inventory uncertainty is affected by: hot emission factors [eEF]: NOx (76%), PM (72%), VOC (63%), CO (44%), FC (43%), CO2 (40%), CH4 (13%) cold emission factors [eEFratio]: CH4 (61%), N2O (59%), CO (19%), FC (11%), CO2 (10%), VOC (5%) mileage of HDV [milHDV]: NOx (12%), PM (8-9%), FC (9%), CO2 (9%). mean trip length [ltrip]: VOC (8%), N2O (6%), CO (5%)

35. Final Presentation of the Project, 21 Jan 2010 35 Conclusions – 3 The Polish inventory uncertainty is affected by: mileage parameter [eM0]: FC (68%), CO2 (67%), NOx (35%), VOC (27%), PM (25-31%), CO (22%), N2O (14%). cold emission factors [eEFratio]: CH4 (56%), N2O (48%), CO (15%), VOC (8%). hot emission factors [eEF]: PM (52-55%), NOx (49%), VOC (20%), CO (15%), CH4 (12%), N2O(11%), FC (10%), CO2 (9%). mean trip length [ltrip]: VOC (23%), CO (20%). Note: the technology classification appears important for the uncertainty in conjunction to other variables

36. Final Presentation of the Project, 21 Jan 2010 36 More Information Review revised (still draft) Road Transport Guidebook Chapter at: http://transportpanel.jrc.it –It contains an updated “4. Data Quality” section Ask Leon (leon@auth.gr) or Penny (panagiota.dilara@jrc.ec.europa.eu) for full report of this work (no hyperlink yet  ). Ask Leon for simulation software to apply in other countries

37. Final Presentation of the Project, 21 Jan 2010 37 Recommendations There is little the Italian expert can do to reduce uncertainty. Most of it comes from emission factors Better stock and mileage description is required for Poland to improve the emission inventory.

38. Final Presentation of the Project, 21 Jan 2010 38 Thank you for your attention! Questions?