Modelling of traffic flow and air pollution emission with application to Hong Kong Island Liping Xia * , Yaping Shao Department of Physics and Materials Science, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon, Hong Kong SAR, China Received 22 November 2003; received in revised form 16 July 2004; accepted 9 August 2004 Abstract In this study, we propose a Lagrangian model for the simulation of traffic flow on a complex road network. This simple approach is quite efficient if adequate road network data are available and statistical constraints are applied to confine the model behavior. We have established a traffic information database for Hong Kong Island and applied the model for traffic flow simulation. It is shown that by specifying three types of traffic routes (random turn, preferred turn and shortest path) and providing traffic flow data at selected stations, the model is capable of simulating traffic flow on the road network. This is confirmed by comparing model simulated and observed traffic flow patterns at several monitoring stations. The simulated traffic flow is then used as the basis for the estimation of traffic induced emission of air pollutants on the island. Using empirical emission factors for a number of vehicle categories, the emission rates of major air pollutants, CO, NO x and PM 10, are estimated. The predicted emission rates are compared with measurements for several air quality monitoring stations. Ó 2004 Elsevier Ltd. All rights reserved. Keywords: Traffic; Traffic flow model; Road network; Traffic emission; Urban air pollution 1. Introduction Traffic generated air pollution is of great concern to the general public. Motor vehicles emit nitrogen oxides (NO x ), carbon monoxide (CO), volatile organic com- pounds (VOC) and particulate matter (PM), which constitute a major source of air pollution in large cities, such as Hong Kong. Traffic generated air pollutants, such as NO 2 and PM, are of health concern; and traffic generated greenhouse gases, such as carbon dioxide (CO 2 ), may contribute to global warming. As motor vehicles are the major contributor to urban air pollution, controlling strategies need to be developed that minimize the environmental impacts but maximize the efficiency of motorized transport. In order to provide a viable method for quantifying the contribution of traffic emission to regional air quality, we develop an integrated Traffic Emission Information System (TEIS) which allows the prediction of traffic induced air pollution in real-time. More details on TEIS are given in Section 3.4. As the key components of TEIS, the traffic flow model and traffic emission model are developed and presented in this study. The emission factor based approach is widely used in modelling traffic-related pollution emission (e.g. Salles et al., 1996; Mensink et al., 2000; Lin and Lin, 2002; Jensen et al., 2001). The accuracy of this approach depends very much on the reliability of traffic data (traffic volume and velocity, their temporal and spatial variations, on road vehicle composition etc.) and the choice of emission factors. The methodology to derive these two types of data is consequently critic to emission factor based modelling of traffic pollution emissions. * Corresponding author. Tel.: C852 27889482; fax: C852 27887830. E-mail address: [email protected](L. Xia). 1364-8152/$ - see front matter Ó 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.envsoft.2004.08.003 www.elsevier.com/locate/envsoft Environmental Modelling & Software 20 (2005) 1175–1188
Modelling of traffic flow and air pollution emissionwith application to Hong Kong Island
Liping Xia*, Yaping Shao
Department of Physics and Materials Science, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon, Hong Kong SAR, China
Received 22 November 2003; received in revised form 16 July 2004; accepted 9 August 2004
Abstract
In this study, we propose a Lagrangian model for the simulation of traffic flow on a complex road network. This simple approachis quite efficient if adequate road network data are available and statistical constraints are applied to confine the model behavior. Wehave established a traffic information database for Hong Kong Island and applied the model for traffic flow simulation. It is shown
that by specifying three types of traffic routes (random turn, preferred turn and shortest path) and providing traffic flow data atselected stations, the model is capable of simulating traffic flow on the road network. This is confirmed by comparing modelsimulated and observed traffic flow patterns at several monitoring stations. The simulated traffic flow is then used as the basis for theestimation of traffic induced emission of air pollutants on the island. Using empirical emission factors for a number of vehicle
categories, the emission rates of major air pollutants, CO, NOx and PM10, are estimated. The predicted emission rates are comparedwith measurements for several air quality monitoring stations.� 2004 Elsevier Ltd. All rights reserved.
Traffic generated air pollution is of great concern tothe general public. Motor vehicles emit nitrogen oxides(NOx), carbon monoxide (CO), volatile organic com-pounds (VOC) and particulate matter (PM), whichconstitute a major source of air pollution in large cities,such as Hong Kong. Traffic generated air pollutants,such as NO2 and PM, are of health concern; and trafficgenerated greenhouse gases, such as carbon dioxide(CO2), may contribute to global warming. As motorvehicles are the major contributor to urban airpollution, controlling strategies need to be developedthat minimize the environmental impacts but maximizethe efficiency of motorized transport.
1364-8152/$ - see front matter � 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.envsoft.2004.08.003
In order to provide a viable method for quantifyingthe contribution of traffic emission to regional airquality, we develop an integrated Traffic EmissionInformation System (TEIS) which allows the predictionof traffic induced air pollution in real-time. More detailson TEIS are given in Section 3.4. As the key componentsof TEIS, the traffic flow model and traffic emissionmodel are developed and presented in this study.
The emission factor based approach is widely used inmodelling traffic-related pollution emission (e.g. Salleset al., 1996; Mensink et al., 2000; Lin and Lin, 2002;Jensen et al., 2001). The accuracy of this approachdepends very much on the reliability of traffic data(traffic volume and velocity, their temporal and spatialvariations, on road vehicle composition etc.) and thechoice of emission factors. The methodology to derivethese two types of data is consequently critic toemission factor based modelling of traffic pollutionemissions.
1176 L. Xia, Y. Shao / Environmental Modelling & Software 20 (2005) 1175–1188
Traffic data are generally obtained by either in-situobservation or numerical modelling. The former mostaccurately reflects traffic conditions in real-time, but isusually carried out on selected road links only, e.g.highways and artery roads. The amount of observeddata is often insufficient for adequately quantifying thetraffic on a road network. Further, in-situ measurementsare usually done on a daily or even a monthly basis. Thistemporal resolution is insufficient for refined (usuallyhourly) emission modelling. A complement is to maketemporal and spatial extrapolation with many assump-tions to allocate traffic volume, e.g. Salles et al. (1996)and Jensen et al. (2001). Another approximate method-ology previously adopted, as pointed out by Cohen et al.(2004), is to distribute traffic emission over model gridcells, resulting in improper grid-based averaging emis-sion rate instead of that along actual mobile source. Linand Niemeier (1998) used observed traffic data toestimate hourly allocation factors and disaggregatedtraffic volume into hourly values. These indirectmethods inevitably lead to inaccuracies in emissionmodelling. In theory, numerical modelling of traffic flowon road can provide every detail required for thecalculation of traffic emissions. Unfortunately, previousefforts failed to do this because of road networkcomplexity and, as we will see below, difficulties insolving the traffic flow equations.
Continuum hydrodynamics was firstly introduced totraffic flow theory in the 1950s (Lighthill and Whiteman,1955). Prigogine and Herman (1971) applied statisticalmethods, as in classic fluid dynamics, to traffic flowstudies. The work of Prigogine and Herman, known asthe kinetic theory of traffic, considered vehicles on roadas interacting particles in traffic flow which can bedescribed by one-dimensional compressible fluid equa-tions. Suppose there is neither creation nor destructionof vehicles on road, the continuity equation and theequation of motion for traffic flow can be written as:
vr
vtCv
vr
vsZ0 ð1Þ
vv
vtCv
�vv
vs
�Z
1
r
�v
vs
�mvv
vs
�� vp
vs
�CI ð2Þ
where r is density (number of cars per unit road length),n is traffic flow velocity, m is viscosity, and p is localpressure. The first term on the right hand of Eq. (2)models viscosity, a presumed tendency to adjust vehiclespeed to that of the surrounding traffic (Nagatani, 1998).The last term I is all inner forces due to interactionbetween individual cars (Kerner and Konhauser, 1993).In practice, the continuum hydrodynamic approach isdifficult to implement for two reasons. One is that thequantities such as m, I and p are not well defined andcannot be readily determined, and the other is that the
numerical solution of Eqs. (1) and (2) requires theirdiscretization for complex road networks. The numer-ical treatments for the diffusion and advection terms arerather cumbersome.
As an alternative, some researchers establishedequilibrium relations between traffic density and trafficflow velocity for the closure of Eq. (1) instead of usingEq. (2). By definition, traffic flow is the product of trafficdensity and velocity. If traffic density is zero, then trafficflow is also zero; and when traffic density reaches themaximum, i.e., traffic is congested, traffic velocitydecreases to zero, so traffic flow is also zero. Newell(1993), Daganzo (1994) and Wong and Wong (2002)suggested piecewise-linear flow–density relationships.De Angelis (1999) studied nonlinear hydrodynamicmodelling of traffic flow in theory. The linear diffusionterm was taken into account in the governing equations.De Angelis found that a second order flow–densityrelation gives a satisfactory fitting to the experimentalresults of Leutzbach (1988). Critical analysis on a similarmodel but with additional phenomenological relationbetween density and velocity was presented by Bonzani(2000) and Marasco (2002). Velan and Florian (2002)explored the implications of nonsmooth equilibriumflow–density relationships. However, all these studieswere concerned with traffic flows on individual high-ways. We are not aware of traffic model applications tocomplex road networks.
Our approach is different. In contrast to thecontinuum hydrodynamic approach, we consider themotion of individual vehicles and determine the macro-scopic traffic flow quantities on the basis of vehiclemovement. Although the problem of traffic on networkis highly complicated, the movement of individualvehicles is quite simple. Vehicle movement is analogousto that of gaseous molecules. However, while moleculesmove randomly, vehicles are confined to the roadnetwork and follow certain designated paths. Hence,the movement of individual vehicles is predictable.
We are therefore motivated to track vehicles on roadnetwork using the Lagrangian methodology. Thisapproach requires no predefined velocity–density re-lationship. Instead, we introduce a critical traffic densityand two time scales. The motion of an individual vehicleis governed by a first-order ordinary differential equationwhich can be solved by using, for example, the Runge–Kutta method. Macroscopic traffic flow quantities, suchas traffic flow velocity and traffic density, can be esti-mated once the velocity and position of individualvehicles are known. The Lagrangian approach is verysimple in theory and involves little mathematical diffi-culties. However, we recognize that the implementationof such a model on a road network requires the knowl-edge of designated paths for individual vehicles. Fora given road network, we may be dealing with millionsof vehicles and it is impossible to determine the
designated paths for all vehicles. However, it is possibleto force the designated paths to comply with certainstatistical conditions.
Another difficulty in modelling traffic in practice isthe requirement for data, in order to quantify the roadnetwork, including road type, junctions, traffic lightsand the sources and sinks of vehicles. In TEIS, coupledwith a traffic GIS (geographic information system), theintegration of dynamic models with vehicle and roadnetwork data is achieved.
The traffic emission simulation presented in this studyis on the basis of the traffic flow simulation for HongKong Island. Since the emission factors specifically forHong Kong are yet to be determined, those derivedfrom COPERT II emission inventory programme(Ahlvik et al., 1997) are used in this study. COPERTII was recommended by the European EnvironmentAgent and widely adopted in Europe for emissionestimates from road transport (e.g. Mensink et al., 2000;Reynolds and Broderick, 2000).
The modelling results of traffic flow and majorpollutant emission rates in Hong Kong Island networkare compared with the traffic data obtained at severalcounting stations and air quality monitoring at roadsidestations, as presented in Section 4.
2. Model description
Several databases are established for the modelling.These include (i) Hong Kong Island road networkdatabase, in essence an attribution table for the specifi-cation of network connectivities, speed limits and roadclasses [roads in HongKong are classified into cataloguesof Tunnel (TUN), Main road (MRD), Secondary road(SRD) and Trail (TAL)]; (ii) vehicle database for thespecification of vehicle characteristics and probabilitydistribution of vehicle types; (iii) traffic emission factordatabase and (iv) air quality monitoring database.
2.1. Traffic flow model
According to traffic density, we introduce two trafficmodes on a road network: a free traffic mode anda congestion mode. Let rc be a critical traffic density,separating the free traffic mode from the congestionmode. If r! rc on a road segment, then traffic is in thefree traffic mode and vehicles would accelerate to a speedlimit vlim. If rO rc, then traffic is in congestion modeand vehicles would decelerate to zero speed. According-ly, the equation of motion for a vehicle can be written as:
dvidt
Z
( vlim � vita
r%rc
�vitb
rOrc
ð3Þ
where ta and tb are vehicle acceleration and decelerationresponse times, respectively. These response times aremainly functions of traffic density, as they do not differto a great degree among vehicles.
Suppose a road length is L and the minimum allowedseparation (on average) is D, then the maximum allowednumber of vehicles on L in free traffic mode is NZ L/D.Therefore, the critical traffic density rc is:
rcZN
LZ
1
Dð4Þ
Eq. (3) is a first-order ordinary differential equation,which can be easily solved (e.g. using the Runge–Kuttamethod), assuming initial traffic density r0j and speed v0jon road segment j and vlim,j are known.
By definition r is the number of vehicles per unit roadlength. Suppose we start counting vehicle number attime t on road segment j of length Dsj. If the countinginterval is dt, then over a time period Dt, m (ZDt/dt)counts N1, N2,.,Nm are made. The average trafficdensity over Dt at segment j is:
rjðtCDtÞZXmiZ1
Ni
Dsj=Dt
dtZ
dt
DsjDt
XmiZ1
Ni ð5Þ
Substituting rj into Eq. (3), the speed of car i on segmentj, vi;jZvi;j
�rj; tCDt
�can be determined. The overall
traffic speed Vj and traffic flow Qj on segment j during tto tCDt are:
VjZvi;j ð6Þ
QjZrjVj ð7Þ
This procedure is repeated for each vehicle on theroad network from start time t0 to end time T.Combining the determined trips with statistic con-straints to traffic assignment (see Section 3.2), trafficdensity, velocity and flow at each road segment in timeinterval Dt are simulated over the entire network.
2.2. The traffic emission model
Traffic emission rate is related to a number of vehiclecharacteristics: model, engine size, age, annual mileageby vehicle age and exhaust control equipment (Bachmanet al., 2000). To estimate the emission of a vehicle fleet,the vehicle population is divided into several categories.Seven categories are used in this study:
� Motor Cycle (MC): motor-propelled 2- or 3-wheeledvehicle;
1178 L. Xia, Y. Shao / Environmental Modelling & Software 20 (2005) 1175–1188
� Public Light Bus (PLB): passenger carrying vehiclewith capacity of 17 seats. CCO 2.01;
� Passenger Van (PV): dual purpose gasoline van withcapacity not exceeding 17 seats with weight less than3.5 t;
� Bus (BUS): diesel urban buses and coaches;� Light Goods Vehicle (LGV): four wheeled lorry ordual purpose van which is not provided with sidewindows covering the full length of the vehicle body,with weight less than 3.5 t;
� Heavy Goods Vehicle (HGV): lorry with more thanfour wheels, including fire engines, refuse vans,military trucks, petrol tanks and other similarvehicles, with weight exceeding 3.5 t.
For vehicle category i, the emission rate of pollutant jon road k is calculated by:
EijkZCijVik ð8Þ
where Vik is the traffic flow of vehicles type i on road kand Cij is the emission factor of pollutant j emitted byvehicles type i. Further, Vik can be expressed as:
VikZPikVk
where Pik and Vk are the fraction of vehicle type i andthe traffic flow of all vehicle types on road k,respectively. The emission rate of pollutant j of thevehicle fleet on road k can then be calculated by:
EjkZXn
iZ1
EijkZXn
iZ1
PikVkCijZVk
Xn
iZ1
PikCij ð9Þ
where Vk is calculated using the traffic flow model, andPik is estimated based on in-situ investigation conductedby local government departments. Table 1 shows theproportion of various vehicles by counting at the exit ofthree cross-harbour tunnels of Hong Kong in year 2000(HKTD, 2001). The values listed are averages over 16 h(0700–2300). In the traffic census of Hong Kong, thereare 203 counting stations in total on the island, althoughwe only listed data from three stations.
The emission factors Cij are estimated by COPERT IImethodology (Ahlvik et al., 1997). The reason we choseCOPERT II for Hong Kong is that Europe emissionstandards have been implemented in Hong Kong since1995 under Air Pollution Control Regulations. TheHong Kong government introduced the EURO I and
Table 1
Counted fraction (%) of vehicles in cross-harbour tunnels of Hong
Kong
Vehicle type MC PC, TX PV PLB BUS LGV HGV
Cross Harbor Tunnel 3.6 59.6 1.7 0.6 7.8 22.4 4.3
Eastern Tunnel 2.9 65.9 1.6 1.3 4.1 20.4 3.8
Western Tunnel 1.5 70.1 1.7 4.1 8.0 11.9 2.7
EURO II emission standards in 1995 and 1997,respectively. Furthermore, the government tightenedthe emission standards for newly registered motorvehicles (design weight less than 3.5 t) to EURO IIIlevel in 2001.
Three emission modes are taken into account forcalculation of emission factors: (i) hot emissions, theseare the emissions from vehicles after they have warmedup to their normal operating temperature; (ii) cold-startemissions, these are the emissions from vehicles whilethey are warming up and the water temperature is below70 �C; and (iii) evaporative emissions, these are associ-ated with the relevant quantities for gasoline vehicles inthe form of no-methane VOC (subtracting CH4 fromVOC) emissions.
The resultant hot emission factors of CO, NOx andPM10 adopted in this study are listed in Tables 2–4.‘Conventional’ vehicle category is applied for all except93/59/EEC for LGV in the form of NOx emission inTable 3 and 91/441/EEC for PLB in the form of CO andNOx emissions in Tables 2 and 3. The cold emissions aretaken into account as additional emissions per kilometerby introducing cold to hot ratio of emissions, ecold/ehot,and the fraction of mileage, b, driven with cold enginesor catalyst operated below the light-off temperature.They are the function of ambient temperature and theaverage trip length. The calculation formula for theseparameters can be found in Ahlvik et al. (1997).
3. Model application to Hong Kong
We have applied the models to the simulation oftraffic flow and traffic-related emission on Hong KongIsland. The road network on the island is shown inFig. 1, which is represented by over 6000 line features inArcGIS. The traffic on the island is an isolated systemwith only three harbour tunnels linking to the out-sidedi.e., Kowloon and the New Territory. Thissignificantly simplifies the simulation. The traffic dataobtained in the three tunnelsdfrom left to right inFig. 1, Western Tunnel, Cross Harbour Tunnel andEastern Tunneldare used as boundary conditions forthe modelling.
The year 2000 traffic data for Hong Kong Island,published in the Annual Traffic Census 2000 by theTransport Department of Hong Kong (HKTD), areused for specifying the initial and boundary conditionsof the model. The data are also used for modelvalidation.
3.1.1. Traffic counting stationsIn the Annual Traffic Census 2000, traffic flow is
surveyed at 806 counting stations over Hong Kong, ofwhich 203 are on Hong Kong Island. In this census, theroad network on the island is divided into fourcategories:
(a) Urban 1: areas corresponding more or less to thebusiness district;
(b) Urban 2 (major roads): areas including major roadlinks in other urban areas;
(c) Recreational: areas including much of the Peak andthe central part of the beach areas;
(d) Remote: South-eastern part of Hong Kong Island.
We chose three counting stations on the island(Fig. 1) for model comparison (Section 4.1), each ofwhich represents a traffic category:
(a) Station 1001: Harcourt Road, representing theUrban 1 category;
(b) Station 1002: Victoria Park Road, representing theUrban 2 category; and
(c) Station 1011: Repulse Bay Road & Stanley GapRoad, representing the Recreation category.
Table 3
Speed dependency of NOx emission factors
Vehicle
class
Vehicle
category
Speed V
[km h�1]
Emission factor [g km�1]
1 MC 10–60 0.00003V2� 0.002VC 0.064
60–110 �0.00002V2C 0.0049V� 0.157
2 PC, TX 10–130 0.000247V2C 0.0014VC 1.387
3 PV 10–130 0.000094V2� 0.0079VC 1.9391
4 PLB 10–130 0.0001015V2� 0.0107VC 0.4767
5 BUS 0–50 89.174V�0.5185
6 LGV 10–130 0.000127V2� 0.01674VC 0.9037
7 HGV!7.5 t 0–50 50.305V�0.7708
50–100 0.0014V2� 0.1737VC 7.5506
Table 4
Speed dependency of PM10 emission factors
Vehicle
class
Vehicle
category
Speed
[km h�1]
Emission factor [g km�1]
5 BUS 0–50 7.8609V�0.736
6 LGV 10–130 0.0000125V2� 0.000577VC 0.288
7 HGV 0–100 4.5563V�0.7070
The observed traffic flow data is divided into twomain groups: one represents the average traffic flow onworking days (Mon–Fri) and the other average trafficflow on weekends (Sat–Sun).
3.1.2. Traffic flow entering Hong Kong IslandFig. 2 shows the averaged hourly inward (south
bound) traffic flow on weekdays in three tunnels. Thereexists a morning (0800 and 0900) and an afternoon peakhour (1800 and 1900) in both Western and EasternTunnel, while a relative steady daytime flow appears inCross Harbour Tunnel from 0800 to 1900. All threetunnels present minimum inward flow at dawn. In total,the Cross Harbour Tunnel has the highest traffic flow(61,816 veh day�1), while the Western Tunnel has thelowest traffic flow (22,165 veh day�1) during weekdays.
Fig. 3 presents the average incoming traffic flowduring weekend in the three tunnels. The variationpattern for Cross Harbour Tunnel is similar toweekdays with the lowest traffic flow occurring around0500. The morning peak time is delayed until 1000. Infact, traffic flow remains high from 1000 to 1900 beforebecoming less at 2100. After that, a night peak hourappears from 2200 to 2300. The traffic in both WesternTunnel and Eastern Tunnel show similar patterns toCross Harbour Tunnel, but with much lower traffic flowduring daytime. Again, among the three tunnels, theCross Harbour Tunnel has the highest traffic flow(61,480 veh day�1) and the Western Tunnel has thelowest traffic flow (18,013 veh day�1) at weekends.
3.2. Vehicle behavior on network
Every vehicle on road has its own origin anddestination (O–D). The traveling route of the vehicledepends on the driver’s need and traffic conditions.Numerous traffic assignment models have been de-veloped aiming at determining the network flow patternsin order to provide route guides during times of concern(e.g. morning peak hours). A review of dynamic trafficassignment (DTA) models can be found in Peeta andZiliaskopoulos (2001). A discussion of DTA problemswas recently presented by Peeta and Yang (2003).Existing DTA models generally involve high complexity.In some studies, the problem is much simplified withspecifications (Papageorgiou, 1990) such as: steady-stateconditions (Sheffi, 1985); single destination (Sarachikand Ozguner, 1982; Wie, 1988; Ziliaskopoulos, 2000);fixed routes (D’Ans and Gazis, 1976) etc. The disad-vantage of the simplifications is that they do notadequately reflect vehicle behavior on road network.
To overcome the problem, we propose a combinationof deterministic and statistic constraints of three typicaltravel options for all vehiclesdrandom turning trip,preferred turning trip and shortest path. In this study,we assume the origin of all vehicles be one of the three
1180 L. Xia, Y. Shao / Environmental Modelling & Software 20 (2005) 1175–1188
Fig. 1. Hong Kong Island network and locations of traffic counting and AQM stations.
cross-harbour tunnels, ignoring the sources on HongKong Island.
3.2.1. Random turning tripIn this travel option, a vehicle travels randomly on
road, i.e., it turns randomly to any linked road whenencountering a traffic light or being at the intersection ofmultiple roads, only subject to traffic rules. Among allvehicles on a road network, only a small proportion ofvehicles adopts this option of travel. We assume suchproportion to be 10%. Fig. 4 shows an example of this
option. A vehicle origins from the Western Tunnel andends the trip on South Lane, Sham Wan, a no-throughroad.
3.2.2. Preferred turning tripIn this study, all roads are classified into four
categories: tunnels, main roads, secondary roads andtrails. For the preferred turning option, a vehicle at anintersection turns preferably to a higher class road. Weassume 20% of the vehicles on road behave this way.Fig. 5 shows the driving route of a vehicle making
0
500
1000
1500
2000
2500
3000
3500
4000
0 2 4 6 8 10 12 14 16 18 20 22 24t (hr)
Traffic flo
w (vh
/h
r)
Cross Harbour TunnelWestern TunnelEastern Tunnel
Fig. 2. Incoming traffic flow in tunnels on weekdays.
Fig. 3. Incoming traffic flow in tunnels during weekends.
a preferred turning trip. It also origins from the WesternTunnel and, after exit from the tunnel, it travels alongConnaught Road West toward Central.
3.2.3. Shortest path tripMost drivers make a trip as short as possible to reach
their destination. Our model is capable of finding theshortest path between the origin and destination ofa trip using the Dijkstra algorithm, which is based onthe Bellman optimality principle (e.g. Kreyszig, 1988).
Fig. 6 shows two shortest paths originating from theCross Harbour Tunnel. One destination is located atintersection of High Street and Western Street, Sai YingPun and another destination is at King’s Road, Taikoo
Shing. The shortest ways determined by the model arequite reasonable for given O–D.
3.3. Air quality monitoring
3.3.1. Site informationAir quality monitoring (AQM) in Hong Kong is
carried out regularly at monitoring stations by HongKong Environmental Protection Department (HKEPD).This network covers Hong Kong Island, Kowloonand New Territory. The monitored pollutants includesulphur dioxide (SO2), nitric oxide (NO), nitrogen di-oxide (NO2), carbon monoxide (CO), respirable sus-pended particulates (RSP) and ozone (O3). We use thehourly monitoring data of CO, NOx and RSP at three
Fig. 4. An example for random turning trip.
1182 L. Xia, Y. Shao / Environmental Modelling & Software 20 (2005) 1175–1188
Fig. 5. Typical preferred driving route.
roadside monitoring stations on Hong Kong Island forvalidating the model estimates.
The site information of three monitoring stations is asfollows (HKEPD, 2000):
(a) Central/Western: located in a residential area andthe sampling height is 18 m (4 floors) above ground;
(b) Causeway Bay: located in a busy commercial areaand the sampling height is 2 m above ground;
(c) Central: located in a busy commercial/financial areaand the sampling height is 4.5 m above ground.
The locations of the stations are shown in Fig. 1.
3.3.2. Observed pollution concentrationThe monitored hourly concentrations of CO, NOx
and RSP (PM10) at the three stations in year 2000 areplotted in Fig. 7. The diurnal variation of the concen-trations shows a degree of similarity with that of theworkday’s traffic flow observed in the same functiondistrict (Fig. 7d). The concentrations in Fig. 7a–cpresent the maximum values at 0800–0900 and 1700–1800, corresponding to the morning and afternoon peakhours of traffic in the area. This temporal feature isclearly observed at Central and Central/Western. AtCauseway Bay, although the morning maximum of CO(Fig. 7a) and the afternoon maximum of NOx (Fig. 7b)
Fig. 6. Typical shortest path between specific origin and destination.
Fig. 7. Monitoring concentrations of (a) CO, (b) NOx and (c) PM10. (d) Diurnal variation of traffic flow.
are not observed, another maximum is still notable inthe plots. The minimum concentrations of pollutantsappear at 0300–0500, when the traffic flow is at thelowest as well.
3.4. GIS and models
The models discussed in this study are the keycomponents of the Traffic Emission Information System
Fig. 8. Framework of TEIS.
1184 L. Xia, Y. Shao / Environmental Modelling & Software 20 (2005) 1175–1188
(TEIS). TEIS consists of three modules: a traffic flowmodel, a traffic emission model and a pollution disper-sion model. Together with the databases and post-processors, TEIS is integrated into a GIS framework.ArcGIS is used for the maintenance of model databaseand the visualization and analysis of model results.Fig. 8 illustrates the framework of the modelling system,in which the system database, including road network,territorial data, traffic features and vehicle character-istics as well as the meteorological and geographic data,and model output are stored, maintained and eventuallyvisualized and analyzed in ArcGIS. The externalmodels, traffic flow model, traffic emission model andair dispersion model coded in Fortran, are integratedinto GIS by ArcInfo AML.
4. Results and discussions
4.1. Simulation of traffic flow
4.1.1. Traffic boundary and initial conditionsAs Hong Kong Island is connected to the outside
only with three cross-harbour tunnels, all vehiclesentering our simulated domain are through the threetunnels. Therefore, the incoming traffic flow in thetunnels discussed in Section 3.1 is assigned as boundaryconditions for the simulation. Our aim is to simulatetraffic flow on network over 24-h time span. Thesimulation starts at 0:00 am, and the initial traffic flowon the entire network is assumed to be zero.
4.1.2. Traffic flow in tunnelsWe first examine the simulation of traffic flows in the
tunnels, at locations about 200–900 m away from thetunnel entries. Fig. 9a shows the simulated and observedtraffic flow on weekdays at the examination point in theCross Harbour Tunnel. The simulated traffic flow wellreproduces the observations. Also, the simulations forthe Eastern Tunnel and the Western Tunnel also showgood agreement with the observed traffic flow (Fig. 9band c).
4.1.3. Traffic flow at counting stationsWe also simulated 24-h traffic flows for weekdays
outside the tunnels. The simulated results are comparedwith traffic flow data obtained at three counting stationsdescribed in Section 3.1.
Fig. 10 shows the simulated traffic flow (normalizedby its maximum value) at counting stations forweekdays. The observed traffic flow is also plotted forcomparison. Station 1001 belongs to the Urban 1category. The daytime traffic flow during weekdays(Fig. 10a) is characterized by the morning and afternoonrush hours around 0800–0900 and 1700–1800. Fig. 10ashows that the simulation reproduces the observed 24-h
traffic pattern satisfactorily although the traffic flowfrom 1000 to 2400 is slightly underestimated.
A similar comparison for Station 1002 is shown inFig. 10b. Station 1002 falls into the Urban 2 category.The road links within this category are mainly used fortraveling to and from work on weekdays. They are alsoheavily used during weekends for recreational and socialactivities (HKTD, 2001). There are more obvious trafficpeaks observed between 0800 and 1700. The modelsuccessfully simulated the bi-peak structure of theobserved traffic flow although the first traffic peak at0800 is somewhat over predicted. Also, the simulatedtraffic flow at night (after 1800) is somewhat lower thanthe observed values. This discrepancy may be caused byignoring the sinks and sources on the network in thiswork, remaining as an important consideration toimprove the modelling performance in future.
The simulated and observed traffic flow at Station1011 is compared in Fig. 10c. Station 1011 falls into the
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
Traffic flo
w (vh
/h
r)
Traffic flo
w (vh
/h
r)
Traffic flo
w (vh
/h
r)
Simulated
Real-time
(a)
Simulated
Real-time
(b)
0500
1000
1500200025003000
35004000
0
500
1000
1500
2000
2500
3000
3500
4000
0
500
1000
1500
2000
2500Simulated
Real-time
(c)
Fig. 9. Simulated against real-time traffic flow in tunnels (weekdays):
(a) Cross Harbour Tunnel; (b) Eastern Tunnel; (c) Western Tunnel.
Recreational category. While most vehicles passingthrough this station on weekdays are also used forwork trips, the observed traffic flow presents a differentpattern in comparison to those at Stations 1001 and1002 (Fig. 10a and b). In contrast to Stations 1001 and1002, only one traffic peak is observed in the afternoon(1600) at Station 1011, while the simulated traffic flowpresents the peak values during both morning (0800)and afternoon (1600). It is shown in Fig. 10c that theobservation at this station is well produced by themodel.
4.1.4. Sensitivity testsModel simulations are also made for weekend cases.
As discussed in Section 3.1, traffic patterns in the three
0
0.2
0.4
0.6
0.8
1
1.2
No
rm
alized
traffic flo
wN
orm
alized
traffic flo
wN
orm
alized
traffic flo
w
Simulated
Real-time
(a)
0
0.2
0.4
0.6
0.8
1
1.2Simulated
Real-time
(c)
0
0.2
0.4
0.6
0.8
1
1.2
1.4Simulated
Read-time(b)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
Fig. 10. Simulated against real-time traffic flow at counting stations
(weekdays): (a) 1001; (b) 1002; (c) 1011.
tunnels during weekends differ from those observedduring weekdays. In particular, no obvious traffic peaksare observed in the Western Tunnel and the EasternTunnel during the daytime. As most traffic travel fordifferent purposes on weekends, different traffic patternsare also observed at the counting stations.
Fig. 11 shows the traffic flow patterns at threestations on weekends. Both observed and simulatedtraffic flow are plotted for comparison. Again, thesimulation reproduces the observed traffic pattern quitewell. Unlike the weekday cases, there is no obvioustraffic peak during daytime at Stations 1001 and 1002(Fig. 11a and b). This is understandable as there are lessbusiness trips to these areas on weekends. The peakhour at Station 1011 is at 1600 (Fig. 11c). Thisreasonably reflects the preferred time when the most
0
0.2
0.4
0.6
0.8
1
1.2
No
rmal
ized
tra
ffic
flo
w
0
0.2
0.4
0.6
0.8
1
1.2
No
rmal
ized
tra
ffic
flo
w
0
0.2
0.4
0.6
0.8
1
1.2
No
rmal
ized
tra
ffic
flo
w
Simulated
Real-time
(a)
Simulated
Real-time
(b)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
Simulated
Real-time
(c)
Fig. 11. Simulated and observed weekend traffic flows at three
counting stations: (a) 1001; (b) 1002; (c) 1011.
1186 L. Xia, Y. Shao / Environmental Modelling & Software 20 (2005) 1175–1188
traffic for recreational purposes are on roads to therecreational area.
4.2. Simulation of traffic emission
Traffic emission rates (kg h�1 km�1) are calculatedusing Eq. (9). The predicted diurnal variations of CO,NOx and PM10 at Central, Causeway Bay and Central/Western are plotted in Fig. 12. It is observed that thevariation of traffic flow in Fig. 7d is completely reflectedin the curves in Fig. 12. This means the linear relationbetween traffic flow and traffic induced emission rates.
Figs. 7a–c and 12a–c are not directly comparablebecause pollution concentrations are measured at themonitoring stations while the model predictions arepollution emission rates. Instead, we analyze the
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Em
issi
on
rat
e (k
g h
r-1 k
m-1
)E
mis
sio
n r
ate
(kg
hr-1
km
-1)
Em
issi
on
rat
e (k
g h
r-1 k
m-1
)
Central
Causeway Bay
(a)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Central
Causeway Bay
Central/Western
(b)
0
0.005
0.01
0.015
0.02
0.025
0.03
Central
Causeway Bay
Central/Western
(c)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
0 2 4 6 8 10 12 14 16 18 20 22 24
t (hr)
Fig. 12. Modelling emission rates of (a) CO, (b) NOx and (c) PM10.
correlation between the predicted emission rates andmeasured concentrations as a validation of the emissionmodel.
The predicted hourly emission rates (kg h�1 km�1)against the observed concentrations (mg m�3) of CO,NOx and PM10 at Central, Causeway Bay and Central/Western are shown in Figs. 13–15. The linear regressionequations and the correlation coefficients R2 are alsoshown in the charts.
Figs. 13–15 indicate that the observed hourlypollution concentrations have a close linear correlationwith the predicted traffic emissions at the three stations.For CO, the correlation coefficients, R2, at bothCauseway Bay and Central are larger than 0.8. The R2
of NOx at Central is as high as 0.86. While it issomewhat lower at Causeway Bay and Central/Western,it still has the value of 0.78 and 0.74. The R2 of PM10 atthe three stations are between 0.82 and 0.85. Thecorrelation coefficients are summarized in Table 5.These results confirm the good performance of boththe traffic flow model and the traffic emission model.
5. Conclusions
A Lagrangian traffic flow model and an emissionfactor based traffic-related air pollution emission modelhave been developed in this study. The traffic flow model
CO : Central
y = 0.0017x - 1.0037R2 = 0.8446
0
0.2
0.4
0.6
0.8
1
1.2
1.4
600 800 1000 1200
Concentration (µg m-3
)
Concentration (µg m-3
)
CO : Causeway Bay
y = 0.0015x - 1.4317R2 = 0.7997
0
0.2
0.4
0.6
0.8
1
1.2
1.4
900 1100 1300 1500 1700 1900Em
issio
n rate (kg
h
r-1 km
-1)
Em
is
sio
n rate (kg
h
r-1 km
-1)
Fig. 13. CO concentration vs. traffic emission rate.
is simple, but has been found to be quite efficient. Withthe specification of travel behavior, the model is capableof simulating traffic flow on a road network. The modelhas been applied successfully to Hong Kong Island. Thesimulated traffic flows in three cross-harbour tunnelsand at three counting stations on the island forweekdays and weekends have been compared withobservations. Good agreement has been found. Thetemporal variations of traffic flow in the cross-harbourtunnels and at the counting stations are reproduced bythe model at satisfactory level.
Using the simulated traffic flow and empirical vehicleemission factors, the hourly emission rates of CO, NOx
and PM10 are predicted and compared with thecorresponding pollution concentrations at three airquality monitoring stations. It is found througha correlation analysis that the two data sets are well
NOx : Central
y = 0.001x - 0.0056R2 = 0.8638
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 100 200 300 400 500 600
Concentration (µg m-3
)
Concentration (µg m-3
)
NOx : Causeway Bay
y = 0.0012x - 0.1745R2 = 0.7752
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 200 400 600
Concentration (µg m-3
)
800
NOx : Central/Western
y = 0.0006x - 0.0209R2 = 0.7408
0
0.02
0.04
0.06
0.08
0 50 100 150Em
issio
n rate (kg
h
r-1 km
-1)
Em
issio
n rate (kg
h
r-1 km
-1)
Em
issio
n rate (kg
h
r-1 km
-1)
Fig. 14. NOx concentration vs. traffic emission rate.
correlated. This shows that the emission factor basedapproach for the prediction of traffic induced pollutionemission in urban area is adequate.
In addition to providing traffic flow data for traffic-related pollution simulation, the traffic flow modelpresented in this work can also be used to predict thecongestion cases at select traffic black points due to suchimpacts as vast traffic amount, design faults of signalsystem and management.
PM : Central
y = 0.0002x - 0.0055R2 = 0.8482
0
0.002
0.004
0.006
0.008
0.01
20 40 60 80 100
PM : Causeway Bay
y = 0.0004x - 0.0208R2 = 0.8294
0
0.005
0.01
0.015
0.02
0.025
0.03
40 60 80 100 120 140 160
PM : Central/Western
y = 8E-05x - 0.003R2 = 0.8228
0.E+00
4.E-04
8.E-04
1.E-03
2.E-03
2.E-03
30 40 50 60 70
Concentration (µg m-3
)
Concentration (µg m-3
)
Concentration (µg m-3
)
Em
issio
n rate (kg
h
r-1 km
-1)
Em
issio
n rate (kg
h
r-1 km
-1)
Em
issio
n rate (kg
h
r-1 km
-1)
Fig. 15. PM10 concentration vs. traffic emission rate.
Table 5
Summary of correlation between hourly concentrations and traffic
emissions
Station CO NOx PM10
Causeway Bay 0.80 0.78 0.83
Central 0.84 0.86 0.85
Central/Western – 0.74 0.82
1188 L. Xia, Y. Shao / Environmental Modelling & Software 20 (2005) 1175–1188
Acknowledgements
This study is supported by the Strategic ResearchGrant of City University of Hong Kong (SRG7001254).Traffic flow data were provided by the TransportDepartment of Hong Kong and the air pollution datawere provided by the Environment Protection De-partment of Hong Kong.
References
Ahlvik, P., Eggleston, S., GoriBen, N., Hassel, D., Hickman, A.J.,
Joumard, R., Ntziachristos, L., Rijkeboer, R., Samaras, Z.,
Zierock, K.H., 1997. COPERT II Computer Programme to
Calculate Emissions from Road Transport. Technical Report No.
6 of European Environment Agency (EEA).
Bachman,W., Sarasua,W., Hallmark, S., Guensler, R., 2000.Modeling
regional mobile source emissions in a geographic information
system framework. Transportation Research Part C 8, 205–229.
Bonzani, I., 2000. Hydrodynamic models of traffic flow: Drivers’
behaviour and nonlinear diffusion. Mathematical Computer
Modeling 31, 1–8.
Cohen, J., Cook, R., Bailey, C.R., Carr, E., 2004. Relationship
between motor vehicle emissions of hazardous pollutants, roadway
proximity, and ambient concentrations in Portland, Oregon.
