- Email: [email protected]
Uncertainties of modelling emissions from road transport
Atmospheric Environment 34 (2000) 4603}4610
Uncertainties of modelling emissions from road transport J. KuK hlwein*, R. Friedrich Institute of Energy Economics and the Rational Use of Energy (IER), University of Stuttgart, He}bru( hlstr. 49a, D-70565 Stuttgart, Germany Received 3 September 1999; accepted 10 May 2000
Abstract To determine emission data from road transport, complex methods and models are applied. Emission data are characterized by a huge variety of source types as well as a high resolution of the spatial allocation and temporal variation. So far, the uncertainties of such calculated emission data have been largely unknown. As emission data is used to aid policy decisions, the accuracy of the data should be known. So, in the following, the determination of uncertainties of emission data is described. Using the IER emission model for generating regional or national emission data, the uncertainties of model input data and the total errors on di!erent aggregation levels are exemplarily investigated for the pollutants NO and NMHC in 1994 for the area of West Germany. The results of statistical error analysis carried out for V annual emissions on road sections show variation coe$cients (68.3% con"dence interval) of 15}25%. In addition, systematic errors of common input data sets have been identi"ed especially a!ecting emissions on motorway sections. The statistical errors of urban emissions with warm engine on town level amount to 35%. Therefore they are considerably higher than the errors outside towns. Error ranges of additional cold start emissions determined so far have been found in the same order. Additional uncertainties of temporally highly resolved (hourly) emission data depend strongly on the daytime, the weekday and the road category. Variation coe$cients have been determined in the range between 10 and 70% for light-duty vehicles and between 15 and 100% for heavy-duty vehicles. All total errors determined here have to be regarded as lower limits of the real total errors. 2000 Elsevier Science Ltd. All rights reserved. Keywords: Road tra$c; Statistical error propagation; Systematic error analysis; Temporally highly resolved emission data; Spatial intersection
1. Introduction Emission data from road transport are used as input data for modelling atmospheric processes of transport, transformation and deposition. Therefore, they are essential for numeric calculations of concentrations and depositions of primary and secondary pollutants and origin for developments of clean air strategies. Up to now there has been only little knowledge about the reliability of such calculated emission data. It was not state of the art to state uncertainty ranges together with the emission data; instead only qualitative information about uncertainties (e.g. as ranking) was given.
* Corresponding author. E-mail address: [email protected] (J. KuK hlwein).
Past estimates of uncertainties of European VOC-emissions from motor cars have been carried out e.g. by (Eggleston, 1993) and (Andrias et al., 1993) for less detailed emission calculation methods and with older emission factors than used here. The analyses described here look with extended detail into the statistical deviations of the input data used for calculating emissions. To estimate uncertainties of modelled emissions, it is necessary to determine the statistical error ranges of input data using suitable methods and to combine them to a total error by sensitivity studies and statistical error calculations. The results presented in the following are related to the high-resolution emission factors and the associated tra$c data which are available for the area of West Germany. The investigations are restricted to the pollutants non-methane hydrocarbons (NMHC) and NO (NO and NO ) as they take main V responsibility for the creation of atmospheric oxidants.
1352-2310/00/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved. PII: S 1 3 5 2 - 2 3 1 0 ( 0 0 ) 0 0 3 0 2 - 2
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
4604
The underlying emission calculation model was developed at the IER, university of Stuttgart (John, 1999), and is similar in structure to the emission models used in European scale (MEET, 1999), but more detailed. The model is based on current parameters for regional emission modelling with high temporal and spatial resolution. Emissions are calculated for all segments of non urban roads and urban areas. The basis of the method used here is the `UBA-Handbooka (Umweltbundesamt, 1999), that provides detailed emission factors as function of vehicle layers (de"ned by vehicle category, type of engine, age, pollution reduction technology, exhaust gas regulation and class of cubic capacity resp. vehicle mass), tra$c situations (driving patterns described by distributions of driving conditions) and road gradients. The average daily tra$c volumes of seven di!erent vehicle categories are available for nearly all individual non-urban main road sections in Germany from tra$c counts that take place in a "ve year rotation. The appropriate driving patterns to be used for a road segment are estimated from information on road category, number of lanes, road gradient and mean tra$c volume. Information about individual road sections are available from o$cial road databases of some German states. The temporal resolution of the tra$c volume data is derived from a detailed analysis of data from automatic permanent counting stations that continuously measure the number of heavy- and light-duty vehicles on selected locations in Germany. If no speci"c tra$c volume data for an individual road section is available, mean values from comparable road sections are transferred. Further di!erentiations of tra$c volume data from vehicle categories to vehicle layers are achieved by transferring mean results of license plate number evaluations (Umweltbundesamt, 1999) that are available for three di!erent road types (motorways, rural and urban driving). A simpli"ed functional relation of input parameters to calculate hourly emissions of one selected road section (outside towns) is given by
E"N¸ IT MDT T T T
dp (dp EF ) Z , (1) T N T J J N E F P N J where E is the calculated hourly emission, N the number of days in 1994 ("365), L the length of road section, IT the interpolation factor for calculation of tra$c T volume in 1994 from counts carried out in 1990 and 1995, di!erentiated by vehicle category v, MDT the T mean daily tra$c volume between 1990 and 1995, di!erentiated by vehicle category v, dp the share of driving T N performance of one driving pattern p in the total driving performance of a vehicle category v (driving pattern mix), dp ¸ the share of driving performance of one vehicle T layer l in the total driving performance of a vehicle
category v (#eet composition), EF the emission factor, J N E di!erentiated by vehicle layer l, driving pattern p and road gradient class g, and Z the fraction of annual F P emission emitted in hour h for road category r. With the results of uncertainty calculations, it is possible to "lter out those input data whose errors contribute most to the error of total emissions. Therefore, it is possible to "nd out the weak points of the used emission model and to do e$cient model improvements in future. Here the term `uncertaintya is used as a general expression of unknown possible deviations of true emissions from calculated emission data. If uncertainties can be quanti"ed the term `errora is used. We have to distinguish between statistical and systematic errors. Statistical errors may be expressed e.g. as standard deviations. Quantifying systematic errors is a much more di$cult question than calculating statistical errors, because independent data sets are necessary to do that. Such additional data sets are not available in each case. The authors found some systematic errors in commonly used input data sets and present the results below. These results can be used to correct modelled emission data from road transport based on these input data sets. Only a few selected results of the many error analyses carried out by the authors can be presented here. A full account is given in (KuK hlwein et al., 1999).
2. Statistical methods for error estimation The quanti"cation of extensive emissions from road transport (or even emissions from single road sections) by measurements requires high e!orts. So, usually emissions are determined by model calculations based on results from mediating observations. The emission E of a pollutant can be expressed as a function of input parameters x , x , 2, x . These input parameters are emission fac L tors, tra$c parameters like tra$c volume, vehicle #eet composition, driving pattern mix etc. and road speci"c parameters like speed limit, number of tra$c lanes, course of road, gradient etc. (see Eq. (1)). The spreads of statistical errors when measuring simple physical quantities can be calculated by basic statistical methods such as standard deviation (standard error of single values) or as standard deviation of the arithmetic mean of sample, if the single measuring values of the random sampling are known. By variation of a single input parameter x in the G calculation model, the sensitivity *E/*x can be deterG mined on the basis of partial derivation of the emission E with respect to the chosen input parameter. The absolute error of emission *E related to the error of the input G parameter x results from the multiplication of the sensiG tivity *E/*x with the calculated or estimated absolute G error dx or *x . G G
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
With the help of the error propagation law on conditions that E the error estimations of the input parameters are correct, E the input parameters are statistically independent from each other and E there is a high degree of linearity between E and x G (*E/*x + const.) or the errors of the input parameters G *x are small compared with x , G it is possible to calculate the mean total error *E of the emission E for any functional relations between the input parameters x according to (Hartung et al., 1995) G
L *E L *x " (*E ) (2) G G *x G G G where n is the number of input parameters, *E/*x the G sensitivity of calculated total emission related to the input parameter x , *x the error of input parameter x G G G (e.g. standard deviation) and *E the single error of G calculated total emission E related to *x G In case of emission modelling the calculation of error ranges of the input parameters in the form of standard deviations and the determination of systematic errors is not always possible. The main reasons are, that: *E"
E the error of the input parameter (and partially the parameter itself) is not quanti"able because of its complex physical structure and E for parts of the data base neither the single measured values nor the standard deviations are known. In these cases plausible assumptions about the statistical error ranges of the input parameter have to be made, and the e!ects of these errors on the emission have to be estimated by model runs. The following results of statistical calculations are given as standard deviations (resp. variation coe$cients). This means that the estimated error ranges describe a 68.3% con"dence interval. In case of emission modelling usually no higher degree of con"dence (e.g. 95%) is reported because of the high general level of errors of calculated data.
3. Results 3.1. Error ranges for emissions from diwerent road categories Error estimations of calculated annual emissions have been carried out for di!erent road categories according to the method described. In addition to the statistical errors, systematic errors have been quanti"ed and documented separately, as far as data for the estimation
4605
of these errors were available. The error calculations have been carried out for the year 1994 and for the area of West Germany. The statistical error ranges of the "ve main input parameters have been quanti"ed as follows: Mean gradients for individual road sections are available in gradient classes of 1% steps from the road databases, resulting in a maximum error of $0.5%. Errors in tra$c volume data are caused by count errors and by errors due to extrapolation from count samples to annual values. They have been estimated on a random basis by comparisons of the tra$c volume data used for modelling with annual tra$c volume data from automatic permanent counting stations. Usually, information about the driving pattern mixes of individual road sections is missing, because sectionspeci"c data about speed limits, road courses, etc., are not available. Plausible assumptions have been made to estimate possible ranges of driving patterns for each road category. Fleet compositions vary from one region to another. The standard errors and the e!ects of these spatial variations on the emission rates have been quanti"ed by evaluating vehicle stock data of di!erent administrative districts in West Germany. Statistical errors of commonly used emission factors have been investigated by the TUG V Rheinland (Hassel et al., 1998a, b). Standard errors from the single values of extensive dynamometer measurements have been calculated per vehicle layer. These highly resolved interim results have been aggregated to standard errors of the most important driving patterns. 3.1.1. Emissions from vehicles with warm engine outside towns The results of the error estimations for the road categories motorway and federal road are presented in Fig. 1. The results are related to road sections on the following conditions: E gradient class: 0}1%, E full extent counts at the o$cial German federal tra$c census in 1990 and 1995 (8 counting days, 36 counting hours), E estimated (not locally recorded) section-speci"c route situation (speed limit, bends, etc.), E estimated (not measured) section-speci"c tra$c data (driving pattern mix and #eet composition). The determined statistical errors (variation coe$cients) of calculated annual emissions have been found in the range of 21}26% for NMHC and 16}22% for NO V for both road categories. Especially, on motorway sections a negative error (emissions are calculated too high) is more probable than a positive one because of asymmetrical e!ects of the errors of the driving pattern mix.
4606
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
Fig. 1. Error ranges of annual emissions on road sections in West Germany, 1994: syst. } systematic error (%); stat. } statistical error: variation coe$cient (%).
The emission factors represent the most important source of error especially for NMHC. Since the frequency distribution of the driving pattern mix is not normal distributed, the statistical error due to driving pattern mix is not symmetrical. In addition to the statistical errors, two important sources of systematic errors (#eet composition and emission factors) have been found. These are described in the following: The term `#eet compositiona describes the further di!erentiation of tra$c volume which is already di!erentiated to vehicle categories by manual tra$c counts. The tra$c volume is now further distributed among vehicle layers which are de"ned by the criterions: type of engine, exhaust gas puri"cation system, legislative exhaust gas limit, age, cubic capacity and vehicle mass. These highly resolved tra$c volume data are available e.g. from the `UBA-Handbooka (Umweltbundesamt, 1999) for the three road categories motorways, rural and urban, related to the German situation. Comparing vehicle stock data (Kraftfahrt-Bundesamt, 1994) that are projected to the dynamic #eet composition by concept- and sizerelated projection factors (Steven, 1995), with the data from the `UBA-Handbooka shows, that shares of vehicles without or with older exhaust gas puri"cation systems are much higher in the projected stock data.
Assuming that our data about #eet compositions are nearer to reality than the commonly used UBA-Handbook data (that could not be checked as the derivation is not published), we conclude, that the use of the UBA data leads to calculated emissions, that are systematically too low. Systematic deviations of emission factors can be partially deduced by interpreting investigations made by TUG V Rheinland (Hassel et al., 1994). Emission factors for di!erent driving cycles are usually constructed by "rst dividing the test driving cycle and the corresponding emissions into parts (modal analysis of e.g. 1 s each) that are de"ned by present speed (v) and acceleration (a) of the tested vehicle. The measured emissions in high temporal resolution are classi"ed by a v}a-grid. To calculate the integral emission factor of any driving pattern, it is necessary to add up the mean emission values of the di!erent grid cells by the speci"c frequency distribution of driving conditions. These added up emission factors are published in large databases (Umweltbundesamt, 1999) and are used for emission modelling. Some of these modal measured emission factors have been exemplarily compared to integral values (bag analysis), measured over the whole driving cycle. Occuring deviations are caused by systematic errors of the method used for determining emission factors.
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
The resulting systematic errors are conspicuous. Whereas these two error groups have di!erent signs at the federal road sections, which means that they compensate each other, at motorway sections the same signs lead to systematic deviations of about #38% (NMHC) and #22% (NO ). This means that there is a systematic V underestimation of emissions on motorways when using the usual input data sets. 3.1.2. Emissions from vehicles with warm engine inside towns In contrast to non-urban tra$c, no detailed standardized tra$c volume data are available for individual road sections inside towns. (Of course, in many towns tra$c counts have been made } but these are not standardized on a regional or national level, so it takes too much e!ort to use them.) Total tra$c volumes on town level are available for some towns only. So additional infrastructure data (e.g. number of inhabitants) have to be included in the emission model. With it, total urban tra$c volumes on federal or regional level are distributed to the individual towns. As shown in Fig. 2, the statistical total errors have been determined to be approx. 37% (NMHC) and 35% (NO ). V Thus, they are considerably higher than the errors of
4607
road sections outside towns. There is an additional systematic error of about !11% for NMHC and !8% for NO . The two input parameters contributing most to the V total error are the tra$c volume and the emission factors. The step of distributing total tra$c volumes to town level using the number of inhabitants per town involve a variation coe$cient of about 23% (Schmitz et al., 1997). Compared to this, the determination of the total tra$c volume inside towns for all motor vehicles (e.g. for the old West German states) is quite reliable (variation coe$cient: approx. 11%) and the errors of di!erentiation of the tra$c volume to di!erent vehicle categories are relatively small (3.8% in case of NMHC and 11.9% in case of NO ). The uncertainties of emission factors for urban V transport are slightly higher than those for non-urban road sections. 3.1.3. Cold start emissions A calculation model, which contains mean distributions for the driving pattern mix, the #eet composition, parking times before start, driving distances, outdoor temperatures and the daily course of starting frequencies (Umweltbundesamt, 1999), lays the foundations for the error estimations of additional cold start emissions. The number of starts per town are deduced from the tra$c volume and the population data. The errors of additional cold start emissions are mainly marked by the spatial di!erentiation of the number of starts on town level and the cold start emission factors. Error contributions of more than 10% have been determined for the number of drives related to tra$c volume and the annual distribution of temperatures (only NMHC). Similar to the tra$c volume in case of determining emissions with warm engine inside towns, necessary information about the number of starts is not available on town level. So the total number of starting processes is determined from the total driving performance of passenger cars (in West Germany) and a mean driving distance. Then the total number of starts is distributed to the individual towns according to the numbers of inhabitants. The statistical uncertainties of the spatial distribution of starts correspond with those of the spatial distribution of tra$c volume. They amount to 23% with a slight asymmetry (Fig. 3). The statistical total errors up to now amount to about 39% in case of NMHC and 33% in case of NO . SystemV atic deviations have been determined from the available data as about !7% for NMHC and !19% for NO . V 3.2. Uncertainties of temporal resolution of emissions
Fig. 2. Error ranges of annual emissions with warm engine inside towns in West Germany, 1994: syst. } systematic error (%); stat. } statistical error: variation coe$cient (%).
In order to model temporally highly resolved emission data, mean annual, weekly and daily courses per road category are applied. Temporally highly resolved tra$c volume data (hourly intervals) are available from tra$c counts by automatic permanent counting stations.
4608
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
Comparing hourly shares in annual tra$c volumes at de"nite hours, deviation ranges between di!erent road sections of the same road category can be derived. The deviations of variation coe$cients caused by the daily courses can be roughly separated into two "elds: (1) late morning till evening, and (2) night till early morning. Besides the deviations caused by the daily courses there are signi"cant di!erences between the weekdays (Monday}Friday, Saturday, Sunday). Deviations derived from the data available from automatic permanent counting stations in West Germany in 1994 for the road categories motorway, federal road and state road have been summarized and are shown in Table 1. In general, the variation coe$cient increases with decreasing tra$c volume. So an increase with decreasing order of road category (motorwayPstate road) can be seen. The deviations on weekends are higher than those on working days. An exception was found for LDV during the night hours where variation coe$cients stay unchanged, respectively, decrease slightly on weekends. For HDV in all examined cases higher deviation ranges occur compared to LDV.
4. Conclusions and outlook
Fig. 3. Error ranges of annual additional cold start emissions from passenger cars in West Germany, 1994: syst. } systematic error (%); stat. } statistical error: variation coe$cient (%).
The results of statistical error analyses carried out for annual emissions in 1994 on road sections show variation coe$cients (68.3% con"dence interval) between 21 and 26% in case of NMHC and between 16 and 22% in case of NO . In addition systematic errors of commonly used V
Table 1 Error ranges of hourly di!erentiation of annual tra$c volumes for di!erent road categories (outside towns), weekdays and daytime Road category
Motorway
LDV/HDV
LDV HDV
Federal road
LDV HDV
State road
LDV HDV
Day/night
Day Night Day Night Day Night Day Night Day Night Day Night
Light-duty vehicles (LDV); heavy-duty vehicles (HDV). Day: late morning till evening; night: night till early morning.
Variation coe$cient (%) Monday}Friday
Saturday
Sunday
10}15 20}45 20}30 30}45 10}15 25}50 15}25 35}70 10}25 25}60 30}40 50}95
10}20 15}25 30}40 35}55 10}20 20}35 25}40 45}75 10}35 30}45 35}65 60}100
20}30 20}30 30}70 45}70 20}30 25}50 45}60 55}90 30}50 20}45 65}90 70}100
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
input data sets have been identi"ed especially a!ecting emissions on motorway sections with#38% for NMHC and#22% for NO . The statistical uncertainties of anV nual emissions with warm engine inside towns on town level amount to 37% (NMHC) and 35% (NO ); they are V much higher than the errors outside towns. Error ranges of additional cold start emissions determined so far are in the same order of magnitude. For two reasons all determined total errors have to be regarded as lower limit of the real total errors : (1) Because of the incomplete data situation the uncertainties of some input parameters could be quanti"ed only partially. There are considerable data gaps especially for emission factors (vehicles with newer technology, cold start, road gradients, heavy duty vehicles, motorcycles, running losses), the driving patterns on ordinary roads and inside towns and the tra$c volume data. Additional investigations are necessary to close these gaps. (2) The mean total errors have been determined according to the method of error propagation on condition that the input parameters are statistically independent from each other. Taking into account possible correlations between the input parameters, the true total error will be found between the mean total error (Eq. (2)) and the maximum total error that results from adding up the amounts of all single errors *E . In most cases it is not possible to calcuG late the true total statistical errors taking into account possible correlations because currently available data is not su$cient to do essential quanti"cations of covariances. However, it is the authors' impression, that the true errors in general are closer to the mean total errors than to the maximum total errors, because of the small in#uences of possible correlations. Signi"cant systematic errors could be recognized for the input parameters emission factors and #eet composition by including additional data sets. To clarify these systematic deviations further dynamometric measurements are necessary. As the errors of the emission factors quanti"ed so far are caused mainly by the method of modal analysis (Hassel et al., 1994) the methodology of deriving emission factors from dynamometer testings should be improved. Experimental determinations of dynamic #eet compositions should be carried out on di!erent road categories (e.g. by license plate number evaluations). The uncertainties of emission data with high temporal resolution (hourly values) depend very strongly on the daytime and strongly on the weekday and the road category. Variation coe$cients amount from 10 to 70% for light-duty vehicles and from 15 to 100% for heavyduty vehicles. Low error ranges occur during the day
4609
hours with high tra$c volume, and higher error ranges occur during the night hours with low tra$c volume. To avoid these peaks of uncertainty, additional tra$c volume data related to the individual road sections in a high temporal resolution are required.
References Andrias, A., Samaras, Z., Zachariadis, T., Zierock, K.H., 1993. Assessment of random and systematic errors associated with the Corinair/Copert methodology for estimating VOCs emitted from road tra$c. In: Proceedings of the TNO/EURASAP Workshop on the reliability of VOC emission data bases, 9*10 June, 1993, Delft, the Netherlands. IMW-TNO Publication, P 93/040, pp. 75}87. Eggleston, H.S., 1993. Uncertainties in the estimates of emissions of VOCs from motor cars. In: Proceedings of the TNO/ EURASAP Workshop on the reliability of VOC emission data bases, 9}10 June, 1993, Delft, the Netherlands. IMWTNO Publication, P 93/040, pp. 59}73. Hartung, J., Elpelt, B., KloK sener, K.-H., 1995. Statistik}Lehrund Handbuch der angewandten Statistik. 10. Au#age. R. Oldenbourg Verlag, MuK nchen. Hassel, D., Jost, P., Weber, F.-J., Dursbeck, F., Sonnborn, K.-S., Plettau, D., 1994. Das Abgas-Emissionsverhalten von Personenkraftwagen in der Bundesrepublik Deutschland im Bezugsjahr 1990. Technischer UG berwachungsverein Rheinland. Berichte Umweltbundesamt 8/94. Erich Schmidt Verlag, Bonn. Hassel, D., Sonnborn, K.-S., Weber, F.-J., 1998a. Vertrauensbereich der Abgas-Emissionsfak-toren von PKW in der Bundesrepublik Deutschland. TUG V Rheinland Sicherheit und Umweltschutz GmbH. FoK rderschwerpunkt TroposphaK renforschung (TFS) des BMBF, Leitthema 2, Projekt A.2. Hassel, D., Sonnborn, K.-S., Weber, F.-J., 1998b. Vertrauensbereich der Abgas-Emissionsfak-toren von schweren Nutzfahrzeugen in der Bundesrepublik Deutschland. TUG V Rheinland Sicherheit und Umweltschutz GmbH. FoK rderschwerpunkt TroposphaK renforschung (TFS) des BMBF, Leitthema 2, Projekt A.2. John, C., 1999. Emissionen von Luftverunreinigungen aus dem Stra{enverkehr in hoher raK umlicher und zeitlicher AufloK sung. Untersuchung von Emissionsszenarien am Beispiel Baden-WuK rttembergs. Dissertation, Institute for Energy Economics and the Rational Use of Energy (IER), University of Stuttgart, Forschungsberichte Band 58. Kraftfahrt-Bundesamt, 1994. Statistische Mitteilungen. Reihe 2: Kraftfahrzeuge, Sonderheft 2: Bestand an Kraftfahrzeugen u. KraftfahrzeuganhaK ngern am 1. Juli 1994 nach Zulassungsbezirken in Deutschland. Metzler-Poeschel-Verlag, Stuttgart. KuK hlwein, J., John, C., Friedrich, R., Obermeier, A., 1999. AbschaK tzung und Bewertung der Unsicherheiten hochaufgeloK ster NO - und NMVOC-Emissionsdaten. IER, University of V Stuttgart. Commissioned by PEF } Projekt EuropaK isches Forschungszentrum fuK r Ma{nahmen zur Luftreinhaltung, Karlsruhe, Berichtsreihe FZKA-BWPLUS 3, Report No. 30, Juli, http://bwplus.fzk.de/berichte/SBer/PEF296002SBer.pdf.
4610
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
MEET, 1999. Meet } Methodology for calculating transport emissions and energy consumption. European Commission. Transport Research. Fourth framework programme. Strategic research, DG VII - 99. Schmitz, S., Haserich, D., Oppermann, F., Otto, I., PuK tz, T., Siedho!, M., Viktorin, D., 1997. Entwicklung eines planungsrelevanten Emissionskatasters Stra{enverkehr. Bundesforschungsanstalt fuK r Landeskunde und Raumordnung, Bonn. Materialien zur Raumentwicklung, Heft 80.
Steven, H., 1995. Emissionsfaktoren fuK r verschiedene Fahrzeugschichten, Stra{enkategorien und VerkehrszustaK nde (1). 2. Zwischenbericht zum Forschungsvorhaben 105 06 044: Erarbeitung von Grundlagen fuK r die Umsetzung von A 40.2 BImSchG. Commissioned by Umweltbundesamt, Berlin (2nd Edition). Umweltbundesamt, 1999. Handbuch fuK r Emissionsfaktoren des Stra{enverkehrs, Version 1.2. Umweltbundesamt Berlin, Bundesamt fuK r Umwelt, Wald und Landschaft Bern, Infras AG Bern (published as software on CD-ROM).
Uncertainties of modelling emissions from road transport J. KuK hlwein*, R. Friedrich Institute of Energy Economics and the Rational Use of Energy (IER), University of Stuttgart, He}bru( hlstr. 49a, D-70565 Stuttgart, Germany Received 3 September 1999; accepted 10 May 2000
Abstract To determine emission data from road transport, complex methods and models are applied. Emission data are characterized by a huge variety of source types as well as a high resolution of the spatial allocation and temporal variation. So far, the uncertainties of such calculated emission data have been largely unknown. As emission data is used to aid policy decisions, the accuracy of the data should be known. So, in the following, the determination of uncertainties of emission data is described. Using the IER emission model for generating regional or national emission data, the uncertainties of model input data and the total errors on di!erent aggregation levels are exemplarily investigated for the pollutants NO and NMHC in 1994 for the area of West Germany. The results of statistical error analysis carried out for V annual emissions on road sections show variation coe$cients (68.3% con"dence interval) of 15}25%. In addition, systematic errors of common input data sets have been identi"ed especially a!ecting emissions on motorway sections. The statistical errors of urban emissions with warm engine on town level amount to 35%. Therefore they are considerably higher than the errors outside towns. Error ranges of additional cold start emissions determined so far have been found in the same order. Additional uncertainties of temporally highly resolved (hourly) emission data depend strongly on the daytime, the weekday and the road category. Variation coe$cients have been determined in the range between 10 and 70% for light-duty vehicles and between 15 and 100% for heavy-duty vehicles. All total errors determined here have to be regarded as lower limits of the real total errors. 2000 Elsevier Science Ltd. All rights reserved. Keywords: Road tra$c; Statistical error propagation; Systematic error analysis; Temporally highly resolved emission data; Spatial intersection
1. Introduction Emission data from road transport are used as input data for modelling atmospheric processes of transport, transformation and deposition. Therefore, they are essential for numeric calculations of concentrations and depositions of primary and secondary pollutants and origin for developments of clean air strategies. Up to now there has been only little knowledge about the reliability of such calculated emission data. It was not state of the art to state uncertainty ranges together with the emission data; instead only qualitative information about uncertainties (e.g. as ranking) was given.
* Corresponding author. E-mail address: [email protected] (J. KuK hlwein).
Past estimates of uncertainties of European VOC-emissions from motor cars have been carried out e.g. by (Eggleston, 1993) and (Andrias et al., 1993) for less detailed emission calculation methods and with older emission factors than used here. The analyses described here look with extended detail into the statistical deviations of the input data used for calculating emissions. To estimate uncertainties of modelled emissions, it is necessary to determine the statistical error ranges of input data using suitable methods and to combine them to a total error by sensitivity studies and statistical error calculations. The results presented in the following are related to the high-resolution emission factors and the associated tra$c data which are available for the area of West Germany. The investigations are restricted to the pollutants non-methane hydrocarbons (NMHC) and NO (NO and NO ) as they take main V responsibility for the creation of atmospheric oxidants.
1352-2310/00/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved. PII: S 1 3 5 2 - 2 3 1 0 ( 0 0 ) 0 0 3 0 2 - 2
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
4604
The underlying emission calculation model was developed at the IER, university of Stuttgart (John, 1999), and is similar in structure to the emission models used in European scale (MEET, 1999), but more detailed. The model is based on current parameters for regional emission modelling with high temporal and spatial resolution. Emissions are calculated for all segments of non urban roads and urban areas. The basis of the method used here is the `UBA-Handbooka (Umweltbundesamt, 1999), that provides detailed emission factors as function of vehicle layers (de"ned by vehicle category, type of engine, age, pollution reduction technology, exhaust gas regulation and class of cubic capacity resp. vehicle mass), tra$c situations (driving patterns described by distributions of driving conditions) and road gradients. The average daily tra$c volumes of seven di!erent vehicle categories are available for nearly all individual non-urban main road sections in Germany from tra$c counts that take place in a "ve year rotation. The appropriate driving patterns to be used for a road segment are estimated from information on road category, number of lanes, road gradient and mean tra$c volume. Information about individual road sections are available from o$cial road databases of some German states. The temporal resolution of the tra$c volume data is derived from a detailed analysis of data from automatic permanent counting stations that continuously measure the number of heavy- and light-duty vehicles on selected locations in Germany. If no speci"c tra$c volume data for an individual road section is available, mean values from comparable road sections are transferred. Further di!erentiations of tra$c volume data from vehicle categories to vehicle layers are achieved by transferring mean results of license plate number evaluations (Umweltbundesamt, 1999) that are available for three di!erent road types (motorways, rural and urban driving). A simpli"ed functional relation of input parameters to calculate hourly emissions of one selected road section (outside towns) is given by
E"N¸ IT MDT T T T
dp (dp EF ) Z , (1) T N T J J N E F P N J where E is the calculated hourly emission, N the number of days in 1994 ("365), L the length of road section, IT the interpolation factor for calculation of tra$c T volume in 1994 from counts carried out in 1990 and 1995, di!erentiated by vehicle category v, MDT the T mean daily tra$c volume between 1990 and 1995, di!erentiated by vehicle category v, dp the share of driving T N performance of one driving pattern p in the total driving performance of a vehicle category v (driving pattern mix), dp ¸ the share of driving performance of one vehicle T layer l in the total driving performance of a vehicle
category v (#eet composition), EF the emission factor, J N E di!erentiated by vehicle layer l, driving pattern p and road gradient class g, and Z the fraction of annual F P emission emitted in hour h for road category r. With the results of uncertainty calculations, it is possible to "lter out those input data whose errors contribute most to the error of total emissions. Therefore, it is possible to "nd out the weak points of the used emission model and to do e$cient model improvements in future. Here the term `uncertaintya is used as a general expression of unknown possible deviations of true emissions from calculated emission data. If uncertainties can be quanti"ed the term `errora is used. We have to distinguish between statistical and systematic errors. Statistical errors may be expressed e.g. as standard deviations. Quantifying systematic errors is a much more di$cult question than calculating statistical errors, because independent data sets are necessary to do that. Such additional data sets are not available in each case. The authors found some systematic errors in commonly used input data sets and present the results below. These results can be used to correct modelled emission data from road transport based on these input data sets. Only a few selected results of the many error analyses carried out by the authors can be presented here. A full account is given in (KuK hlwein et al., 1999).
2. Statistical methods for error estimation The quanti"cation of extensive emissions from road transport (or even emissions from single road sections) by measurements requires high e!orts. So, usually emissions are determined by model calculations based on results from mediating observations. The emission E of a pollutant can be expressed as a function of input parameters x , x , 2, x . These input parameters are emission fac L tors, tra$c parameters like tra$c volume, vehicle #eet composition, driving pattern mix etc. and road speci"c parameters like speed limit, number of tra$c lanes, course of road, gradient etc. (see Eq. (1)). The spreads of statistical errors when measuring simple physical quantities can be calculated by basic statistical methods such as standard deviation (standard error of single values) or as standard deviation of the arithmetic mean of sample, if the single measuring values of the random sampling are known. By variation of a single input parameter x in the G calculation model, the sensitivity *E/*x can be deterG mined on the basis of partial derivation of the emission E with respect to the chosen input parameter. The absolute error of emission *E related to the error of the input G parameter x results from the multiplication of the sensiG tivity *E/*x with the calculated or estimated absolute G error dx or *x . G G
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
With the help of the error propagation law on conditions that E the error estimations of the input parameters are correct, E the input parameters are statistically independent from each other and E there is a high degree of linearity between E and x G (*E/*x + const.) or the errors of the input parameters G *x are small compared with x , G it is possible to calculate the mean total error *E of the emission E for any functional relations between the input parameters x according to (Hartung et al., 1995) G
L *E L *x " (*E ) (2) G G *x G G G where n is the number of input parameters, *E/*x the G sensitivity of calculated total emission related to the input parameter x , *x the error of input parameter x G G G (e.g. standard deviation) and *E the single error of G calculated total emission E related to *x G In case of emission modelling the calculation of error ranges of the input parameters in the form of standard deviations and the determination of systematic errors is not always possible. The main reasons are, that: *E"
E the error of the input parameter (and partially the parameter itself) is not quanti"able because of its complex physical structure and E for parts of the data base neither the single measured values nor the standard deviations are known. In these cases plausible assumptions about the statistical error ranges of the input parameter have to be made, and the e!ects of these errors on the emission have to be estimated by model runs. The following results of statistical calculations are given as standard deviations (resp. variation coe$cients). This means that the estimated error ranges describe a 68.3% con"dence interval. In case of emission modelling usually no higher degree of con"dence (e.g. 95%) is reported because of the high general level of errors of calculated data.
3. Results 3.1. Error ranges for emissions from diwerent road categories Error estimations of calculated annual emissions have been carried out for di!erent road categories according to the method described. In addition to the statistical errors, systematic errors have been quanti"ed and documented separately, as far as data for the estimation
4605
of these errors were available. The error calculations have been carried out for the year 1994 and for the area of West Germany. The statistical error ranges of the "ve main input parameters have been quanti"ed as follows: Mean gradients for individual road sections are available in gradient classes of 1% steps from the road databases, resulting in a maximum error of $0.5%. Errors in tra$c volume data are caused by count errors and by errors due to extrapolation from count samples to annual values. They have been estimated on a random basis by comparisons of the tra$c volume data used for modelling with annual tra$c volume data from automatic permanent counting stations. Usually, information about the driving pattern mixes of individual road sections is missing, because sectionspeci"c data about speed limits, road courses, etc., are not available. Plausible assumptions have been made to estimate possible ranges of driving patterns for each road category. Fleet compositions vary from one region to another. The standard errors and the e!ects of these spatial variations on the emission rates have been quanti"ed by evaluating vehicle stock data of di!erent administrative districts in West Germany. Statistical errors of commonly used emission factors have been investigated by the TUG V Rheinland (Hassel et al., 1998a, b). Standard errors from the single values of extensive dynamometer measurements have been calculated per vehicle layer. These highly resolved interim results have been aggregated to standard errors of the most important driving patterns. 3.1.1. Emissions from vehicles with warm engine outside towns The results of the error estimations for the road categories motorway and federal road are presented in Fig. 1. The results are related to road sections on the following conditions: E gradient class: 0}1%, E full extent counts at the o$cial German federal tra$c census in 1990 and 1995 (8 counting days, 36 counting hours), E estimated (not locally recorded) section-speci"c route situation (speed limit, bends, etc.), E estimated (not measured) section-speci"c tra$c data (driving pattern mix and #eet composition). The determined statistical errors (variation coe$cients) of calculated annual emissions have been found in the range of 21}26% for NMHC and 16}22% for NO V for both road categories. Especially, on motorway sections a negative error (emissions are calculated too high) is more probable than a positive one because of asymmetrical e!ects of the errors of the driving pattern mix.
4606
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
Fig. 1. Error ranges of annual emissions on road sections in West Germany, 1994: syst. } systematic error (%); stat. } statistical error: variation coe$cient (%).
The emission factors represent the most important source of error especially for NMHC. Since the frequency distribution of the driving pattern mix is not normal distributed, the statistical error due to driving pattern mix is not symmetrical. In addition to the statistical errors, two important sources of systematic errors (#eet composition and emission factors) have been found. These are described in the following: The term `#eet compositiona describes the further di!erentiation of tra$c volume which is already di!erentiated to vehicle categories by manual tra$c counts. The tra$c volume is now further distributed among vehicle layers which are de"ned by the criterions: type of engine, exhaust gas puri"cation system, legislative exhaust gas limit, age, cubic capacity and vehicle mass. These highly resolved tra$c volume data are available e.g. from the `UBA-Handbooka (Umweltbundesamt, 1999) for the three road categories motorways, rural and urban, related to the German situation. Comparing vehicle stock data (Kraftfahrt-Bundesamt, 1994) that are projected to the dynamic #eet composition by concept- and sizerelated projection factors (Steven, 1995), with the data from the `UBA-Handbooka shows, that shares of vehicles without or with older exhaust gas puri"cation systems are much higher in the projected stock data.
Assuming that our data about #eet compositions are nearer to reality than the commonly used UBA-Handbook data (that could not be checked as the derivation is not published), we conclude, that the use of the UBA data leads to calculated emissions, that are systematically too low. Systematic deviations of emission factors can be partially deduced by interpreting investigations made by TUG V Rheinland (Hassel et al., 1994). Emission factors for di!erent driving cycles are usually constructed by "rst dividing the test driving cycle and the corresponding emissions into parts (modal analysis of e.g. 1 s each) that are de"ned by present speed (v) and acceleration (a) of the tested vehicle. The measured emissions in high temporal resolution are classi"ed by a v}a-grid. To calculate the integral emission factor of any driving pattern, it is necessary to add up the mean emission values of the di!erent grid cells by the speci"c frequency distribution of driving conditions. These added up emission factors are published in large databases (Umweltbundesamt, 1999) and are used for emission modelling. Some of these modal measured emission factors have been exemplarily compared to integral values (bag analysis), measured over the whole driving cycle. Occuring deviations are caused by systematic errors of the method used for determining emission factors.
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
The resulting systematic errors are conspicuous. Whereas these two error groups have di!erent signs at the federal road sections, which means that they compensate each other, at motorway sections the same signs lead to systematic deviations of about #38% (NMHC) and #22% (NO ). This means that there is a systematic V underestimation of emissions on motorways when using the usual input data sets. 3.1.2. Emissions from vehicles with warm engine inside towns In contrast to non-urban tra$c, no detailed standardized tra$c volume data are available for individual road sections inside towns. (Of course, in many towns tra$c counts have been made } but these are not standardized on a regional or national level, so it takes too much e!ort to use them.) Total tra$c volumes on town level are available for some towns only. So additional infrastructure data (e.g. number of inhabitants) have to be included in the emission model. With it, total urban tra$c volumes on federal or regional level are distributed to the individual towns. As shown in Fig. 2, the statistical total errors have been determined to be approx. 37% (NMHC) and 35% (NO ). V Thus, they are considerably higher than the errors of
4607
road sections outside towns. There is an additional systematic error of about !11% for NMHC and !8% for NO . The two input parameters contributing most to the V total error are the tra$c volume and the emission factors. The step of distributing total tra$c volumes to town level using the number of inhabitants per town involve a variation coe$cient of about 23% (Schmitz et al., 1997). Compared to this, the determination of the total tra$c volume inside towns for all motor vehicles (e.g. for the old West German states) is quite reliable (variation coe$cient: approx. 11%) and the errors of di!erentiation of the tra$c volume to di!erent vehicle categories are relatively small (3.8% in case of NMHC and 11.9% in case of NO ). The uncertainties of emission factors for urban V transport are slightly higher than those for non-urban road sections. 3.1.3. Cold start emissions A calculation model, which contains mean distributions for the driving pattern mix, the #eet composition, parking times before start, driving distances, outdoor temperatures and the daily course of starting frequencies (Umweltbundesamt, 1999), lays the foundations for the error estimations of additional cold start emissions. The number of starts per town are deduced from the tra$c volume and the population data. The errors of additional cold start emissions are mainly marked by the spatial di!erentiation of the number of starts on town level and the cold start emission factors. Error contributions of more than 10% have been determined for the number of drives related to tra$c volume and the annual distribution of temperatures (only NMHC). Similar to the tra$c volume in case of determining emissions with warm engine inside towns, necessary information about the number of starts is not available on town level. So the total number of starting processes is determined from the total driving performance of passenger cars (in West Germany) and a mean driving distance. Then the total number of starts is distributed to the individual towns according to the numbers of inhabitants. The statistical uncertainties of the spatial distribution of starts correspond with those of the spatial distribution of tra$c volume. They amount to 23% with a slight asymmetry (Fig. 3). The statistical total errors up to now amount to about 39% in case of NMHC and 33% in case of NO . SystemV atic deviations have been determined from the available data as about !7% for NMHC and !19% for NO . V 3.2. Uncertainties of temporal resolution of emissions
Fig. 2. Error ranges of annual emissions with warm engine inside towns in West Germany, 1994: syst. } systematic error (%); stat. } statistical error: variation coe$cient (%).
In order to model temporally highly resolved emission data, mean annual, weekly and daily courses per road category are applied. Temporally highly resolved tra$c volume data (hourly intervals) are available from tra$c counts by automatic permanent counting stations.
4608
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
Comparing hourly shares in annual tra$c volumes at de"nite hours, deviation ranges between di!erent road sections of the same road category can be derived. The deviations of variation coe$cients caused by the daily courses can be roughly separated into two "elds: (1) late morning till evening, and (2) night till early morning. Besides the deviations caused by the daily courses there are signi"cant di!erences between the weekdays (Monday}Friday, Saturday, Sunday). Deviations derived from the data available from automatic permanent counting stations in West Germany in 1994 for the road categories motorway, federal road and state road have been summarized and are shown in Table 1. In general, the variation coe$cient increases with decreasing tra$c volume. So an increase with decreasing order of road category (motorwayPstate road) can be seen. The deviations on weekends are higher than those on working days. An exception was found for LDV during the night hours where variation coe$cients stay unchanged, respectively, decrease slightly on weekends. For HDV in all examined cases higher deviation ranges occur compared to LDV.
4. Conclusions and outlook
Fig. 3. Error ranges of annual additional cold start emissions from passenger cars in West Germany, 1994: syst. } systematic error (%); stat. } statistical error: variation coe$cient (%).
The results of statistical error analyses carried out for annual emissions in 1994 on road sections show variation coe$cients (68.3% con"dence interval) between 21 and 26% in case of NMHC and between 16 and 22% in case of NO . In addition systematic errors of commonly used V
Table 1 Error ranges of hourly di!erentiation of annual tra$c volumes for di!erent road categories (outside towns), weekdays and daytime Road category
Motorway
LDV/HDV
LDV HDV
Federal road
LDV HDV
State road
LDV HDV
Day/night
Day Night Day Night Day Night Day Night Day Night Day Night
Light-duty vehicles (LDV); heavy-duty vehicles (HDV). Day: late morning till evening; night: night till early morning.
Variation coe$cient (%) Monday}Friday
Saturday
Sunday
10}15 20}45 20}30 30}45 10}15 25}50 15}25 35}70 10}25 25}60 30}40 50}95
10}20 15}25 30}40 35}55 10}20 20}35 25}40 45}75 10}35 30}45 35}65 60}100
20}30 20}30 30}70 45}70 20}30 25}50 45}60 55}90 30}50 20}45 65}90 70}100
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
input data sets have been identi"ed especially a!ecting emissions on motorway sections with#38% for NMHC and#22% for NO . The statistical uncertainties of anV nual emissions with warm engine inside towns on town level amount to 37% (NMHC) and 35% (NO ); they are V much higher than the errors outside towns. Error ranges of additional cold start emissions determined so far are in the same order of magnitude. For two reasons all determined total errors have to be regarded as lower limit of the real total errors : (1) Because of the incomplete data situation the uncertainties of some input parameters could be quanti"ed only partially. There are considerable data gaps especially for emission factors (vehicles with newer technology, cold start, road gradients, heavy duty vehicles, motorcycles, running losses), the driving patterns on ordinary roads and inside towns and the tra$c volume data. Additional investigations are necessary to close these gaps. (2) The mean total errors have been determined according to the method of error propagation on condition that the input parameters are statistically independent from each other. Taking into account possible correlations between the input parameters, the true total error will be found between the mean total error (Eq. (2)) and the maximum total error that results from adding up the amounts of all single errors *E . In most cases it is not possible to calcuG late the true total statistical errors taking into account possible correlations because currently available data is not su$cient to do essential quanti"cations of covariances. However, it is the authors' impression, that the true errors in general are closer to the mean total errors than to the maximum total errors, because of the small in#uences of possible correlations. Signi"cant systematic errors could be recognized for the input parameters emission factors and #eet composition by including additional data sets. To clarify these systematic deviations further dynamometric measurements are necessary. As the errors of the emission factors quanti"ed so far are caused mainly by the method of modal analysis (Hassel et al., 1994) the methodology of deriving emission factors from dynamometer testings should be improved. Experimental determinations of dynamic #eet compositions should be carried out on di!erent road categories (e.g. by license plate number evaluations). The uncertainties of emission data with high temporal resolution (hourly values) depend very strongly on the daytime and strongly on the weekday and the road category. Variation coe$cients amount from 10 to 70% for light-duty vehicles and from 15 to 100% for heavyduty vehicles. Low error ranges occur during the day
4609
hours with high tra$c volume, and higher error ranges occur during the night hours with low tra$c volume. To avoid these peaks of uncertainty, additional tra$c volume data related to the individual road sections in a high temporal resolution are required.
References Andrias, A., Samaras, Z., Zachariadis, T., Zierock, K.H., 1993. Assessment of random and systematic errors associated with the Corinair/Copert methodology for estimating VOCs emitted from road tra$c. In: Proceedings of the TNO/EURASAP Workshop on the reliability of VOC emission data bases, 9*10 June, 1993, Delft, the Netherlands. IMW-TNO Publication, P 93/040, pp. 75}87. Eggleston, H.S., 1993. Uncertainties in the estimates of emissions of VOCs from motor cars. In: Proceedings of the TNO/ EURASAP Workshop on the reliability of VOC emission data bases, 9}10 June, 1993, Delft, the Netherlands. IMWTNO Publication, P 93/040, pp. 59}73. Hartung, J., Elpelt, B., KloK sener, K.-H., 1995. Statistik}Lehrund Handbuch der angewandten Statistik. 10. Au#age. R. Oldenbourg Verlag, MuK nchen. Hassel, D., Jost, P., Weber, F.-J., Dursbeck, F., Sonnborn, K.-S., Plettau, D., 1994. Das Abgas-Emissionsverhalten von Personenkraftwagen in der Bundesrepublik Deutschland im Bezugsjahr 1990. Technischer UG berwachungsverein Rheinland. Berichte Umweltbundesamt 8/94. Erich Schmidt Verlag, Bonn. Hassel, D., Sonnborn, K.-S., Weber, F.-J., 1998a. Vertrauensbereich der Abgas-Emissionsfak-toren von PKW in der Bundesrepublik Deutschland. TUG V Rheinland Sicherheit und Umweltschutz GmbH. FoK rderschwerpunkt TroposphaK renforschung (TFS) des BMBF, Leitthema 2, Projekt A.2. Hassel, D., Sonnborn, K.-S., Weber, F.-J., 1998b. Vertrauensbereich der Abgas-Emissionsfak-toren von schweren Nutzfahrzeugen in der Bundesrepublik Deutschland. TUG V Rheinland Sicherheit und Umweltschutz GmbH. FoK rderschwerpunkt TroposphaK renforschung (TFS) des BMBF, Leitthema 2, Projekt A.2. John, C., 1999. Emissionen von Luftverunreinigungen aus dem Stra{enverkehr in hoher raK umlicher und zeitlicher AufloK sung. Untersuchung von Emissionsszenarien am Beispiel Baden-WuK rttembergs. Dissertation, Institute for Energy Economics and the Rational Use of Energy (IER), University of Stuttgart, Forschungsberichte Band 58. Kraftfahrt-Bundesamt, 1994. Statistische Mitteilungen. Reihe 2: Kraftfahrzeuge, Sonderheft 2: Bestand an Kraftfahrzeugen u. KraftfahrzeuganhaK ngern am 1. Juli 1994 nach Zulassungsbezirken in Deutschland. Metzler-Poeschel-Verlag, Stuttgart. KuK hlwein, J., John, C., Friedrich, R., Obermeier, A., 1999. AbschaK tzung und Bewertung der Unsicherheiten hochaufgeloK ster NO - und NMVOC-Emissionsdaten. IER, University of V Stuttgart. Commissioned by PEF } Projekt EuropaK isches Forschungszentrum fuK r Ma{nahmen zur Luftreinhaltung, Karlsruhe, Berichtsreihe FZKA-BWPLUS 3, Report No. 30, Juli, http://bwplus.fzk.de/berichte/SBer/PEF296002SBer.pdf.
4610
J. Ku( hlwein, R. Friedrich / Atmospheric Environment 34 (2000) 4603}4610
MEET, 1999. Meet } Methodology for calculating transport emissions and energy consumption. European Commission. Transport Research. Fourth framework programme. Strategic research, DG VII - 99. Schmitz, S., Haserich, D., Oppermann, F., Otto, I., PuK tz, T., Siedho!, M., Viktorin, D., 1997. Entwicklung eines planungsrelevanten Emissionskatasters Stra{enverkehr. Bundesforschungsanstalt fuK r Landeskunde und Raumordnung, Bonn. Materialien zur Raumentwicklung, Heft 80.
Steven, H., 1995. Emissionsfaktoren fuK r verschiedene Fahrzeugschichten, Stra{enkategorien und VerkehrszustaK nde (1). 2. Zwischenbericht zum Forschungsvorhaben 105 06 044: Erarbeitung von Grundlagen fuK r die Umsetzung von A 40.2 BImSchG. Commissioned by Umweltbundesamt, Berlin (2nd Edition). Umweltbundesamt, 1999. Handbuch fuK r Emissionsfaktoren des Stra{enverkehrs, Version 1.2. Umweltbundesamt Berlin, Bundesamt fuK r Umwelt, Wald und Landschaft Bern, Infras AG Bern (published as software on CD-ROM).