Content uploaded by Michael Clairotte
Author content
All content in this area was uploaded by Michael Clairotte on Apr 12, 2017
Content may be subject to copyright.
1
Martin Weiss, Elena Paffumi, Michaël Clairotte,
Yannis Drossinos, Theodoros Vlachos,
Pierre Bonnel, Barouch Giechaskiel
2017
An assessment of cold-start
frequencies and emission
effects
Including cold-start emissions in the
Real-Driving Emissions (RDE) test
procedure
EUR 28472 EN
This publication is a Science for Policy report by the Joint Research Centre (JRC), the European Commission’s
science and knowledge service. It aims to provide evidence-based scientific support to the European policy-
making process. The scientific output expressed does not imply a policy position of the European Commission.
Neither the European Commission nor any person acting on behalf of the Commission is responsible for the use
which might be made of this publication.
JRC Science Hub
https://ec.europa.eu/jrc
JRC 105595
EUR 28472 EN
PDF
ISBN 978-92-79-62229-8
ISSN 1831-9424
doi:10.2760/70237
Luxembourg: Publications Office of the European Union, 2017
© European Union, 2017
Reproduction is authorised provided the source is acknowledged.
All images © European Union 2017.
Title: Including cold-start emissions in the Real-Driving Emissions (RDE) test procedure
Abstract
We document two independent analyses that were conducted to support the inclusion of cold-start emissions in
the Real-Driving Emissions (RDE) test procedure. First, we present the results of a scoping review on cold-start
frequencies and trip distances in Europe. Second, we present a scenario analysis that aims to quantify the impact
of modifications in the RDE data pre-processing and evaluation on the calculated NOX emissions over the urban
part of an on-road test. We find that some 27 ± 5% of trips in Europe may contain a cold start. The driving
distance between two consecutive cold starts reaches 36 ± 16 km (mean) and 30 ± 13 km (median),
respectively. Our scenario analysis suggests that a simple inclusion of cold start into the regulatory RDE data
evaluation procedure may not capture cold-start NOX emissions in a robust manner. However, combining
modifications of the RDE data pre-processing and the RDE data evaluation can capture at least part of the
incremental cold-start NOX emissions. A more systematic assessment of European driving data an d an expansion
of the scenario analysis presented here could substantiate the findings of this report.
ii
Content
List of Tables ........................................................................................................ ii
List of figures ....................................................................................................... iii
List of abbreviations, acronyms, and units ............................................................... iv
Acknowledgements ............................................................................................... vi
Executive summary ............................................................................................. vii
1 Introduction ..................................................................................................... 1
2 Background ...................................................................................................... 2
3 Methods ........................................................................................................... 7
3.1 Analysis of driving distance between two consecutive cold starts ....................... 7
3.2 Scenario analysis - cold-start inclusions ......................................................... 7
4 Results .......................................................................................................... 11
4.1 Driving distance between two consecutive cold starts .................................... 11
4.1.1 GPS car data: Driving patterns in Modena and Florence (Italy) .............. 11
4.1.2 Green eMotion data ......................................................................... 14
4.1.3 Driving data from the Netherlands ..................................................... 16
4.1.4 Driving data from Sweden ................................................................ 17
4.1.5 WLTP data base and miscellaneous data sources ................................. 19
4.1.6 Handbook Emission Factors for Road Transport (HBEFA 3.2) ................. 21
4.1.7 CADC and related trip analyses ........................................................... 23
4.2 Scenario analysis – cold-start inclusion ........................................................ 25
4.2.1 Baseline scenario and simple RDE cold-start inclusion .......................... 25
4.2.2 Modifications of RDE data pre-processing ........................................... 26
4.2.3 Modifying the weighting of moving averaging windows ......................... 28
4.2.3 Combining modifications in the pre-processing and evaluation
of NOX emissions ............................................................................. 29
4.2.4 Summary of the scenario analysis ..................................................... 31
5 Discussion and conclusions ............................................................................... 34
5.1 General aspects ........................................................................................ 34
5.2.1 Conclusions on the driving distance between two cold-starts ................. 35
5.2.2 Conclusions on the scenario analysis – cold-start inclusions .................. 37
6 References ..................................................................................................... 38
ii
List of Tables
Table 1: Principal options for adaptating RDE provisions to ensure a robust
coverage of cold-start emissions ............................................................ 5
Table 2: On-road tests used for the scenario analysis ........................................... 10
Table 3: Distribution of parking durations in the province of Modena ...................... 13
Table 4: Preliminary review of European trip distances based on
miscellaneous sources .......................................................................... 20
Table 5: Frequency distribution of trip distances in 6 European countries ................ 22
Table 6: Frequency distribution of parking time in 6 European countries ................. 23
Table 7: Percentage of driving distance covered by trips that start with a
cold or partially warmed up engine ........................................................ 24
Table 8: Distance of trips containing cold start .................................................... 24
Table 9: Coefficients of the linear regression model explaining MAWP as a
function of CSPI .................................................................................. 33
Table 10: Overview of results .............................................................................. 36
iii
List of figures
Figure 1: Province of Modena (Italy), second-by-second calculations ....................... 12
Figure 2: Frequency distribution of trip lengths ..................................................... 13
Figure 3: Generic cool-down curve for engine coolant ............................................ 14
Figure 4: Distribution of trip distances as obtained from the Green eMotion data ....... 15
Figure 5: Distribution of parking times as obtained from the Green eMotion data ...... 16
Figure 6: Frequency distribution of trip distances in Sweden ................................... 18
Figure 7: Frequency distribution of parking durations ............................................ 18
Figure 8: Distribution of urban trip distances in the EU based on miscellaneous
data sources ....................................................................................... 21
Figure 9: NOX emissions after the evaluation of the urban part of an RDE trip
with and without the inclusion of cold-start ............................................. 25
Figure 10: NOX emissions effect of modifying the RDE data pre-processing ................ 27
Figure 11: NOX emissions effect of modifying the weighting of moving averaging
windows ............................................................................................. 29
Figure 12: NOX emissions effect of adapting the pre-processing of cold-start
emissions and weighting of moving averaging windows ............................ 30
Figure 13: Relationship between MAWP and CSPI expressed through a simple linear
model ................................................................................................ 32
iv
List of abbreviations, acronyms, and units
# - Number
% - Percent
BMVI - Bundesministerium für Verkehr und digitale Infrastruktur
(Federal Ministry for Transport and Digital Infrastructure, Germany)
CADC - Common Artemis Driving Cycle
CBS - Statistics Netherlands (Centraal Bureau voor de Statistiek)
CSPI - Cold Start Performance Indicator [%]
dcold - actual driving distance during cold start [km]
durban - reference driving distance, representing the typical distance driven by
Europeans between two consecutive cold starts [km]
ECS - average NOX emissions during cold start [mg/km]
EUrban - average NOX emissions during urban driving with a warm engine
[mg/km]
EC - European Commission
e.g. - exempli gratia (example given)
EU - European Union
GPS - Global Positioning System
GSM - Global System for Mobile Communications
h - hour
HBEFA - HAndbuch Emissions FAktoren (HAndBook Emission FActors)
i.e. - id est (that is)
JRC - Joint Research Centre
km - kilometre
Mcold - distance specific pollutant emissions during cold start [mg/km]
Mhot,urban - distance specific pollutant emissions during warm-engine operation as
determined according to Appendices 5 or 6 of Regulation 2016/427
[mg/km]
Murban - final distance-specific pollutant emissions over the urban part of a RDE
trip [mg/km]
MAW - Moving Averaging Window
MAWP - Moving Averaging Window Performance [%]
mg - milligram
MS - EU Member States
NO2 - nitrogen dioxide
NOX - nitrogen oxides
v
NOXRDEMod0a - NOX emissions calculated with the baseline scenario Mod0a [mg/km]
NOXRDEModi - NOX emissions calculated with the respective scenario i
PEMS - Portable Emissions Measurement System
PN - Particle Number
RDE - Real-Driving Emissions
s - second
SCR - Selective Catalytic Reduction
w - weighting factor for emissions from cold-start versus warm-engine
operation
WLTC - Worldwide harmonized Light-duty vehicles Test Cycle
WLTP - Worldwide harmonized Light-duty vehicles Test Procedure
vi
Acknowledgements
We thank Lars-Henrik Björnsson, Sten Karlsson, Jörg Kühlwein, Norbert Ligterink, and Heinz
Steven for supporting our analysis of European driving data. We are grateful to Massimo
Carriero, Rinaldo Colombo, Fausto Forni, Marcos Otura Garcia, Gaston Lanappe, Philippe Le-
Lijour, François Montigny, Mirco Sculati, Germana Trentadue, and Theodoros Vlachos for
planning and executing vehicle tests at the Vehicle Emissions Laboratory (VeLA) of the JRC.
We thank Panagiota Dilara, Vicente Franco, and Zlatko Kregar for providing comments on
earlier drafts of this report.
vii
Executive summary
The European Union implemented in 2016 the first two packages of the Real-Driving
Emissions (RDE) test procedure as Regulations 2016/427 and 2016/646. The third
regulatory RDE package addressing cold-start emissions, the testing of hybrid vehicles, and
the measurement of particle number emissions was approved by a technical committee of
experts from Member States in December 2016. During the stakeholder consultations on
the third RDE package, several options for the inclusion of cold-start into the RDE test
procedure were discussed. Member States supported a simple inclusion of the cold-start
period into the normal RDE data evaluation. The Joint Research Centre, by contrast,
proposed a separate calculation of distance-specific pollutant emissions for (i) the cold-start
period and (ii) the warm-engine operation and a subsequent weighting of the resulting
emissions by a factor that accounts for the distance typically driven by vehicle users
between two consecutive cold starts.
The objective of this report is to document two independent analyses conducted by the JRC
in support of the RDE stakeholder consultations on cold start. First, we conduct a scoping
review of cold-start frequencies and trip distances in Europe. Second, we conduct a scenario
analysis to investigate the effect of modifications in the RDE data pre-processing and
evaluation procedures on the calculated NOX emissions for urban driving. This analysis is
based on actual on-road NOX emissions data obtained from vehicle tests conducted at the
JRC.
The reviewed driving data obtained from seven major sources suggest that some 27 ± 5%
of trips are driven in Europe after vehicle parking of at least 3 to 8 h and may thus contain a
cold start. Based on all collected data, the average driving distance between two
consecutive cold starts reaches 36 ± 16 km (mean) and 30 ± 13 km (median), respectively.
If only urban trips are considered, the distance between two consecutive cold starts reaches
25 ± 16 km (mean) and 27 ± 8 km (median), respectively. These findings are, in the strict
sense, valid for the relatively conservative assumption that cold starts occur after a
minimum parking duration of 3-8 h. Given the heterogeneity of temperature and driving
conditions across the European continent, cold-starts may occur in real-word driving after
longer/shorter parking durations, e.g., if the engines and after-treatment systems are
thermally encapsulated/if vehicle operation occurs at ambient temperatures lower than
15oC.
The inclusion of cold start into the normal RDE data evaluation for 7 on-road trips yields
both higher and lower overall urban NOX emissions, although all tested vehicles show higher
NOX emissions during cold start than during warm-engine operation. This observation
suggests that a simple inclusion of cold start into the normal RDE data evaluation may not
capture cold start NOX emissions in a robust manner. Modifications of (i) the data pre-
processing (duplication of cold-start phase or change in the order of emission events) and
(ii) the weighting of moving averaging windows (assuming a constant or linearly decreasing
weighting factor for windows containing cold-start emissions) each increases on average the
overall urban NOX emissions but may also result for individual tests in decreasing NOX
emissions relative to the exclusion of cold start from the RDE evaluation. However, a
combination of modifications in the RDE data pre-processing and evaluation might capture
at least part of the incremental cold-start emissions.
We regard our findings as robust. Yet, a more systematic collection and assessment of
driving data and an expansion of the scenario analysis covering more tests and vehicles
could complement the findings presented here.
1
1 Introduction
The European Union (EU) has implemented in spring 2016 the first two packages of the
Real-Driving Emissions (RDE) test procedure as Regulations 2016/427 and 2016/646 (EC,
2016a,b). RDE constitutes world-wide the first on-road test for the type approval of light-
duty vehicles. In an initial monitoring phase, vehicles are tested on the road with so-called
Portable Emissions Measurement Systems (PEMS) and their pollutant emissions have to be
documented. From September 2017 onwards, binding not-to-exceed emission limits apply.
In parallel to the implementation of the first two RDE packages, the European Commission
has started working on two additional RDE package that address cold-start emissions, the
testing of hybrid vehicles, the on-road measurement of particle number (PN) emissions, and
the periodical regenerations of after-treatment systems (third RDE package) and
administrative provisions for in-service conformity testing and market surveillance (fourth
RDE package).
Cold start can contribute substantially to the overall vehicle emissions, specifically in urban
areas, where trips are short, cold starts frequent, and air quality problems most severe
(EEA, 2015). Moreover, nitrogen oxides (NOx) emissions during cold start could become a
major contributor to the overall NOx emissions of diesel cars, once large parts of the diesel
fleet are equipped with selective catalytic reduction (SCR) and lean NOx-trap after-
treatment systems. The European Commission (EC) therefore intends to implement
dedicated cold-start provisions as part of the 3rd regulatory RDE package.
Several options for the inclusion of cold-start into the existing RDE Regulations 2016/427
and 2016/646 (EC, 2016a,b) have been discussed with Member States and other
stakeholders since January 2016 in the RDE working group. Several Member States have
supported a simple inclusion of cold-start into the normal RDE data evaluation. The Joint
Research Centre (JRC), by contrast, had proposed a separate calculation of the distance-
specific pollutant emissions for (i) the cold-start phase and (ii) the warm-engine operation
and a subsequent weighting of the resulting emissions by means of a factor that accounts
for the distance typically driven by vehicle users between two consecutive cold starts.
The objective of this report is to provide rationale for the discussions about the inclusion of
cold start into the RDE test procedure. To this end, we present two separate analyses. First,
we conduct a scoping review of available information on the typical trip distance and the
frequency of cold starts in Europe. The results of this analysis could help defining a factor
for the weighting of cold-start versus warm-engine emissions measured during the urban
part of a RDE test. Second, we conduct a scenario analysis that investigates the effect of
cold-start inclusions in and modifications of the RDE data evaluations on the calculated NOX
emissions. This analysis is based on selected vehicle tests and actual on-road NOX emissions
data. The results can help identifying elements in the data evaluation that could be
amended to allow for a robust coverage of cold-start emissions by the RDE test procedure.
The report continues with additional background information (Section 2) and a short
description of our research methods (Section 3). We present the results for the two
independent analyses in Section 4. The report ends with a discussion and conclusions for
policy makers in Section 5.
2
2 Background
Cold start in the context of RDE is defined by Regulation 2016/427 as the first 5 minutes
after the initial start of the combustion engine
1
. If the engine coolant temperature can be
reliably determined, the cold start period ends once the coolant has reached 343 K (70 °C)
for the first time but no later than 5 minutes after initial engine start (EC, 2016a). Cold start
emissions have to be recorded but are excluded from the emissions calculation until specific
requirements are defined. At present, there are no requirements for the preconditioning of
vehicles before an RDE test. However, the provisions of the recently approved third RDE
package foresee that vehicles are driven for at least 30 min, parked with doors and bonnet
closed and kept in engine-off status within moderate or extended altitude and temperatures
for between 6
2
and 56 hours. The vehicle soak prior to RDE testing will be in line with the
requirements for Type I testing in the laboratory that demands a minimum soak duration of
6 h. For the low-temperature Type IV test, the soak duration is set to 12 h (UNECE, 2015).
In January 2016, the European Commission has expressed the intention to cover cold-start
emissions by the RDE test procedure; dedicated provisions for inclusion into the 3rd
regulatory RDE package where then discussed throughout 2016 (EC, 2016c). The cold-start
provisions contain three independent elements: (i) the preconditioning of vehicles including
a precondition drive and vehicle soak, (ii) specific requirements for the driving conditions
during cold-start and (ii) the evaluation of cold-start emissions.
First, to ensure cold-start is indeed part of RDE testing, the third regulatory RDE package
foresees that vehicles will have to be preconditioned as described above at moderate or
extended altitude (up to 700 and 1300 m, respectively) and temperatures (minimum
temperature 0oC and -7oC, respectively
3
). Exposure to extreme atmospheric conditions
(e.g., heavy snowfall, storm, hail) and excessive amounts of dust during the parking period
should be avoided. If the vehicle was conditioned for the last three hours prior to the test at
an average temperature that falls within the extended temperature, the pollutant emissions
during cold start can be divided by a factor of 1.6, even if the running conditions are not
within the extended temperature range.
Second, to ensure unbiased driving the 3rd regulatory RDE package requires for the cold-
start period:
an average vehicle speed (including stops) of 15-40 km/h;
a maximum speed no more than 60 km/h;
idling after the first ignition of the combustion engine not to exceed 30 s;
1
Regulation 2016/427 does not contain provisions for vehicle conditioning. However, the 3rd regulatory RDE
package requires that vehicles are parked for at least 6 h before entering an RDE test. After such preconditioning,
the temperature of the engine oil and coolant as well as of the emissions after-treatment technologies resembles
that of the ambient air.
2
Vehicle conditioning of 6 h might not in all cases ensure a complete engine cooldown, especially for larger
engines or vehicles equipped with thermal engine encapsulation. Yet, a 6 h precondition period allows to
precondition and test vehicles in one working day and could force the adoption of advanced thermal encapsulation
that may yield real-world emission benefits.
3
Until 5 years after effectiveness of Regulation 715/2009 (EC, 2007), moderate temperature conditions are limited
to 3°C and extended temperature conditions to -2°C.
3
total durations of all stops not to exceed 90 s, i.e., 30% of the cold-start duration as
defined in Regulation 2016/427 (EC, 2016a).
Third, the evaluation of cold-start emissions had been subject to intensive discussions
among RDE stakeholders. Two options were analysed in greater detail: (i) the inclusion of
cold-start emissions into the normal RDE evaluation of the urban part of a trip as prescribed
in Appendices 5 and 6 of Regulation 2016/427 (EC, 2016a) and (ii) a separate assessment
of cold-start emissions and warm-engine emissions for the urban part of a trip followed by
weighting of the two results.
Option 1 is simple and can easily be implemented into the existing regulatory RDE text but
might not always assess cold-start emissions in a robust manner:
The composition of a RDE trip is evaluated based on the realized vehicle speed, that
is, Point 6 of Regulation 2016/427 and the RDE data evaluation defined in
Appendices 5 and 6 of the same regulation use vehicle speed to classify events a
urban, rural, or motorway driving. The speed-based classification leads to a situation
where parts of the trip that are actually driven in a rural environment or on the
motorway may be classified as urban driving if the instantaneous vehicle speeds or
the average speed of moving averaging windows does not exceed 60 km/h and 45
km/h, respectively (EC, 2016a). Depending on route design and traffic conditions,
RDE trips lasting some 90-120 min may thus cover longer urban distances than
those typically driven by vehicle users between two consecutive cold starts. This
observation is problematic as pollutant emissions during low-speed driving with fully
warmed-up engine and after-treatment systems tend to be substantially lower than
cold-start emissions.
The data weighting applied in Appendices 5 and 6 of Regulation 2016/427 can lead
to an under-representation of cold-start emissions. Appendix 5 weighs or excludes
pollutant emissions if these belong to moving averaging windows those average CO2
emissions deviate by more than 25% from the CO2 reference curve (established by
driving the vehicle on the chassis dynamometer with the WLTC). Appendix 6
categorizes three-second averages of pollutant emissions into power bins that are
weighed based on a factor derived from a pre-defined frequency distribution, which
gives relatively little weight to bins of high power. Related to the functioning of the
data evaluation in Appendices 5 and 6, cold-start tests could be defeated by
purposefully driving the cold start in an aggressive manner, which would lead to very
high CO2 emissions and the allocation of driving events into high power bins and, in
turn, to a de facto exclusion of cold start emission from the evaluation of urban
emissions.
Pertinent to the moving averaging window method of Appendix 5, data points at the
beginning and end of a test are contained in fewer averaging windows than data
points in the middle of a test. When averaging the emissions of windows, cold start
receives a disproportionately low weight in the overall data evaluation. Moreover,
Regulation 2016/427 does not specify the order in which moving averaging windows
have to be calculated. A window calculation starting from the end of a test entails the
risk that cold-start data are not covered by any window.
Pertinent to the power-binning method of Appendix 6 are challenges in evaluating
hybrid vehicles; as the wheel power cannot be reliably estimated from the CO2
emissions of such vehicles, the third regulatory RDE package specifies that Appendix
6 can only be applied to mild and full hybrid vehicles if the wheel power is measured,
e.g., by a wheel torque sensor.
4
Taken together, we see a clear risk that the simple inclusion of cold start into the normal
RDE data evaluation, without implementing complementary provisions, may (i) under-
represent cold-start emissions compared to their actual contribution to real-world driving
emissions and (ii) trigger biased driving during the cold-start period to simply defeat the
RDE test. In view of these shortcomings, the JRC has proposed as Option 2 to separately
assess cold-start emissions and the warm-engine emissions during urban driving and then
apply a weighting factor for the calculation of the final emissions results as follows:
where:
Murban - final distance-specific pollutant emissions over the urban part of a RDE
trip [mg/km]
Mcold - distance specific pollutant emissions during cold start [mg/km]
Mhot,urban - distance specific pollutant emissions during warm-engine operation as
determined according to Appendices 5 or 6 of Regulation 2016/427
[mg/km]
w - weighting factor
dcold - actual driving distance during cold start
durban - reference driving distance, representing the typical distance driven
between two consecutive cold starts in the EU
Option 2 would ensure that the weight of cold-start emissions in the final RDE result for
urban driving that is comparable to the weight of cold start in the overall driving of
European citizens. However, while Option 2 might capture cold-start emissions accurately,
its implementation would add complexity to the RDE regulation and its effectiveness would
hinge on the establishment of an appropriate weighting factor.
Several EU Member States have therefore expressed in the RDE meeting on 8 and 9
September 2016 their preference for Option 1, highlighting its simplicity and the
observation that to date, no data had been presented that point to specific challenges
related to cold-start emissions that are not yet covered by the existing Type 1 emissions
tests.
If Option 1 is implemented in the 3rd regulatory RDE package, three areas for intervention
could be considered, to make the evaluation of cold-start emissions more robust (Table 1):
Ensuring the cold-start share in the total urban distance driven in a RDE trip matches
the share of cold start in real-world driving. This objective could be achieved, e.g.,
by a map-based evaluation of the urban portion of a trip or by introducing a fixed
distance after test start that should be considered for the evaluation of urban
emissions.
5
Adapting the pre-processing of data in Appendix 4 of Regulation 2016/427 (EC,
2016a) but leaving the actual evaluation of emissions data according to Appendices 5
and 6 unchanged. Options include (i) re-arranging of events in the recorded data
stream, e.g., by cutting cold start from the beginning of a test and placing it into the
middle of urban driving or (ii) the duplication of cold-start data, or (iii) the
application of weighting factors to specific data segments.
Adapting the data evaluation methods themselves, e.g., by applying a circular
calculation of moving averaging windows to ensure events at the beginning and the
end of urban driving are covered by a similar number of windows than events in the
middle of the urban part of a trip.
Table 1: Principal options for adapting RDE provisions to ensure a robust coverage of
cold-start emissions (according to Option 1, i.e., inclusion of cold start into
the RDE data evaluation)
Adaptation
Options for implementation
Controlling for
urban driving
distance
- map-based evaluation of urban driving
- implementation of a specific distance threshold, e.g., minimum RDE
urban distance of 16 km or WLTC distance of 23 km to separate urban
from rural driving
Pre-processing
of data
(Appendix 4)
- cutting the cold start section (i.e., first 300 s after engine start) and
placing it into the middle of the data stream for urban driving, e.g., at
a point when the first moving averaging window has been completed
(≈10km)
- duplicating the cold-start section and placing the duplicate, e.g., at
the end of the test, at the end of urban driving
- multiplying instantaneous cold-start emissions with a fixed or variable
correction factor that accounts for the actual urban distance of a RDE
trip
Adaptations of
the data
evaluation
(Appendix 5)
- circular continuation of the determination of moving averaging
windows to ensure data points at the beginning and end of urban
driving are present in a similar number of windows than data points in
the middle of urban driving
- multiplying windows that contain cold-start emissions with a
correction factor
Various combinations of the three adaptation types could be envisaged and any of these
could be implemented into the RDE test procedure without complex adaptations of the
existing regulatory text. However, depending on the choice, any of the suggested
interventions could impact the severity of the RDE test procedure with respect to the
importance of cold start relative to vehicle operation with a warm engine. Studying the
emission effects might be straightforward in case of the first two adaptation types.
However, studying the emissions effect of adapting the data evaluations in Appendix 5 of
Regulation 2016/427 (EC, 2016a) might require somewhat more complex software
reprogramming. In the following sections, we investigate in a preliminary scenario analysis
the emission effects of a few selected adaptations of the RDE data pre-processing and the
data evaluation with the moving averaging window method (see Sections 3.2 and 4.2).
6
7
3 Methods
3.1 Analysis of driving distance between two consecutive cold
starts
In the first part of the analysis, we conduct a scoping review of information and statistical
data on the length of trips and the frequency of cold starts with the objective to establish a
first-order estimate of the distances typically driven by European vehicle users between two
consecutive cold starts. Our analysis includes data from openly available scientific reports,
peer-reviewed articles, technical presentations, and road vehicle emission models such as
the Handbook of Emission Factors for Road Transport (HBEFA 3.2). This way, we identify
seven sources of information that we present and discuss separately in Section 4.1.
Throughout this report, we define trips as driving events that are delimited by longer
parking durations. Trips may contain multiple short trips, interrupted by stops at traffic
lights or in congested traffic. The reviewed literature applies a variety of methodological
choices to differentiate trips that contain a cold start from those that start with a warm
engine or an engine that has not yet completely cooled down. The cool-down of engines and
after-treatment systems (and their subsequent end-point temperature) generally depends
on vehicle-specific design characteristics, on the driving pattern prior to vehicle parking and
the ambient temperature. As a first, and relatively conservative proxy, we assume based on
own assessments depicted in Figure 3 that after a parking duration of 3-8 h
4
, engine and
after-treatment systems have cooled down. Depending on the specific background
information available from the various studies, we therefore assume that trips following a
parking duration in the range of some 3-8 h contain a cold start. Throughout this report, we
present error margins to indicate the standard deviation of values in a given data sample.
3.2 Scenario analysis - cold-start inclusions
In the second part of the analysis, we focus on modifications in the RDE data pre-processing
according to Appendix 4 and in the actual data evaluation according to Appendix 5 of
Regulation 2016/427 (EC, 2016a). This analysis excludes considerations on the actual urban
driving distance covered by an RDE trip (see first bullet point on Page 3 and Table 1)
because decisions on the respective driving distance that could be taken into account for the
evaluation of urban emissions can be made independently of the applied data evaluation
method, e.g., when designing an RDE trip. Instead, we seek to conduct a scenario analysis
of the NOX emissions effect of various modifications related to (i) the re-arrangement of
emission events in the recorded data stream during pre-processing and (ii) modification of
the weighting approach applied in Appendix 5 individual moving averaging windows. These
modifications address the second and third bullet points on Page 3 as well as Columns 2 and
3 in Table 1 and should ensure cold-start emissions are sufficiently covered in individual
windows and window containing cold-start emissions receive sufficient weight in the
subsequent data evaluation.
4
We acknowledge that this assumption represents a simplification; in reality, the cool-down characteristics of
engines and after-treatment systems differ from each other and should ideally be dealt with separately.
8
As baseline scenario of our analysis, we assume the current RDE provisions (Regulations
2016/427 and 2016/646) that exclude cold start (i.e., pollutants emitted during the first
300 s of a trip) from the data evaluation. This scenario is referred to hereinafter as Mod0a.
We then establish four scenarios that assess modifications of the data pre-processing
according to Appendix 4. These scenarios are aimed at increasing the number of averaging
windows that contain cold-start emissions but leave the RDE data evaluation (i.e., the
weighting of moving averaging windows according to Appendix 5) unchanged:
Scenario Mod0b includes cold start into the normal RDE data evaluation according to
Appendix 5 of Regulation 2016/427 but abstains from re-arranging emission events
in the recorded data stream. This scenario is equivalent to the cold-start proposal of
DG GROW for the third RDE package as of November 2016.
Scenario Mod1a includes cold start into the RDE data evaluation and duplicates the
cold-start segment in the recorded data stream. This scenario thus increases the
number of windows that contain cold-start emissions compared to Scenario Mod0b.
Scenario Mod1b includes cold start into the RDE data evaluation after cutting it from
the beginning of the test and pasting into the middle of the urban part of the
recorded data.
Scenario Mod1c includes cold start into the RDE data evaluation after duplicating the
cold-start segment (i.e., the first 300 s of a trip), cutting the duplicate from the
beginning of the test and pasting it into the middle of the urban part of the recorded
data. This scenario combines Scenarios Mod1a and Mod1b and would increase even
further the number of windows that contain cold-start emissions compared to the
baseline and previous two scenarios.
In a second stop of our scenario analysis, we assess two scenarios that modify the
weighting approach of Appendix 5 to ensure sufficient weight is given to windows containing
cold start. For these scenarios, we include cold start into the normal RDE data evaluation as
done in scenario Mod0b but we do not re-arrange events in the recorded data stream:
Scenario Mod2a applies a constant weighting factor of 1 to the first 300 moving
averaging windows (i.e., the windows that do likely contain cold start emissions),
irrespective of the actual CO2 emissions of these windows. This scenario could
prevent that the RDE data evaluation excludes cold-start windows with very high or
low CO2 emissions that may result from biased driving.
Scenario Mod2b applies a linearly decreasing weighting factor from 2 to 1 for the
first 300 MAWs. This scenario puts additional weight on the first windows of a trip
that contain relatively large shares of cold-start emissions.
In a third step, we explore two selected combinations of scenarios that combine the pre-
processing and the weighting of windows:
Scenario Mod3a combines the duplication of cold start of scenario Mod1c with the
application of a constant weighting factor of 1 for the first 300 MAWs in scenario
Mod2a.
Scenario Mod3b combines the duplication of cold start of scenario Mod1c with a
linearly decreasing weighting factor from 2 to 1 applied to the first 300 moving
averaging windows in scenario Mod2b.
We would like to emphasize that the scenarios are chosen to obtain a first and preliminary
insight into the emissions effect of feasible and easily implementable modifications of the
RDE data pre-processing and evaluation. The scenarios are chosen for a somewhat semi-
9
qualitative assessment of modifications but do not constitute a rigid evaluation of concrete
proposals for the modification of the RDE data evaluation. Alternative, and equally relevant,
modifications could be assessed in the future, e.g., a dedicated evaluation of emissions over
the first moving averaging window of a trip or over a distance typically driven by vehicle
users between two consecutive cold starts.
We demonstrate impact of the various scenarios on the calculated NOX emissions by
determining a cold start performance indicator (CSPI) [%] as:
where:
ECS - average NOX emissions during cold start, i.e., the first 5 min of a RDE
trip [mg/km] (as determined by applying the different scenarios)
EUrban - average NOX emissions during urban driving with a warm engine
[mg/km]
The CSPI is calculated based on the instantaneous emissions data without the application of
the RDE data evaluation methods; the CSPI can be expected to be high/low for vehicles
with high/low cold start emissions compared to the average warm-engine emissions over
the urban part of an RDE trip. The CSPI is a vehicle-specific, technology-specific, and trip-
specific parameter.
We express the relative difference in the NOX emissions between each modelled scenario
and the baseline scenario Mod0a (cold start exclusion from RDE evaluation) as the moving
averaging window performance (MAWP) [%] as:
where:
NOXRDEMod0a - urban NOX emissions calculated with the baseline scenario Mod0a
[mg/km] (using the MAW method according to Regulation 427/2016)
NOXRDEModi - urban NOX emissions calculated with the respective scenario i [mg/km]
(using the MAW method)
We analyse the NOX emissions effect of the various scenarios based on seven on-road tests,
conducted with four vehicles on four test routes (
Table 2). The selected tests were already conducted in 2013 and do not fully comply with
the RDE requirements according to Regulations 2016/427 and 2016/646 (EC, 2016a,b).
Still, we consider our selection to be fit for purpose as vehicles show higher NOx emissions
during cold start than during warm-engine operation for all tests.
10
Table 2: On-road tests used for the scenario analysis
Vehicle
Route
Test name
Gasoline Euro 6
#1
Test 1
Diesel Euro 6 #1
#1
Test 2
Diesel Euro 6 #1
#1
Test 3
Diesel Euro 6 #1
#1
Test 4
Diesel Euro 6 #2
#2
Test 5
Diesel Euro 6 #3
#3
Test 6
Diesel Euro 6 #3
#4
Test 7
11
4 Results
4.1 Driving distance between two consecutive cold starts
4.1.1 GPS car data: Driving patterns in Modena and Florence (Italy)
De Gennaro et al. (2014) and Paffumi et al. (2015) analysed trip characteristics based on a
comprehensive set of GPS car data obtained for the Italian provinces of Modena and
Florence. The data were acquired by on-board loggers whereby a GPS device (used to locate
vehicles) sends data to a remote server via GSM (Global System for Mobile
Communications). The data set comprises 28,000 vehicles, 4.5 million trips, and a total of
36 million vehicle-kilometres. The data were obtained over a one-month period in May
2011.
The analyses of De Gennaro et al. (2014) and Paffumi et al. (2015) suggest that the mean
trip distance in the two provinces is 8 ± 3 km (mean trip distances are 7.8 km and 8.0 km
in Modena and Florence, respectively). Approximately 20% of the parking events lasted 6 h
or longer (Figures 1 and 2; Table 3). Applying the relatively conservative criterion
5
of 6 h or
longer parking durations to distinguish trips that constitute cold starts from those being
started with a warm or semi-warm engine, the data suggest that the mean distance
between two consecutive cold starts (kilometres travelled per number of cold starts as
analysed by Paffumi et al., 2015) is (8 ± 3) km/20% = 40 ± 15 km.
This approximation is based on average trip distances and neglects that the probability
distribution of parking durations and trip distances are positively skewed (non-symmetric).
In fact, Figure 2 suggests that the median trip distance between two consecutive cold starts
is somewhat shorter than the mean distance (i.e., in the range of some 30 km).
5
The longer the cut-off time, the more conservative the criterion. Applying a parking duration shorter that 6 h as
criterion to distinguish between trips with and without cold start would increase the percentage of trips containing a
cold start. At an ambient temperature of 15oC, the engine coolant temperature may have approximately reached
the ambient temperature (see Figure 3). We acknowledge that the level of pollutant emissions depends on the
temperature of the after-treatment systems (but not directly on the temperature of the engine). As the catalyst
likely cools down faster than the engine coolant does, parking durations shorter than 6 h could be assumed to
differentiate trips with and without cold start from each other.
12
Figure 1: Province of Modena (Italy), second-by-second calculations: (a) Frequency
distribution of parking durations in hours (0.5 h bin size); (b) Cumulative
parking duration per day; (c) Rate of parking during the day (hours); (d)
Share of the fleet sample parked during the day (Source: Paffumi et al.,
2015)
After the parking of a vehicle, the cooling of engine and after-treatment systems typically
follows an exponential trend with temperatures decreasing at a declining rate over time,
eventually approaching the ambient temperature asymptotically. Figure 3 shows that at an
ambient temperature of 15oC, the engine coolant temperature (a proxy for the temperature
of the engine and after-treatment systems) might have fallen to below 30oC after a parking
duration of 3 h
6
, which appears to represent some 30% of all parking events in Modena (De
Gennaro et al. (2014); Table 3). Applying the more stringent 3 h criterion to distinguish
between trips with and without cold start yields an average distance travelled between two
consecutive cold starts of (8 ± 3) km/30% = 27 ± 8 km and a median distance between
two consecutive cold starts of 30 km*20%/30% = 20 km.
6
Again, we acknowledge that the level of pollutant emissions depends on the temperature of the after-treatment
systems, which likely cools down faster than the engine coolant does. Moreover, the cooling of the engine depends
on the individual vehicle. Figure 3 therefore constitutes a schematic sketch that could be complemented by more
detailed and vehicle-specific analyses.
13
Figure 2: Frequency distribution of trip lengths, i.e., driving distance between two
consecutive cold starts after a parking duration of at least 6 h (Source:
Paffumi et al., 2015)
Table 3: Distribution of parking durations in the province of Modena (Italy; Source:
De Gennaro et al., 2014)
Parking duration in h
Percentage
Cumulative percentage
0-0.5
43.5
43.5
0.5-1
10.4
53.9
1-1.5
6.6
60.5
1.5-2
4.5
65.0
2-2.5
3.2
68.2
2.5-3
2.5
70.7
3-6
10.5
81.2
6-12
9.3
90.5
12-24
7.5
98.0
>24
2.0
100
14
Figure 3: Generic cool-down curve for engine coolant (Source: EC, 2015)
4.1.2 Green eMotion data
Donati et al. (2015) complemented the driving data analysed by De Gennaro et al. (2014)
and Paffumi et al. (2015) with those from hybrid and electric vehicles that participated in
the Green eMotion project. The Green eMotion data contain information about the driving
patterns of hybrid and electric cars, motorcycles, and transporters recorded by on-board
data loggers in the period between March 2011 and December 2013. The vehicles were
driven in 11 demonstration regions, in various cities of six European countries (Denmark,
France, Germany, Ireland, Italy and Sweden). The data set comprises 457 vehicles and a
total of 65,799 trips. The mean and median trip distances travelled are 7.8 km and 4.8 km,
respectively (Figure 4). Donati et al. (2015) argue that the mean trip distance is negligibly
shorter than the one identified by De Gennaro et al. (2014) and Paffumi et al. (2015; see
Section 4.1.1) because the Green eMotion vehicles were mainly propelled electrically and
thus tend to be driven predominantly within cities. This observation may make the driving
data obtained by Donati et al. (2015) specifically appropriate for characterizing the EU-wide
driving pattern in urban environments.
15
Figure 4: Distribution of trip distances as obtained from the Green eMotion data; bars
(blue) denote the relative frequency of trip distances; the red line denotes the
cumulative frequency distribution of trip distances (Source: Donati et al.,
2015)
As the distribution of trip distances, also the distribution of parking times is skewed towards
shorter parking durations (Figure 5). Parking times reach on average 3.3 h with some 27%
and 20% of parking events being longer than 3 h and 6 h, respectively. These observations
are well in line with the findings of De Gennaro et al. (2014) and Paffumi et al. (2015). If we
apply the 27% and 20% criteria to distinguish trips with and without cold starts, we obtain
the following driving distances between two consecutive cold starts:
Mean distance between two consecutive cold starts (≥3 h parking):
7.8 km/27% = 29 km.
Mean distance between two consecutive cold starts (≥6 h parking):
7.8 km/20% = 39 km.
Median distance between two consecutive cold starts (≥3 h parking):
4.8 km/27% = 18 km.
Median distance between two consecutive cold starts (≥6 h parking):
4.8 km/20% = 24 km.
Trip Distance
(valid trips=65,799)
distance (Km)
Relative freq.
010 20 30 40 50 60
0.0 0.1 0.2 0.3 0.4
Average 7.79
Median 4.75
Mode 2.5
0.0 0.2 0.4 0.6 0.8 1.0
cumulative scale
~80% at d=10 km
16
Figure 5: Distribution of parking times as obtained from the Green eMotion data; bars
(blue) denote the relative frequency of parking times; the red line denotes
the cumulative frequency distribution of parking times (Source: Donati et al.,
2015)
4.1.3 Driving data from the Netherlands
Klein et al. (2015) presented data on the trip distance and frequency of cold starts in a
mobility study conducted already in 1995 by Statistics Netherlands (CBS). The study
consists of interviews of a large, random group of car owners about the use of their vehicle
on particular days. According to Klein et al. (2015), the average distance of trips in the
Netherlands is 14.5 km
7
. Approximately 60% of trips are assumed to contain a cold start
8
;
the total number of cold starts per travelled kilometre is therefore 0.04 (i.e., a cold start
happens on average every 24 km). Approximately 95% of all cold starts take place within
urban areas. In 1995, according to Statistics Netherlands, 25% of the passenger vehicle
kilometres were driven within urban areas and about 35% on rural roads; the number of
cold starts per passenger car kilometre on urban roads is approximately 0.15 (i.e., a cold
start happens on average every 6.7 km) but, due to longer trip distances, on rural roads
only 0.05
9
(i.e., a cold start happens every 20 km).
We note that the cold-start frequency of 60% as identified by Klein et al. (2015) differs
considerably from the 20-30% presented in Sections 4.1.1 and 4.1.2. Moreover, the mean
travelled distance of 14.5 km identified by Klein et al. (2015) is almost double that of 8 km
and identified by De Gennaro et al. (2014), Paffumi et al. (2015), and Donati et al. (2015).
A possible explanation is that these authors considered predominantly urban driving, and
that the Green eMotion data consider urban trips of electric vehicles. The deviations in the
7
More recent data for the year 2015 suggests that Dutch drivers make on average 0.85 car trips per day, thereby
covering a distance of 18 km (Ligterink (2016) based on CBS (2016)).
8
Klein et al. (2015) refer to cold start as driving with a cold engine (presumably at ambient temperature) without,
however, specifying after which parking duration the conditions for a cold-start are satisfied.
9
Klein et al. (2015) specify the number of cold starts per passenger car kilometre for rural roads to be
approximately 0.005. Personal communication with Ligterink (2016) suggests the actual number of cold starts is in
fact a factor ten lower, i.e., 0.05.
Parking Time Duration
(valid stops=47,093)
duration (h)
Relative frequency
0 5 10 15 20
0.0 0.1 0.2 0.3 0.4 0.5
Average 3.96
Median 1.19
Mode 0.25
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
cumulative scale
17
frequency of cold starts is most likely caused by different assumptions regarding the
minimum parking time after which the engine has cooled down and a new trip begins with a
cold start. According to Table 3, 60% of parking times in Modena (Italy) are within a
duration of 0.5 h; this time interval is relatively short to allow for engine cool-down (see
also Figure 3). In fact, Klein et al. (2015) do not specify the duration of parking time used
to determine whether a vehicle start is a cold start. Moreover, whereas the GPS car data
(De Gennaro et al., 2014; Paffumi et al., 2015) and Green eMotion data (Donati et al.,
2015) are based on measured trip distances and durations, the data Klein et al. (2015)
analysed are based on surveys. Ligterink (2016) argues that the data presented by Klein et
al. (2015) might be more representative of the average car use than GPS data as the
former also include older cars. In the Netherlands, cars of 8 years and older typically drive
less than 10,000 km per year and are predominantly used for urban driving.
4.1.4 Driving data from Sweden
Karlsson (2013) logged the movements of 432 passenger cars in private use with a GPS for
a research project on car movements in Sweden. The cars had an age of 0-100 months
since registration and were driven in Västra Götaland county (including Gothenburg, the
second-largest city in Sweden) and the Kungsbacka municipality. The cars were randomly
selected from the Swedish vehicle register; loggings were distributed over the seasons in
the period between 2010 and 2012. Karlsson (2013) considers the car sample to be
approximately representative of Sweden in terms of movement patterns, car ownership,
and the coverage of larger and smaller towns and rural areas. The movements of each car
were logged for 1-3 months (58 days on average).
Karlsson (2013) find that the majority of trips are shorter than 5 km (Figure 6) and some
25% of parking events last for 6 h or longer (Figures 7). Meta data provided by Karlsson
and Björnsson (2016) through personal communication suggest that cars were driven on
average 47 ± 25 km (mean) and 42 km (median)
10
between parking events that lasted 6
h or longer.
10
This median represents the median of the mean driving distances covered by each individual car between
parking durations of 6 h or longer.
18
Figure 6: Frequency distribution of trip distances in Sweden (Source: Karlsson, 2013)
Figure 7: Frequency distribution of parking durations (referred to here as break time T)
in Sweden (Source: Karlsson, 2013)
19
4.1.5 WLTP data base and miscellaneous data sources
An analysis of 430,000 km and 35,850 trips of European driving data contained in the entire
data base of trips used for the development of the Worldwide harmonized Light Vehicles
Test Procedure (WLTP) suggests a mean trip length of 10.5 km (Steven, 2016). Donati et al.
(2015) also found a mean of 9.8 km (Modena) and 10.3 (Florence) in their re-analyses of all
the floating-car data, i.e., not restricting their analysis to urban driving. Assuming an mean
trip distance of 10.5 km and a cold-start frequency of 27% (≥3 h parking) and 20% (≥6 h
parking), yields mean distances between two consecutive cold starts of 39 km and 53 km,
respectively.
A back-of-the-envelope calculation proposed by ACEA (2016) suggested an average annual
mileage per car of ≥14,000 km, some ≤2 cold-starts per day (i.e., ≤720 cold starts per
year), and thus an average distance between two consecutive cold starts of ≥19.4 km.
In preparation of the work on the 3rd RDE package, the JRC had compiled a preliminary
review of European trip distances (Table 2). The results are in line with the other findings
presented in Section 4.1.1, suggesting the mean and median distance of trips (including
urban and extra-urban driving) in Europe reach 14 ± 5 km and 12 km, respectively. As
observed previously, the distribution of trip distances is positively skewed, with the majority
of trips being shorter than the mean.
If we assume that 27% and 20% of trips begin with a cold start (De Gennaro et al., 2014;
Paffumi et al., 2015), we obtain the following driving distances between two consecutive
cold starts:
Mean distance between two consecutive cold starts (≥3 h parking):
14 ± 5 km/27% = 52 ± 19 km.
Mean distance between two consecutive cold starts (≥6 h parking):
14 ± 5 km/20% = 70 ± 25 km.
Median distance between two consecutive cold starts (≥3 h parking):
12 km/27% = 44 km.
Median distance between two consecutive cold starts (≥6 h parking):
12 km/20% = 60 km.
20
Table 4: Preliminary review of European trip distances based on miscellaneous sources (not all trips driven in one day may
include a cold start; primary sources of information not included in the list of references)
Source
Year
Country
Trips per day
Distance per trip [km]
Comment
Pasaoglu et al. (2012)
2012
France
2.9
19 commuting; 15 free time1
Based on a web survey of 600
participants in 6 EU Member States
Germany
2.6
25 commuting; 15 free time1
Italy
2.7
17 commuting; 16 free time1
Spain
2.4
25 commuting; 34 free time1
Poland
2.5
24 commuting; 20 free time1
United Kingdom
2.5
18 commuting; 15 free time1
ISFORT (Mobility in Italian
cities)
2008
Italy
3.12 (average of
mobility in big cities)
12 (average of mobility in big
cities)
15,000 interviews per year, age
between 14-80, working days
Città metropolitane: Mobilità,
crisi e cambio modale
2015
Italy
2.7 (average of
mobility in big cities)
11.5 (average of mobility in big
cities)
15000 interviews per year, age
between 14-80, working days
XX Rapporto Aci-CENSIS
2012
Italy
3 on week days; 2.1
on weekends
10 on weekdays; 11 on
weekends
Average of various cities
National Travel Survey:
2013
2013
United Kingdom
2.5
11.5
Survey based on about 9000
households
Mobilität in Deutschland
2008
2008
Germany
3.5
11.8
Survey based on 25000 households
Eurostat: Passenger mobility
in Europe
1999-
2001
Denmark
2.7
12.7
Data collected between1999-2001;
data collected based on populations
of up to 40,000 households
depending on the studied country
Germany
3.4
11.7
France
2.9
12.2
Latvia
1.9
4.6
The Netherlands
3.3
10.2
Austria
3
9.4
Finland
2.9
15.4
Sweden
2.7
16.3
United Kingdom
2.9
11
Switzerland
3.6
13
Norway
3.3
11.5
Traffic patterns
2010
Switzerland
2.7
13.6
Survey based on 63000 individuals
National travel survey
2009
Ireland
2.4
13
Survey based on 7000 households
Car ownership, travel and
land use: a comparison of
the US and Great Britain
2006
United Kingdom
3
8.7
Mean
2.8
13.8
Standard deviation
0.4
4.9
Median
2.7
12.2
1For the calculation of the overall mean, standard deviation, and median, we assume commuting occurs on 5 out of 7 days of the week and driving in free time
on 2 out of 7 days of the week.
21
An analysis conducted by the European Federation for Transport and Environment (T&E
2016a,b) based on a collection of data from national travel surveys of various European
countries (Belgium, France, Germany, Italy, UK) and complemented with data for Latvia and
Sweden suggests that roughly 50% of car trips in urban environments are shorter than 6
km (Figure 8). This observation is consistent with the findings of Karlsson (2013) and
Donati et al. (2015). Assuming that 27% and 20% of trips begin with a cold start (De
Gennaro et al., 2014; Paffumi et al., 2015), we obtain the following median driving
distances between two consecutive cold starts:
Median distance between two consecutive cold starts (≥3 h parking):
6 km/27% = 22 km.
Median distance between two consecutive cold starts (≥6 h parking):
6 km/20% = 30 km.
Figure 8: Distribution of urban trip distances in the EU based on miscellaneous data
sources (Source: T&E, 2016a,b)
4.1.6 Handbook Emission Factors for Road Transport (HBEFA 3.2)
The data contained in HBEFA (version 3.2) suggest the overall mean and median distance of
trips in six European countries reach 8.3 ± 1.6 km and 4.9 km, respectively (Table 5). The
positively skewed distribution of trip distances results in a relatively low median with the
majority of trips being shorter than the mean. The distribution of parking times in HBEFA
3.2 suggests that on average over all countries covered, 31 ± 7 % (mean) and 35%
(median) of all trips are started under cold conditions, i.e., after a parking time of more
than 8 h (Kühlwein, 2016; Table 6). By contrast, hot start after a stand-still time of less
than 1 h accounts for some 37 ± 9% (mean) and 35% (median) of all trips. The remaining
trips are started under warm conditions after parking times between 1 h and 8 h (Kühlwein,
2016).
0%
10%
20%
30%
40%
50%
60%
70%
80%
< 1 km < 2 km < 3 km < 4 km < 5 km < 6 km < 7 km < 8 km < 9 km < 10 km
Proportion of trips in EU km by km
Cars < 10 km All modes
22
Table 5: Frequency distribution of trip distances in 6 European countries (Source:
Kühlwein (2016) based on HBEFA 3.2); primary sources of information not
included in the list of references
Distance
class
Country
Germany
Austria
Switzerland
France
Norway
Sweden
Data source
Not
specified
Not
specified
Not specified
Lille, 1998
Statistics
Norway
from
instrumented
cars, VTI
Average
distance per
class [km]
Frequency [%]
0-1 Km
0.50
10.30
10.00
7.00
11.50
15.45
23.77
1-2 Km
1.50
13.40
13.00
9.00
20.50
13.39
7.43
2-3 Km
2.50
10.90
11.00
12.50
14.00
9.64
7.92
3-4 Km
3.50
7.05
7.00
5.00
10.00
6.30
9.90
4-5 Km
4.50
7.05
7.00
8.00
8.00
7.65
7.92
5-6 Km
5.50
6.05
6.00
9.50
6.00
4.67
3.96
6-7 Km
6.50
6.05
6.00
6.50
4.00
3.40
2.48
7-8 Km
7.50
3.03
3.00
4.50
3.50
3.66
8.91
8-9 Km
8.50
3.03
3.00
3.50
2.50
1.06
5.94
9-10 Km
9.50
3.03
3.00
3.00
2.50
6.52
2.97
10-11 Km
10.50
2.06
2.00
3.50
2.00
0.62
1.39
11-12 Km
11.50
2.06
2.00
3.00
2.00
1.73
1.39
12-13 Km
12.50
2.06
2.00
2.00
2.00
0.79
1.39
13-14 Km
13.50
2.06
2.00
1.50
2.00
0.56
1.39
14-15 Km
14.50
2.06
2.00
1.50
1.00
3.93
1.39
15-16 Km
15.50
0.96
1.00
2.00
1.00
0.64
0.52
16-17 Km
16.50
0.96
1.00
1.00
1.00
0.57
0.52
17-18 Km
17.50
0.96
1.00
1.50
1.00
0.57
0.52
18-19 Km
18.50
0.96
1.00
1.00
1.00
0.16
0.52
19-20 Km
19.50
0.96
1.00
1.00
1.00
4.08
0.52
>20 Km
30.00
15.00
16.00
13.50
3.50
14.60
9.24
Distance of all trips
Mean [km]
8.3 ± 1.61
(all countries)
9.2 km
9.5 km
9.4 km
5.8 km
9.1 km
7.1 km
Median
[km]
4.9
(all countries)
5.2 km
5.3 km
5.9 km
3.7 km
5.1 km
4.2 km
1 Uncertainty margin corresponds to the standard deviation of mean trip distances for all countries.
This observation has two implications. First, when applying the criterion of 8 h parking time,
the average distance between two consecutive cold starts may range between 8.3 ± 1.6
km/31 ± 7% = 27 ± 8 km (mean distance/mean share cold-start trips) and 4.9 km/35% =
14 km (median distance/median share cold-start trips)
11
. Second, starting conditions in
which the engine is not completely cooled down or warmed up constitute indeed a
substantial part of real-world driving. To account for such intermediate conditions at the
beginning of a trip, HBEFA includes specific emission factor functions for each pollutant and
vehicle concept (Kühlwein, 2016).
11
The distances calculated here represent conservative estimates. Considering that after-treatment systems cool
down faster than the engine coolant does, one could assume shorter parking durations, which in turn yields shorter
distances between two cold starts. More detailed scenario analyses could complement the analysis presented here.
23
Table 6: Frequency distribution of parking time in 6 European countries (Source:
Kühlwein (2016) based on HBEFA 3.2); primary sources of information not
included in the list of references; EI-TG – Eco-Innovations Technical
Guidelines (EC, 2015)
Country
Germany
Germany
Austria
Switzerland
France
Norway
Sweden
EI-TG
Data source
DRIVE,
1991
SRV, 1994
DRIVE,
1991
MZ05
MZ10
Lille,
1998
Statistics
Norway
from
instrument
ed cars,
VTI
Time class
[min]
Frequency [%]
<30
14.14
21.79
14.14
20.21
17.07
22.00
42.79
30-60
27.46
12.41
27.46
10.02
10.41
36.00
8.30
13.93
36.00
60-90
3.60
6.79
3.60
6.52
7.06
5.10
1.99
90-120
7.00
6.79
7.00
5.26
4.63
9.00
5.80
4.98
13.00
120-150
2.21
2.98
2.21
3.69
3.63
3.20
5.97
150-180
4.29
2.98
4.29
3.31
2.98
6.00
3.60
1.99
6.00
180-210
1.43
1.78
1.43
2.51
2.57
2.20
2.99
210-240
2.77
1.78
2.77
2.69
2.72
9.00
2.60
1.49
4.00
240-270
0.83
0.88
0.83
2.86
2.56
1.50
1.49
270-300
1.62
0.88
1.62
2.31
2.40
7.00
1.60
1.00
2.00
300-330
0.83
0.74
0.83
1.56
1.68
1.00
1.00
330-360
1.62
0.74
1.62
0.98
1.30
3.00
1.10
0.50
2.00
360-390
0.48
0.77
0.48
0.75
0.83
0.80
1.00
390-420
0.92
0.77
0.92
0.71
0.67
6.00
1.40
0.50
1.00
420-450
0.48
0.51
0.48
0.61
0.67
1.80
0.50
450-480
0.92
0.51
0.92
0.63
0.83
24.00
2.40
0.50
1.00
480-510
29.40
0.91
29.40
0.85
0.94
2.10
0.50
510-540
1.19
1.08
1.23
2.00
0.50
3.00
540-570
1.45
1.36
1.49
1.00
0.50
570-600
1.80
1.48
1.86
0.90
0.50
4.00
600-630
1.72
1.36
1.78
0.90
0.50
630-660
1.42
1.34
1.47
0.90
0.50
3.00
660-690
1.30
1.13
1.34
0.80
0.50
690-720
1.36
1.27
1.40
26.90
13.88
1.00
>72
25.73
25.54
26.53
24.00
COLD
≥8h
29
37
29
35
38
24
36
17
35
HOT ≤1h
42
34
42
30
27
36
30
57
36
Warm
>1 to<8 h
29
29
29
34
35
40
34
26
29
4.1.7 CADC and related trip analyses
The research of André et al. (1999) in support of establishing the Common Artemis Driving
Cycle (CADC) suggests that 69% of trips start with a cold or not fully warmed up engine
(Table 7). This observation is in line with the data of HBEFA 3.2 (Kühlwein, 2016) and the
findings of De Gennaro et al. (2014), Paffumi et al. (2015), and Donati et al. (2015). The
trip distance driven after a cold start reaches some 9.1 km (Table 8). However, this distance
only considers the trip following a cold start and does not consider any driving distance
potentially covered by subsequent trips that start with a warm or partially cooled-down
engine. We would thus argue that the driving distance between two consecutive cold starts
is thus longer than the 9.1 km obtained from André et al. (1999).
24
Table 7: Percentage of driving distance covered by trips that start with a cold or
partially warmed up engine (Source: André et al. (1999) cited from André and
Joumard, 2005)
Average trip
speed [km/h]
Winter
(4 months)
Summer
(4 months)
Intermediate season
(4 months)
Full year
Percentage of total driving distance
<10
61.7
62.7
58.9
61.3
10 to 20
71.9
71.1
56.9
67.7
>20 to 30
71.8
67.1
62.8
67.7
>30 to 40
78.8
68.6
64.8
72.2
>40 to 50
80.9
76.3
66.5
75.6
>50 to 60
77.0
76.7
60.6
71.7
>60 to 70
74.6
67.9
76.6
72.9
>70
67.3
58.9
57.6
62.9
Total
73.4
67.3
63.2
69.0
Table 8: Distance of trips containing cold start (Source: André et al. (1999) cited from
André and Joumard, 2005)
Distance
class
[km]
Average
distance
[km]
Speed category [km/h]
Total
<10
10 to 20
>20 to 30
>30 to 40
>40 to 50
>50
Average speed to reach warm-engine conditions [km/h]
5.4
15.3
24.9
34.6
44.3
60.9
Distance [km/h]
<0.5
0.2
0.3
0.1
0.0
0.0
0.0
0.0
0.1
0.5 to 1
0.8
0.6
0.5
0.2
0.0
0.0
0.0
0.2
>1 to 2
1.5
2.2
2.4
1.7
0.6
0.1
0.0
1.2
>2 to 3
2.5
3.0
4.3
3.9
1.3
0.5
0.0
2.5
>3 to 4
3.4
2.2
3.0
3.2
2.6
1.2
0.2
2.3
>4 to 5
4.5
3.2
4.7
4.7
2.9
1.6
0.7
3.3
>5 to 6
5.5
3.9
5.6
4.9
3.9
1.8
1.2
2.9
>6 to 7
6.5
3.9
4.9
4.2
4.0
1.9
0.8
3.5
>7 to 8
7.4
2.3
3.3
3.4
2.4
3.4
0.7
2.8
>8 to 9
8.5
1.9
2.1
2.3
2.6
1.1
0.7
1.9
>9 to 10
9.5
1.0
1.7
3.1
4.3
2.1
1.3
2.5
>10 to 11
10.5
1.2
3.7
2.4
4.6
3.6
0.9
3.0
>11 to 12
11.5
2.2
2.4
2.8
4.0
1.5
0.9
2.5
>12
34.6
71.9
61.3
63.1
66.8
81.4
92.5
70.3
Total
9.1
100.0
100.0
100.0
100.0
100.0
100.0
100.0
25
4.2 Scenario analysis – cold-start inclusion
4.2.1 Baseline scenario and simple RDE cold-start inclusion
We start out by characterizing the NOX emissions of the selected tests. Figure 9 presents
the RDE results over the urban part of trips for the baseline scenario Mod0a (exclusion of
cold start from the RDE data evaluation in accordance with Regulation (EU) 2016/427) and
scenario Mod0b (inclusion of cold start into the RDE data evaluation).
Figure 9: NOX emissions after the evaluation of the urban part of an RDE trip with and
without the inclusion of cold-start
26
Test 1 on the far left in Figure 9 shows no difference in the NOX emissions between the
baseline scenario (Mod0a) that excludes cold start and the cold-start inclusion into the
normal RDE data evaluation (Mod0b) although this Euro 6 gasoline vehicle (see Table 2),
like the other three vehicles included in our scenario analysis, showed higher NOX emissions
during cold start than during warm-engine operation. Moreover, only for 3 out of 7 tests,
the inclusion of cold start into the normal RDE data evaluation (Mod0b) leads to an
increase of the overall urban NOX emissions.
For three tests, the overall urban NOx emissions even decrease. This observation could
potentially be explained by the weighting of cold-start windows. Cold start is typically
characterized by elevated CO2 emissions that, in turn, increase the average CO2 emissions
of cold-start windows and could thus lead either to an exclusion or a low weighting of the
window-average NOX emissions in the RDE data evaluation. The observation that the
inclusion of cold start into the normal RDE evaluation could has no effect or even decreases
the calculated NOX emissions (although vehicles show higher NOX emissions [mg/km] during
cold start than during warm-engine operation) suggests that a simple inclusion of cold start
into the RDE data evaluation may not cover cold-start emissions in a robust manner.
4.2.2 Modifications of RDE data pre-processing
The emissions effect of scenarios Mod1a (duplication of cold start at the beginning of a
test), Mod1b (cutting the first 300 s of cold start from the beginning of a test and placing it
in the middle of urban driving), and Mod1c (combination of Mod1a and Mod1b) are
displayed in Figure 10.
The duplication of cold start at the beginning of a test (Mod1a) resulted in higher urban
NOX emission in 4 tests, and lower emissions in 2 tests compared to baseline scenario
(Mod0a). The highest increase of about 50% in the urban NOX emissions is observed for the
second and fourth tests displayed in Figure 10. The Euro 6 gasoline car (Test 1) showed no
difference in the NOX emissions between scenarios Mod0a and Mod1a.
The shift of the cold start to the middle of the urban driving (Mod1b) resulted in higher
urban NOX emissions in 6 tests, and very slightly lower in 1 test relative to the baseline
scenario Mod0a. The duplication of cold start and the subsequent placement of the
duplicate into the middle of the urban part of a trip (Mod1c) resulted in higher urban NOX
emissions for 4 tests, and slightly lower NOX emissions for 2 tests relative to the baseline
scenario (Mod0a). For Test 6, no difference between the baseline scenario (Mod0a) and
Mod1c was found.
Scenarios Mod0b and Mod1b both include cold start but at different location in the data
stream. The urban NOX emissions as evaluated with the moving averaging window method
(EC, 2016a) tend to be higher when the cold-start section is included in the middle of urban
driving (Mod1b) in 5 out of 7 tests compared to the scenario in which cold start is located
at beginning of a test (scenario Mod0b).
27
Figure 10: NOX emissions effect of modifying the RDE data pre-processing
Scenarios Mod1a and Mod1c include two times the cold start section, but at different
locations in the data stream. The urban NOX emissions are higher when the second cold-
start section is included in the middle of the urban part (Mod1c) in 5 out of 7 tests
compared to the scenario where the duplicated cold-start is placed at the beginning of a test
(Mod1a).
To conclude, our scenario analysis of modifications in the RDE data pre-processing suggests
that pasting cold start in the middle of urban driving tends to increase the evaluated NOX
emissions relative to the default scenario in which cold start remains at the beginning of the
test data cold. This observation could be explained by the larger number of windows
covering the cold-start emissions if these are placed in the middle of urban driving. Still, the
modifications of the data pre-processing did not yield a consistent increase in the urban NOx
28
emissions for all tests. This observation (i) points again to the potential exclusion or
weighting of cold-start windows if these show comparatively high CO2 emissions and (ii)
highlight the need to modify the moving averaging window method to cover cold start in a
robust manner by the RDE test procedure.
4.2.3 Modifying the weighting of moving averaging windows
The emissions effect of scenarios Mod2a (constant weighting factor of 1 applied for the first
300 MAWs, regardless of distance-specific CO2 emissions) and Mod2b (linearly decreasing
weighting factor from 2 to 1 imposed for the first 300 MAWs) is presented in Figure 11.
Applying a weighting factor of 1 for the first 300 windows (Mod2a) increases the urban NOX
emissions for 4 tests and decreases the emissions for 3 tests relative to the baseline
scenario Mod0a. Scenario Mod2a resulted in equal urban NOX emissions for 5 tests than
scenario Mod0b (inclusion of the cold start into the normal RDE evaluation of urban
driving). This observation suggests that for these 5 tests, the CO2 emissions of all windows
covering the cold-start period were within the 25% primary tolerance around the CO2
reference.
The application of a linearly decreasing weighting factor (Mod2b) resulted in higher urban
NOX emissions for 4 tests and lower emissions for 3 tests, relative to the baseline scenario
Mod0a. The linearly decreasing weighting factor applied in scenario Mod2b results in
higher urban NOX emissions for 4 tests and lower emissions for 3 tests compared to the
application of a fixed weighting factor of 1 in scenario Mod2a. This result suggests that the
linearly decreasing weighting factor (Mod2b) amplifies the emissions effect observed in
scenario Mod2a (application of a weighting factor of 1 for the first 300 windows) compared
to the baseline scenario Mod0a.
To conclude, modifying the weighting of moving averaging windows ensures that all MAWs
containing cold-start emissions are actually included in the calculation of the final RDE
result. However, a modified weighting may not catch for all tests the excess NOx emissions
related to cold start. Overall increasing or decreasing urban NOx emissions as the result of a
modified weighting approach are possible as the application of, e.g., a fixed weighting factor
changes also the warm-engine NOX emissions that are contained in cold-start windows. If
the warm-engine NOX emissions contained in cold-start window are relatively low, the
application of a fixed weighting could decrease the overall urban NOX emissions, even if
cold-start emissions are on average lower than the warm-engine emissions.
29
Figure 11: NOX emissions effect of modifying the weighting of moving averaging windows
4.2.3 Combining modifications in the pre-processing and evaluation of NOX
emissions
Figure 12 depicts the NOX emissions effect of combining scenario Mod1c (duplication of cold
start in the middle of the urban driving) with the two scenarios that adapt the weighting of
MAWs (Mod2a and Mod2b).
The duplication of cold start in the middle of the urban part combined with a weighting
factor of 1 for the first 300 windows (Mod3a) resulted in higher urban NOX emissions for 5
tests and slightly lower emissions for 1 test, relative to the baseline scenario Mod0a. The
duplication of cold start in the middle of the urban part combined with a linearly decreasing
30
weighting factor (Mod3b) resulted in higher urban NOX emissions for 5 tests and slightly
lower emissions for 2 tests, relative to the baseline scenario Mod0a. Scenario Mod3b tends
to increase urban NOX emissions to a large degree than scenario Mod3a; the former
resulting in a maximum emissions increase of more than 60% for 3 out of the 7 tests,
relative to the baseline scenario Mod0a.
Figure 12: NOX emissions effect of adapting the pre-processing of cold-start emissions
and weighting of moving averaging windows
It appears that a combination of modifications of the RDE data pre-processing and the
weighting approach for cold-start windows (as assessed by scenarios Mod3a and Mod3b)
can better capture the excess cold-start NOX emissions than an application of these two
modifications separately. Further scenarios could be explored to substantiate this
observation (see discussion in Section 5).
31
4.2.4 Summary of the scenario analysis
The percentage deviation between the NOX emissions of each modelled scenario and the
baseline scenario Mod0a can be expressed in terms of the moving averaging window
performance (MAWP) and plotted as a function of the actual incremental cold-start
emissions, expressed here as cold start performance (CSPI) of each tested vehicles (Figure
13, Table 9). A robust coverage of cold-start emissions would result in an approximately
linear relationship between the NOX emissions effect (MAWP) of a given scenario and the
incremental cold-start NOX emissions of a given vehicle (expressed in terms of CSPI).
Among the scenarios modifying the data pre-processing, scenarios Mod1a and Mod1c that
include a duplication of the cold-start phase display the highest overall sensitivity to the
cold-start performance of the tested vehicles. Among the scenarios modifying the weighting
of moving averaging windows, the linearly decreasing weighting factor in scenario Mod2b
appears to represent more accurately the cold-start performance of the tested vehicles than
Mod2a does.
Yet, as discussed in Section 4.2.3, the scenarios combining modifications of the data pre-
processing and the weighting of cold-start windows (Mod3a and Mod3b) showed the
highest sensitivity to the actual cold-start performance of vehicles.
This conclusion is supported by a verification the statistical significance of the slope
coefficients displayed in Table 9. Slope coefficients are generally not statistically significant
(p-value > 0.1) and the coefficients of determination (R2) low, given the data variability and
the small data sample used for this analysis. This observation suggests that (taken
individually) the applied modifications are generally not able to reflect the cold start
emissions in a robust manner.
However, exceptions are scenarios Mod2a (p-value = 0.05), Mod2b (p-value = 0.036),
Mod3a (p-value = 0.023) and Mod3b (p-value = 0.013) for which the slope coefficients
are significant. The intercept coefficients for all scenarios are not statistically significant (p-
value > 0.1), thus not significantly different from zero. Overall, our statistical analysis
suggests that the assessment presented is indeed partial; additional modifications (next to
those implemented and tested here) could be investigated to achieve a robust coverage of
cold-start emissions within RDE.
32
Figure 13: Relationship between MAWP and CSPI expressed through a simple linear
model; blue dots depict the gasoline vehicle; red dots depict the diesel
vehicles; coloured areas depict the confidence interval around the regression
line
33
Table 9: Coefficients of the linear regression model fitted to explain MAWP as a
function of CSPI
Scenario
Axis intercept [%]
Slope
R2
Mod0b
-3.8
0.07
0.154
Mod1a
-1.4
0.11
0.129
Mod1b
5.8
0.06
0.072
Mod1c
4.8
0.10
0.087
Mod2a
-9.7
0.15
0.568
Mod2b
-13
0.21
0.619
Mod3a
-9.8
0.32
0.679
Mod3b
-15
0.41
0.742
34
5 Discussion and conclusions
5.1 General aspects
This report addresses two topics that are relevant for the establishment of a cold-start test
procedure as part of the 3rd regulatory RDE package, namely (i) the distance typically
driven by vehicle users between two consecutive cold starts and (ii) scenarios for a robust
coverage of cold-start emissions by the RDE moving averaging window method. The
analyses presented in this report on the two topics are intended to provide rationale for the
stakeholder discussions in the RDE working group but should be considered preliminary. A
more systematic collection and assessment of European driving data and an expansion of
our scenario analysis through alternative modifications of the RDE data pre-processing and
evaluation as well as the inclusion of additional vehicle test could add to the findings
presented here.
We acknowledge that the number of data sources used for this analysis is rather limited in
view of the diversity in driving patterns, vehicle characteristics, and socio-economic
conditions within the EU. In occasions, we obtained data from the ‘grey’, non-peer-
reviewed, literature and through personal communication. A major source of uncertainty
represents the assumed minimum parking durations that are necessary to cool down the
engine and after-treatment systems. As the temperature of these components at the start
of a trip is a function of vehicle-specific parameters, ambient temperature, and the driving
pattern prior to vehicle parking, cold-start frequencies may vary between vehicles and
depending on season and geographical location. Yet, we regard the reviewed driving data to
be suitable and our findings robust as a first order approximation of the typical distances
driven in Europe between two consecutive cold starts.
Our scenario analysis on the modifications of the RDE data pre-processing and evaluation
confirms that a simple inclusion of the cold start into the normal RDE data evaluation of
Appendix 5 (Regulation 2016/427; EC, 2016a) might not capture for all vehicles and tests
conditions the excess NOx emissions from cold start in a robust manner. This observation
can be attributed to the peculiarities of evaluating emissions data with the moving
averaging window method (see Section 2). Given limited availability of resources, we have
addressed here a limited number of scenarios but not yet, e.g., modifications of the actual
calculation procedure of moving averaging windows (starting the calculation of moving
averaging windows from the beginning of a test; applying a circular calculation of windows
to ensure emissions data at the beginning and end of urban driving are contained in the
same number of windows as data located in the middle of urban driving). Our preliminary
analysis has mimicked the potential effects of such modifications to some extent. Yet, we
see scope for assessing additional, and equally relevant, modifications in the future, e.g., a
dedicated evaluation of emissions over the first moving averaging window or over a distance
typically driven by vehicle users between two consecutive cold starts. Moreover, a circular
calculation of windows could be investigated to understand the feasibility and potential
emission effects of looping back the window calculation to the beginning of a test. Assessing
this specific scenario will require some re-programming of the RDE data evaluation tools and
could be combined with a fixed weighting factor of cold-start windows. Additional
assessments could focus on the application of weighting factors directly to the cold-start
emissions as part of the pre-processing of data. Moreover, our analysis has not addressed
the second RDE data evaluation method (i.e., the power-binning method described in
Appendix 6 of Regulation 2016/427). From the discussions in Section 2, we would expect
that also power-binning in its current form may not be able to capture cold-start emissions
in a robust manner. Further scenario analyses could investigate the feasibility of
modifications of this method.
35
5.2.1 Conclusions on the driving distance between two cold-starts
Table 10 summarizes the data collected in our scoping review. Columns 3 and 4 display the
mean and median trip distances. The frequency of parking events longer than, e.g., 3-8 h
are shown in Column 5. Columns 6 and 7 then contain the distance between two
consecutive cold starts, calculated by dividing the distances given in Columns 3 and 4 with
the cold-start frequencies assumed in Column 5. Based on the analysis presented in Section
4.1 and the data summarized in Table 10, we draw the following conclusions:
The distribution of trip distances tends to be positively skewed with the majority of
trips being shorter than the arithmetic mean trip distance. This observation suggests
that the median rather than the mean might represent best the general trend in the
trip distances.
Urban trips (based on all literature sources shown in Table 10: 7 ± 2 km (mean)
and 6 ± 2 km (median)
12
) tend to be shorter than the overall average trip driven
in urban and extra-urban environments (based on all literature sources shown in
Table 10: 10 ± 3 km (mean) and 8 ± 3 km (median)).
Data on the frequency of cold-starts are scarce; the actual frequency depends, e.g.,
on the assumed parking duration, ambient temperature, operating conditions prior to
vehicle parking, and the vehicle-specific design of engine and after-treatment
technologies. Based on the data presented by De Gennaro et al. (2014), Donati et al.
(2015), Paffumi et al. (2015), Karlsson (2013), and HBEFA 3.2 (Kühlwein, 2016) and
under the assumption that a cold start occurs after vehicle parking of some 3 h to 8
h, we conclude that as a first-order approximation, 27 ± 5% of trips may contain
a cold start. The findings of Klein et al. (2015), according to which 60% of trips in
the Netherlands contain a cold start, appear to include parking times substantially
shorter than 3 h.
Averaging the data displayed in Columns 6 and 7 suggests that in Europe the
distance between two consecutive cold starts is 36 ± 16 km (mean) and 30
± 13 km (median). If only urban trips are considered, the distance between two
consecutive cold starts reaches 25 ± 16 km (mean) and 27 ± 8 km
(median).
The choice of a 3-8 h parking duration to differentiate between cold-start and warm-
start trips accounts (to some extent) for the cold-start definition in the RDE test
procedure. In real-word driving, also substantially shorter parking durations may
lead to a cold start, e.g., vehicles are driven over comparatively short trips or if
driving and parking occur at lower ambient temperatures that the 15oC assumed in
Figure 3. Additional research considering the actual cool down of after-treatment
technologies of a sample of vehicles is necessary to verify our results.
We consider our results to represent conservative estimates on the distance vehicle
users actually drive between two consecutive cold starts. More detailed scenario
analyses incorporating additional data could verify the conclusions of this report.
12
Calculated as the mean and the standard deviation of individual medians presented in Table 10.
36
Table 10: Overview of results; numbers in normal type setting obtained from the respective sources; numbers in italics
calculated by the authors of this report
Source
Country
Mean trip distance
[km]
Median trip
distance
[km]
Frequency of daily cold
start; percentage of trips
containing
cold starts(a)
Scenario
calculation:
Mean distance
between two
consecutive
cold starts
Scenario
calculation:
Median distance
between two
consecutive
cold starts
De Gennaro et al.
(2014); Paffumi et
al. (2015)
Italy (Modena and
Florence)
8 ± 3 (urban trips)
8.4
(urban trips)
20% (parking ≥6h)
30% (parking ≥3h, coolant
<30 oC)
40 ± 15
27 ± 8
42,30(c)
28, 20(c)
Donati et al.
(2015)
Various cities in 6
European countries
7.8 (urban trips,
electric)
4.8 (urban
trips, electric)
20% (parking ≥6h)
27% (parking ≥3h, coolant
<30 oC)
39
29
24
18
Klein et al. (2015)
The Netherlands
14.5 (all trips)
60%
24
Klein et al. (2015)
The Netherlands
4 (urban trips)
60%
7
Klein et al. (2015)
The Netherlands
12 (extra-urban
trips)
60%
20(b)
Karlsson (2013)
Sweden
12 ± 6
11(d)
25%
47 ± 25
42(d)
Steven (2016)
WLTP data base
10.5 (all trips)
20% (parking ≥6h)
27% (parking ≥3h, coolant
<30 oC)
53
39
ACEA (2016)
BMVI data for
Germany
≥2
≥19
Misc. data analysed
by JRC
Various European
countries
14 ± 5 (all trips)
12 (all trips)
20% (parking ≥6h)
27% (parking ≥3h, coolant
<30 oC)
70 ± 25
52 ± 19
60
44
T&E (2016)
Various European
countries
6 (urban trips)
20% (parking ≥6h)
27% (parking ≥3h, coolant
<30 oC)
30
22
HBEFA3.2
Various European
countries
8.3 ± 1.6
(all trips)
4.9
(all trips)
31% (mean, parking ≥8 h)
35% (median, parking ≥8 h)
27 ± 8
24 ± 5
16
14
(a) Entries in italics are assumed or calculated by the authors of this report.
(b) Klein et al. (2015) specify the number of cold starts per passenger car kilometre for rural roads to be approximately 0.005. Personal communication with
Ligterink (2016) suggests the actual number of cold starts per passenger kilometre is in fact 0.05, suggesting a driving distance between two cold-starts of 20
km.
(c) Calculated based on the frequency distribution displayed in Figure 2.
(d) Representing the median value of the mean driving distances covered by each individual car between parking durations of 6 h or longer.
37
5.2.2 Conclusions on the scenario analysis – cold-start inclusions
From the scenario analysis presented in Section 4.2, we draw the following conclusions:
Modifying the data pre-processing by duplicating or shifting cold-start emissions to
the middle of the urban part of a trip (Mod1a, Mod1b, Mod1c) generally increases
on average the overall urban NOX emissions but may also result for individual tests in
decreasing calculated emissions compared to the baseline scenario Mod0a
(exclusion of cold start). Taking into account that all vehicles included in our analysis
showed higher NOX emissions during cold start than during warm-engine operation,
we conclude that these proposed modifications of the RDE data pre-processing can
improve the coverage of cold start by the RDE data evaluation but are, by
themselves, insufficient to capture cold-start emissions in a consistent manner.
Modifying the weighting of moving averaging windows by assuming a constant or
linearly decreasing weighting factor for the first 300 windows (Mod2a, Mod2b)
likewise tends to increase on average the overall urban NOX emissions but may also
decrease the urban NOX emissions for individual trips compared to the baseline
scenario (Mod0a). For several tests, the modification of the weighting approach
amplified the emissions effect (i.e., the observed increase or decrease of the overall
urban NOX emissions) observed for the inclusion of cold-start inclusion into the
normal RDE data evaluation (scenario Mod0b). Moreover, applying a linearly
decreasing weighting factor (from 2 to 1 for the first 300 windows in scenario
Mod2b) amplified the emission effects observed for a constant weighting factor of 1
(Mod2a). Therefore, modifying the weighting of windows may at occasions increase
the coverage of cold start by the RDE data evaluation but is, by itself, insufficient to
capture cold-start emissions in a robust and consistent manner.
A combination of modifications in the data pre-processing (e.g., scenario Mod1c that
duplicates cold start) and the data evaluation (e.g., scenarios Mod2a and Mod2b
that apply a constant or linearly decreasing weighting factor for cold-start windows)
as it is modelled by scenarios Mod3a and Mod3b shows the highest responsiveness
to the actual cold-start emissions performance of vehicles. This observation suggests
that a combination of modifications in the data pre-processing and evaluation might
capture the cold-start emissions in a more robust manner than applying the
proposed modifications individually. Yet, our analysis suggests that even a
combination of the proposed modifications may not capture the incremental cold-
start emission of all vehicles. We thus propose to complement our analyses by
additional modification scenarios in the future.
Any of the analysed scenarios could be implemented without major modifications of
the existing provisions of Regulation 2016/427 (EC, 2016a).
38
6 References
ACEA (2016): ACEA RDE – Cold start. Status-update. 18 May 2016. Presentation given to
the RDE working group. ACEA – European Automobile Manufacturer’s Association.
Brussels, Belgium.
André, M., Hammarström, U., Reynaud, I. (1999): Driving statistics for the assessment of
air pollutant emissions from road transport. INRETS report, LTE9906, Bron, France,
pp. 191. http://www.inrets.fr/infos/cost319/index.html.
André, M., Joumard, R. (2005): Modelling of cold start excess emissions for passenger cars.
INRETS, Bron, France.
CBS (2016): Personenmobiliteit in Nederland; vervoerwijzen en reismotieven, regio’s. CBS –
Centraal Bureau voor de Statistiek. Source:
http://statline.cbs.nl/statweb/publication/?vw=t&dm=slnl&pa=83500ned&d1=a&d2=
0&d3=0-2,6&d4=0-1&d5=0,2&d6=4-5&hd=160923-
0938&hdr=t&stb=g1,g4,g3,g2,g5. Retrieved: 12 January 2017.
De Gennaro, M., Paffumi, E., Martini, G., Scholz, H. (2014): A pilot study to address the
travel behaviour and the usability of electric vehicles in two Italian provinces. Case
Studies on Transport Policy 2 (2014) 116–141.
Donati, A. V., Dilara, P., Thiel, C., Spadaro, A., Gkatzoflias, D., Drossinos, Y. (2015):
Individual mobility: From conventional to electric cars. Report JRC97690. European
Commission - Joint Research Centre Ispra, Italy.
EC (2007): Regulation 715/2007. EC – European Commission. Official Journal of the
European Union L 171, pp. 1-16.
EC (2015): Technical guidelines for the preparation of applications for the approval of
innovative technologies pursuant to Regulation (EC) No 443/2009 and Regulation
(EU) No 510/2011.
EC (2016a): Regulation 2016/427. EC – European Commission. Official Journal of the
European Union L 82, pp. 1-98.
EC (2016b): Regulation 2016/646. EC – European Commission. Official Journal of the
European Union L 109, pp. 1-22.
EC (2016c): Real driving emission (RDE) legislation: Next steps. Document presented at the
RDE meeting on 25 January 2016. EC – European Commission. Source:
https://circabc.europa.eu/faces/jsp/extension/wai/navigation/container.jsp?FormPrin
cipal:_idcl=FormPrincipal:_id1&FormPrincipal_SUBMIT=1&id=0171f779-9515-4cc2-
b9fc-
f18bb3ee398c&javax.faces.ViewState=OOvIr3rYicPfsLTwClPOLiNFH%2F8ao7ViKoT%
2F%2FhQ8XQLvBj6pBMzqdqrUXDGVnYHShxnd9mvS8Yg3gewgwI9ryxagKm1rJPixgBb
ZmTejPQCdnuBvFn5EFcYH65OKTo1jVsFoZ25Fioo%2FbpgQciSmBcPZqpzHZj5Zf9MnP
g%3D%3D. Retrieved: 9 December 2016.
EEA (2016): Air quality in Europe – 2016 report. EEA – European Environmental Agency.
Copenhagen, Denmark.
Karlsson S. (2013): The Swedish car movement data project: Final report. PRT report
2013:1 Rev2, Chalmers University of Technology, Gothenburg.
http://publications.lib.chalmers.se/records/fulltext/187380/local_187380.pdf.
Karlsson, S., Björnsson, L.-B. (2016): Personal communication with JRC.
39
Klein, J., Hulskotte, J., Ligterink, N., Molnár, H., Geilenkirchen, G. (2015): Methods for
calculating the emissions of transport in the Netherlands. Statistics Netherlands, PBL,
TNO. Source:
http://www.emissieregistratie.nl/erpubliek/documenten/Lucht%20%28Air%29/Verke
er%20en%20Vervoer%20%28Transport%29/Methoderapporten%20Taakgroep%20V
erkeer%20en%20Vervoer/Klein%20et%20al.%20%282015%29%20Methods%20for
%20calculating%20emission%20from%20transport%20in%20NL.pdf. Accessed: 29
July 2016.
Kühlwein, J. (2016): Data analysis of HBEFA 3.2, shared in personal communication.
Ligterink, N. E. (2016): Personal communication.
Paffumi, E., De Gennaro, M., Martini, G. Scholz, H. (2015): Assessment of the potential of
electric vehicles and charging strategies to meet urban mobility requirements.
Transportmetrica A: Transport Science, 11:1, 22-60, DOI:
10.1080/23249935.2014.913732
Pasaoglu et al. (2012): Driving and parking patterns of European car drivers – a mobility
survey. Source: https://setis.ec.europa.eu/sites/default/files/reports/Driving_and_
parking_patterns_of_European_car_drivers-a_mobility_survey.pdf. Accessed: 29 July
2016.
Steven, H. (2016): Personal communication.
T&E (2016a) T&E compilation of European data. Transport and Environment, Brussels,
Belgium.
T&E (2016b): T&E’s proposals for RDE 3rd package. Presentation given on 20 July 2016.
Transport and Environment, Brussels, Belgium.
UNECE (2015): Addendum 82: Regulation No. 83. Revision 5. UNECE – United Nations
Economic Commission for Europe. Geneva, Switzerland.
Europe Direct is a service to help you find answers to your questions about the European Union
Free phone number (*): 00 800 6 7 8 9 10 11
(*) Certain mobile telephone operators do not allow access to 00 800 numbers or these calls may be billed.
A great deal of additional information on the European Union is available on the Internet.
It can be accessed through the Europa server http://europa.eu
How to obtain EU publications
Our publications are available from EU Bookshop (http://bookshop.europa.eu),
where you can place an order with the sales agent of your choice.
The Publications Office has a worldwide network of sales agents.
You can obtain their contact details by sending a fax to (352) 29 29-42758.
XX-NA-xxxxx-EN-N
doi:10.2760/70237
ISBN 978-92-79-62229-8
KJ-NA-28472-EN-N
