*Summary of the study :*

*Overview and Analysis of the links between "Models of
Mobility" and "Models of Pollutant Emissions from Transport"*

*Benoit GILSON and Vincent FAVREL under the direction of
Dr Walter HECQ.**Study funded by the
European Commission under COST319 Action.*

*Pdf version of complete report will be soon
available.*

Overview of mobility models | Emission models | Linking emission models and mobility models

To modelise mobility demand, two main types of models are available. On the one hand, econometric models which explain a variable related to mobility analysis as a function of the most significant socio-economic variables. On the other hand, mobility models (often called network flow models) which modelise traffic flow on a transport network. These last models are more dis-aggregated and provide more detailed output than the econometric models.

Three econometric models are pointed. They differ following the
*dependant variable* studied : either an indicator of mobility, a fuel
consumption, or a variable related to car fleet. In the case of modelling
indicators of mobility, the dependent variable is a measure of the volume of
transport, e.g. number of passenger-kilometres, vehicle miles travelled, miles
travelled per car (or light trucks or lorries) etc, see Greene (1992). For fuel
consumption, the dependant variable is aggregated and can be expressed in
gasoline, diesel or LPG consumption per capita, per household or per vehicle,
see Epsey (1996). Finally, in the case of car fleet analysis, the dependant
variable can be the motorization rate (mean number of car per adult). The
methodology for the latter case is somewhat different than the previous models
as a demographic approach (longitudinal analysis) is used. This last approach is
based on household expenditure survey and panel data estimation techniques, see
Madre (1995).

All of these models consider income and price/cost as main
*explanatory variables*, introduced as exogenous inputs. Only a few other
socio-economic variables are taken into account in these models but they are
weakly significant (e.g. driven licence in the case of passenger-kilometre
analysis). Data (explanatory and dependant variables) are, in general, easily
available and predictable, but only on a broad aggregated (national)
scale.

The network is represented by links and nodes. Usually links are physical and logical (e.g. transfer between transport modes) connections. Two types of nodes are used : nodes representing a junction where three or more links meet or when a route changes its characteristics and the centroid nodes which represent origin zones and destination zones.

Each link of the network includes a start node, an end node and a link type. Different parameters can be coded on the link : distance, capacity, travel time/speed, even delays for custom formalities, etc. Representations of links for road, rail and air networks are depending on the level of detail covered by the model. O/D matrices represent the number of trips between each centroid node.

Classical transport models are made up of 4 main steps (sub-models): generation, distribution, modal choice and assignment steps. For a more detailed description, see Transport Research-APAS 22 studies.

The first step, which is called the * generation/attraction
model*, estimates the number of trips leaving a zone (generated trips)
and entering a zone (attracted trips) on the basis of socio-economic variables.
Passenger and freight transportation are handled separately.

The second step is the * distribution model* where
O/D matrices are built. These models estimate where the produced trips will go
to and where the attracted trip comes from.

The * modal choice model* constitutes the third step
of transport models where the share of trips following the transport mode used
is estimated. This results in the subdivision of the O/D matrices built in the
previous step into several sub-matrices, one for each transport mode.

The final step of the 4-step model is the * trip assignment
model* where route choices are modelled. Trips, calculated in the
previous steps, are assigned to a network. This results in a loaded network. The
outputs are calculated for each O/D pair as path flows, junction delays, O-D
travel costs. The assignment procedures can be either deterministic or
stochastic. Travellers choose paths which minimise their generalised cost (or
utility) functions (mainly the time parameter). In the stochastic case, a random
term is added to the assignment algorithm.

Finally, a * validation* procedure is often added.
Note that some models include exclusively the assignment phase taking O/D
matrices as exogenous input.

The four-step transport model scheme is used by the most
well-known modelling tools. APAS 22 give an overview of the strategic or
multi-national models available in the European Economic Area. The APAS database
describes 62 passenger models and 43 freight models, collected with several
criteria (e.g. for passenger models) : scale (part of country, one country, part
of Europe, European Union, Europe); area (in square km^{2}) ; scope
((part) of the home country, international); number of O/D matrices (for cars,
public transport, air, sea, bicycle, pedestrian, others); number of trip
purposes (0, 1, 2, ³ 2); number of zones (0-200,
22-500, 55-1000, >1000); number of links in a road network (no network,
0-5000, 5000-10000, 10000-50000, >50000); time basis (year/month, day,
morning peak, evening peak, day + peak, parts of the day average weekday);
etc..

One important aspect for emission assessment is that these models can infer average speed on the links in relation to the traffic flow (number of passenger cars per hour and number of lorries per hour). The speed-flow function depends on link characteristics : capacity, number of lanes, terrain characteristics (slopes, bends).

Concerning non road transportation, models for the calculation of non-road transport emissions have not been specifically considered in a linking perspective in this study.

Considering cold start emissions, apart from meteorological parameters and fuel properties, the data required that could possibly be supplied by mobility models concern : the average trip length per vehicle trip and the total annual kilometres driven by the vehicle of each category (referring to COPERT) or the distance travelled by the vehicle, the number of starts per day and per vehicle and parking duration before the trip (referring to the German/Swiss model).

COPERT II suggests a methodology for evaporative emission calculation. It requires many parameters that are, most of the time unavailable and have to be estimated. These parameters are the following : the fraction of trips finished with hot engines, the fraction of trips finished with cold engines or with the catalyst below its light-off temperature, the yearly average number of trips per vehicle per day and the total annual mileage of each vehicle category. Referring to the German/Swiss model and for the same purpose, other parameters have to be estimated : the number of times the engine is turned off, the frequency distribution of the travelled distance before the engine is turned off and the frequency distribution of the parking duration after the engine is turned off.

The aggregated character implies that we do not have modelisation on mobility per transport mode, per vehicle category, etc. Other econometric models could be built to split, for example, urban from non urban vehicles travelling, provided statistics are available. Further investigation would be requested to assess this possibility. The existing econometric models, which have been developed with other goals than emission assessment, partly satisfy the requirements of emission models, provided simplifying assumptions are made.

When based on fuel consumption, models do not differentiate the different fuel types. Once again, if data on total annual fuel consumption for each type of fuel can be available, models could be built on a time series basis. These models can be linked with emission models, such as COPERT II, which calculates the total annual fuel consumption as a calibration parameter for estimating uncertain parameters (e.g. average annual mileage driven on each road class and for each vehicle category).

Furthermore, econometric models could provide information on car fleet composition or motorisation rate, which is of great interest for all types of emissions, using age cohort models. But these models provide once again only aggregated information on the car fleet as a whole. In fact, emission models require not only the total number of cars per country/region but also the structure of the car fleet, i.e. : the share of diesel, gasoline and LPG cars, or the share of different vehicle cubic capacities and the age categories of each vehicle.

In particular, concerning cold start emissions, the number of starting operations is unknown. To find it, we can make the restrictive assumption that each trip leaving a zone is considered as a start. Concerning parking duration distribution before the trip, further information has been requested from mobility model developers in order to establish if this parameter can be provided one way or another. Cold start and evaporative emissions also depend on the outside temperature which can be different following the parking location of the vehicles (indoor, outdoor). This aspect is not considered in the models and requires additional data concerning the share of vehicles parked in an indoor heated parking. Up to now, neither mobility models nor emission models consider this aspect. New developments in cold start emission modelling [Sérié et Joumard, 1997] consider the driving pattern at the beginning of the trip using the average speed as additional data. This last parameter is available from the mobility models.

Concerning evaporative emission, the fraction of trips finished with hot engines and the fraction of trips finished with cold engines or with the catalyst below its light-off temperature, can be determined knowing the trip length distribution and the ambient temperature. The number of trips per vehicle and per day and the total annual mileage of the vehicle category can also be determined processing the output data of mobility models. The number of times the engine is turned off can be roughly estimated making the assumption that it is equal to the number of trips arriving at a zone.

As mentioned, matching problems arise while considering the linking emission and mobility models. Firstly, mobility models can only distinguish the share of kilometres driven by car, bus/coach and by truck. In order to reconcile them with emission models, two solutions are envisaged :

- to refine modal choice models by splitting existing modes into sub-categories, for instance, by splitting the O/D matrices for cars into sub-matrices differentiating car sub-categories (fuel types and technological concepts). It should be assessed to see the possible level of dis-aggregation that can be achieved and at which cost;
- to use statistical data on the car fleet and to weight the number of vehicles on each O/D pair per the share of the different vehicle categories including year per year considerations. This alternative can easily be operational but needs to assess the accuracy of the method.

- The COPERT II emission model only differentiates three road types (urban, rural and highway);
- The German/Swiss emission model differentiates for three basic road types more than 170 so-called "traffic situations" for the different vehicle categories;
- mobility models like STREAMS differentiate 9 road type links.