Carbon Monoxide Emissions from Passenger Vehicles

Carbon Monoxide Emissions from Passenger Vehicles: Predictive Mapping with an application to

Hamilton, Canada

002 October 2004

D. Potoglou and P.S. Kanaroglou

Carbon Monoxide Emissions from Passenger Vehicles: Predictive Mapping with an application to Hamilton, Canada. D. Potoglou and P.S. Kanaroglou

Centre for Spatial Analysis - Working Paper Series

-1-

Abstract

Integrated urban models are designed to capture and simulate land-use and transportation interactions and to predict traffic volume and vehicle emissions at the link level of the urban transport network. As such, these models can address the weakness of existing systems. The Integrated Model of Urban LAnd-use and Transportation for Environmental analysis (IMULATE) is one of the operational urban models calibrated for the Hamilton Census Metropolitan Area (CMA). This paper extends IMULATE to include air pollution estimation and mapping of vehicle air pollutants, employing a dispersion model (CALINE-4) and spatial data analysis. The proposed approach provides an integrated framework for impact assessment of land-use and transport policies on traffic flows, emissions, and pollutant concentration, enabling the evaluation of population exposure to traffic related pollution. The study illustrates how vehicle-generated carbon monoxide (CO) concentration can be estimated and mapped using the proposed approach under a base-case scenario for the year 2006. Several development and transportation scenaria can be developed and “hot-spots” of traffic-originated air pollution can be identified and visualized within a GIS framework.

1. Introduction Recent evidence indicates that road traffic emissions are a major source of air pollution in urban areas with subsequent adverse human health effects (Faiz, 1993 ; Colvile et al., 2001). Although improvements in vehicle technology play a significant role in reducing traffic emissions at the source, air pollution abatement will remain a challenge because of increasing demand for transportation (Faiz, 1993 ; Colvile et al., 2001 ; World Business Council for Sustainable Development (WBCSD), 2001). Increase in the overall vehicle-kilometres-travelled (VKT) over the past two decades has outweighed any gains in emission reductions achieved through advances in car technologies (Kanaroglou and South, 1999 ; Anderson et al., 1996a). It has been suggested that the changing spatial arrangement of interacting activities in urban areas is mainly responsible for this effect (Kanaroglou and South, 1999). Land-use patterns, accessibility, land-use balance, mix and density have been found to be relevant to travel behaviour and to affect positively the level of VKT (Badoe and Miller, 2000). If planners are to implement policies that will reduce the need for travel in order to improve air quality in urban areas, then it is imperative not only to forecast the spatial pattern of traffic related air pollution, but also to quantify the link between land-use and the level of VKT within a single framework of analysis. Recent approaches for assessing and mapping traffic air pollution combine traffic, meteorology and dispersion of pollutants within an integrated software environment with mapping capabilities (Namdeo et al., 2002 ; McHugh et al., 1997b ; McHugh et al., 1997a). However, most approaches of this kind consider traffic as exogenous information and pay little or no attention to the land-use component (Gualtieri and Tartaglia, 1998 ; Jensen et al., 2001 ; Karppinen et al., 2000b ; Karppinen et al., 2000a). The allocation of land-use activities is highly related to travel demand and consequently, emissions and traffic air



-2-

pollution (Badoe and Miller, 2000). Hence, it is necessary to incorporate land-use and transportation interactions as an integral part of the traffic air quality assessment procedure. Also, for air pollution estimation and mapping, most software packages apply the Gaussian dispersion equation to the whole area of study. This results to low spatial resolution of air pollution estimates and high demand for computational power (European Commission, 1998). This paper demonstrates that the application of the Gaussian dispersion equation in conjunction with spatial data analysis provides plausible results, shortens computational time and potentially eliminates the need for programming. The contribution of this study is twofold. First, this paper extends the Integrated Model of Urban LAnd-use and Transportation for Environmental analysis (IMULATE) for estimating and mapping traffic air pollution. Second, dispersion modeling is integrated with spatial data analysis to estimate air pollution from traffic for the whole area of study. IMULATE estimates traffic flows, average speeds and emissions on a link-by-link basis for a typical morning peak period in five-year intervals. Estimated traffic volumes and emissions are used as input to the dispersion model CALINE-4 to estimate air pollution at specific point locations. Subsequently, a geostatistical interpolation technique, known as universal kriging, is employed to map the concentration over the whole area of study. The combination of models links characteristics of land use and the transportation network to the contribution of traffic to air pollution, opening the door to the evaluation of environmental impacts of various types of land-use and transport policies. Using the proposed framework, this study illustrates how vehicle-generated carbon monoxide (CO) concentration can be estimated and mapped under a base-case scenario for the year 2006. CO is the result of incomplete fuel combustion that characterizes mobile as opposed to stationary pollution sources and can be used as a marker for the contribution of traffic to air pollution (Fenger, 1999). The remainder of this paper is organized as follows; section 2 provides a detailed overview of the current advances regarding vehicle generated air pollution and mapping. Section 3 describes the steps involved in estimating CO concentration as applied to Hamilton Census Metropolitan Area (CMA). Section 4 presents and discusses the simulation results and finally, section 5 summarizes this work and suggests topics for further research.

2. Models for Assessing Traffic Air Pollution In the early nineties, research in urban transportation policy and planning recognized a unified approach, in which the transportation and land-use subsystems are interrelated (Kanaroglou and Scott, 2002). This modelling approach termed “land-use-transportation interaction models” or Integrated Urban Models (IUMs) implies that transport and land-use changes may affect each other, and both are related to the demand for automobiles (Bates, 2000). The prime objective of these models is to capture the important relationships in the urban system so that alternative policy decisions can be projected and studied in



-3-

advance (Kanaroglou and Scott, 2002). Using information of the land-use and transportation infrastructure, the cost of travel opportunities and socio-economic attributes of the agents, integrated urban models are the state-of-practice modelling tools in projecting land-use changes, traffic flows and average speeds. Following that, a methodology to quantify vehicle emissions progresses from vehicle activity, to mobile source emissions models and mobile emission source inventories (Sawyer et al., 2000). Although there are several operational IUMs, a limited number of them have incorporated modules for the assessment of environmental impacts from traffic. On the other hand, several urban air quality models have attempted to incorporate land-use and travel demand considerations within an integrative framework. Some of the most recent such attempts are outlined in Table 1. Table 1. Some integrated and empirically applied integrated urban air models

Although the land-use and transportation interaction is usually absent, several projects (mainly in Europe) have developed sophisticated applications for the dispersion of air pollutants, air quality and exposure assessment from motorized traffic (Gualtieri and Tartaglia, 1998 ; Jensen et al., 2001 ; Karppinen et al., 2000b ; Karppinen et al., 2000a). The main drawback of these approaches is the exogenous treatment of traffic flows, rendering the potential for long-term forecasting limited at best. Implementation of such an approach usually integrates a series of traffic, emission and dispersion models within a commercially available Geographic Information System (GIS), thus enhancing the output of the models with mapping capabilities and allowing for scenaria development. The following discusses some of the basic features of these integrated air quality models. ADMS-Urban. ADMS-Urban (McHugh et al., 1997b ; McHugh et al., 1997a) is an urban air quality assessment and forecasting package, which takes into account the full range of emission source types, such as road traffic, industrial, commercial and domestic emissions.

Model/Package Useful References Application

ADMS-Urban McHugh et al. (1997a and b) A range of cities in the UK and China

GUALTIERY AND TARTAGLIA

Gualtieri and Tartalia (1998) Florence, Italy

SPARTACUS European Commission (1998)

European Union, EC, DGXII

UDM Karppinen et al. (2000 a and b), Kousa et al. (2001 and 2002), Kukkonen et al. (2001)

Helsinki, Metropolitan Area, Finland

AirGIS Jensen et al. (2001) Middelfart, Denmark TEMMS Namdeo et al. (2002) Leeds, UK



-4-

Also, it is fully integrated with a GIS allowing user-friendly emission set-up, output presentation and analysis (van den Hout and Larsen, 2003). GUALTIERI and TARTAGLIA. Using a simple approach, Gualtieri and Tartaglia (1998) developed a modelling package for the estimation of traffic air pollution. The package operates within a GIS for information management, computations and air pollution mapping. Traffic flows of road-vehicles are estimated using a fixed origin-destination (O-D) matrix and the stochastic user - equilibrium traffic assignment algorithm (Sheffi, 1985). Emission factors are a function of vehicle class. Finally, the dispersion of pollutants (i.e., NOx, HC and CO) is based on a modified Gaussian dispersion equation. SPARTACUS. The SPARTACUS modeling system (European Commission, 1998) extended the integrated land-use and transportation model MEPLAN with additional modules for pollutant concentration estimation, population exposure to air pollution and noise. First, MEPLAN estimates a static traffic condition (e.g., peak hour or 24-hour average) within a land-use sub-model, taking account of the urban spatial economic system (Southworth, 1995). Then, to assess air quality and population exposure, SPARTACUS uses a grid-based approach, in which the study area is divided into a finite number of raster-cells. The centre of each cell contains information on traffic characteristics, derived from the disaggregation of the road network. Based on emission functions for different types of vehicles, the emission rate of a pollutant is calculated for each cell. The emission rates allocated to the centre of the raster cell are then fed into a Gaussian dispersion model to calculate concentrations of pollutants (i.e., Nitrogen Oxides (NOx), CO, Particulate Matter (PM)) (European Commission, 1998). UDM. The UDM system for integrated urban quality has been developed by the Finish Meteorological Institute (Karppinen et al., 2000b ; Karppinen et al., 2000a). UDM accounts for both stationary and mobile sources. To estimate traffic volumes and average speeds, UDM uses the EMME/2 transportation planning system. Traffic volumes are projected for future years using aggregated data collected in 1988. Estimated traffic volumes and average speeds are used as input to the LIISA emissions system, which has been calibrated for the Helsinki vehicle fleet. Subsequently, the finite-line source dispersion model CAR-FMI (Kukkonen et al., 2001) uses the Gaussian diffusion equation to estimate concentration of pollutants at a specific location. UDM predictions have been compared with ground measurements of Nitrogen Oxides (NOx) and Nitrogen Dioxide (NO2) in the Helsinki Metropolitan Area (Kousa et al., 2001). The results show that the modelling system tends to under-predict the measured concentrations in convective atmospheric conditions, and over-predict in stable conditions. Finally, UDM applies a mathematical model to determine human exposure to ambient air pollution in an urban area (Kousa et al., 2002). The exposure model combines predicted concentrations, the location of the population, and the time spent at home, at work and other places of activity.



-5-

AIR-GIS. AirGis estimates ambient air pollution levels, and enables mapping of traffic emissions, air quality levels and human exposures (Jensen et al., 2001). Traffic emissions are estimated using average-speed emission factors per vehicle class and hourly traffic loads derived from traffic data measurements at selected locations in the study area. TEMMS. TEMMS is another integrated package for modelling and mapping urban traffic emissions and air quality, which accounts for both stationary and mobile sources. To estimate traffic flows, TEMMS interfaces with the transportation model SATURN (Simulation and Assignment of Traffic to Urban Road Network) (Namdeo et al., 2002). SATURN estimates traffic volumes for each link of the road network assuming a fixed trip matrix. Link based emissions are estimated using ROADFAC, which is an integral emissions model for line sources. Estimated emissions are used as input to the ADMS-Urban dispersion model. Within TEMMS, these models have been integrated - via a database exchanger and a graphical user interface (GUI) - with the MapInfo GIS to allow for modelling and mapping of vehicle emissions and grid-based air quality. In summary, the majority of these approaches involving vehicular-emissions calculate the dispersion of a pollutant applying the Gaussian dispersion equation for the whole area of study. Alternatively, one can apply the dispersion equation to calculate the concentration at specific receptor points. This is the case of CALINE-4 dispersion model. As this paper demonstrates, the estimation of the concentration for the complete area of the study region can be performed using spatial data analysis techniques. With the development of user-friendly GIS and spatial data analysis tools, such as the ArcGIS 8.1 - Geostatistical Analyst (Johnston et al., 2001), estimations require less computer power and computational time, as well as minimal programming. Spatial data analysis techniques for spatially continuous data (e.g., air pollution) rely on the similarity of nearby observations in order to generate a surface. For the purposes of this application, we apply the universal kriging geostatistical technique, which allows for the estimation of a continuous pollution surface and the assessment of the uncertainty (error) of the predictions (Johnston et al., 2001).



-6-

3. Framework and Methodology for Assessing Traffic Air Pollution This section provides a detailed discussion of the operational framework and methodologies for the estimation of a CO pollution surface from vehicular traffic over the Hamilton CMA. Figure 1 graphically depicts the entire procedure for such estimation.

ESTIMATION OF CO AT POINT LOCATIONS

DISPERSION OF POLLUTANTS(CALINE4)

LAND-USE, TRAVEL DEMAND & EMISSIONS(IMULATE)

EMPLOC

TRANDEM

TRAFFICASSIGNMENT

TRAVELTIMES

MOBILE EMISSIONSMOBILE5.C

ESTIMATION OF COOVER THE STUDY AREA

POLLUTION MAPPING(UNIVERSAL KRIGING )

POPMOB

Figure 1. Operational framework for assessing traffic air pollution

In a nutshell, the approach follows three major steps. First, using the transportation component of the integrated urban model (IMULATE), we derive traffic loads on all the links of the transportation network for the morning peak period of a typical day (6:00 - 9:00 am). That component of IMULATE is interfaced to the MOBILE5C emissions model, allowing the automatic translation of traffic volumes into CO emissions. Second, using a generalized form of the existing road network and the results from the first step (i.e., traffic volumes and emission factors per link), the one-hour average CO concentration is estimated at point locations (receptors) using the CALINE-4 dispersion model. Finally, the universal kriging geostatistical technique is used to create the probabilistic map of the concentration of the pollutant over the Hamilton CMA.

3.1 Estimation of Vehicle Emissions IMULATE is an operational integrated urban model calibrated for the Hamilton

CMA, Ontario, Canada (Anderson et al., 1994). The model simulates changes in the spatial distribution of urban households and inter-zonal work, school and discretionary travel. Based on these trip projections and the estimation of the transportation mode choice of trip makers, IMULATE estimates traffic flows and average speeds on a link-by-link basis. Recently, IMULATE has been embedded within a graphical user interface, which employs



-7-

state-of-the art mapping components with the help of the package MapObjects (Buliung et al., 2004 ; ESRI, 1996). This software environment can manage, manipulate and analyze spatial and non-spatial data, using familiar GIS and windows controls (figure 2).

Figure 2. The IMULATE graphical user interface

As shown in Figure 1, IMULATE consists of four sub-models (Anderson et al., 1994 ; Buliung et al., 2004): (a) POPMOB is the land use component, which handles the intra-urban migration and place of work assignment of the resident workforce within a system of 151 census tracts of the Hamilton CMA. A recent modification of the land use module includes an employment location model (EMPLOC), which in combination with POPMOB facilitates the derivation of a place of residence/place of work matrix at the level of the census tract (Maoh et al., 2002). (b) TRANDEM, through a modal split module, estimates the number of trips related to work, school, and discretionary activities by mode. (c) The TRAFFIC ASSIGNMENT sub-model uses a stochastic user equilibrium algorithm to assign inter-zonal automobile trips to the vector model of the road network. (d) The MOBILE EMISSIONS extension accepts the information (i.e., traffic flows and average speeds) from the TRAFFIC ASSIGNMENT sub-model to estimate CO, hydrocarbons (HC) and NOx per link (Anderson et al., 1996b).

MOBILE5.C is the Canadian version of MOBILE5 emissions model, developed by the U.S. Environmental Protection Agency (E.P.A.). MOBILE5 is an average-speed emissions model. For each vehicle class, estimation of emissions depends on average speed on each link, the ambient temperature and vehicle characteristics. A previous study by Anderson et al. (1996) reports on the generation of a full set of emission factors for HC, CO and NOx, specifically for the purposes of IMULATE. Figure 3 presents the resulting CO emissions as a function of cruising speed for an average gasoline vehicle.



-8-

0.000

0.100

0.200

0.300

0.400

0.500

0 20 40 60 80 100

Speed (Km/hr)C

O e

mis

sion

s (K

gr)

Figure 3. CO emissions factors with cruising speed

3.2 Dispersion Modelling CALINE-4 is a line source air quality model developed by the California Department of Transportation (Caltrans). The model is based on a Gaussian diffusion model and employs the mixing zone concept to estimate dispersion of CO near roadways, accounting for a host of parameters such as source strength (i.e., traffic volume per link and emission factors), meteorology and site geometry. The model allows the calculation of CO concentration at predefined point locations within 500m of the roadway (Benson, 1989). CALINE-4 divides an individual link into a series of elements and calculates incremental concentrations for each element. The summation of incremental concentrations forms the total concentration estimate at a specific receptor. Figure 4 depicts the estimation method.

Figure 4. Element series represented by series of equivalent FLS (Benson, 1989) Each element is modelled as an “equivalent” finite line source (FLS) positioned normal to the wind direction and centred at the element’s midpoint. At this stage, the model makes two assumptions: (a) A FLS incorporates the emissions of the corresponding element and (b) the dispersion of the emissions takes place in a Gaussian manner downwind of the

WIND DIRECTION

y

x Receptor

YE

FET

Gaussian Plume

Plume Centerline FET: Receptor Fetch

YE: Plume Centerline Offset

PHI

εo



-9-

element. Equation [1] calculates the concentration at a receptor from an infinitesimal FLS segment dy (Benson, 1989):

σ+−+

σ−−⋅

σ−

⋅σ⋅σ⋅⋅π

⋅= 2

z

2

2z

2

2y

2

zy 2)Hz(exp

2)Hz(exp

2y

expu2

dyqdC [1]

where, dC is the incremental concentration of CO in grams/m 3, q is the linear source strength in grams, u is the wind speed in meters per second, H is source height in meters, z is the height of the receptor in meters and σy, σz are horizontal and vertical dispersion parameters, respectively. The meteorological parameters taken into account by CALINE-4 include: wind direction (degrees), wind speed (meters/sec), wind direction standard deviation, atmospheric stability class (scale 1-7) and ambient temperatures (degree Celsius). The above meteorological parameters are highly dependent on the time-period of the day and the type of the geographic location (e.g., coastal, mountain) (Coe et al., 1998). Emission factors are expressed in grams per mile per vehicle. For relatively inert pollutants, such as CO, emissions factors are directly proportional to the predicted concentrations. On the other hand, a vehicle-induced heat flux component alters the one-to-one relationship between traffic volumes (vehicles per hour) and CO concentration. Initially, the model computes all concentrations in mass per unit volume (µg/m3). Results are converted to a volumetric equivalent (i.e., parts per million (ppm)) for gaseous pollutants. The conversion is achieved by multiplying the concentration values in µg/m3 by the FPPM factors, which account for the effects of both temperature and pressure on the volumetric concentration.

⋅

⋅⋅=T

ALT03417.0exp273T

MOWT02241.0FPPM [2]

where, MOWT is the molecular weight of the pollutant, T is the temperature (oK) and ALT is the altitude in meters (see Benson (1989)).

3.3 Geostatistical Analysis- Universal Kriging The estimated values of CO concentration from CALINE-4 are used as input in modeling the variability and mapping of CO over the Hamilton CMA. The ultimate objective is to estimate CO concentration at locations other than the receptors. For this purpose, we employ a geostatistical technique known as universal kriging. In general, kriging is a widely used interpolation technique for spatially continuous data (e.g., air pollution). Among the different types of kriging, ordinary and universal kriging are the most commonly used techniques. The general concept is that the prediction of the



-10-

value )(Y s (i.e., CO concentration) at any location s (spatial coordinates) is obtained as a weighted average of neighbouring data (Bailey and Gatrell, 1995 ; Kanaroglou et al., 2002):

∑=

=n

1i

)(Y)()(Y ii ssws [3]

In both ordinary and universal kriging, the ultimate goal is to estimate the optimal values of the weights w i(s), i = 1, 2, ..., n. This is accomplished by selecting the weights so that the expected mean square error becomes as small as possible (Kanaroglou et al., 2002). One can show that in matrix notation, the solution to the minimization problem is:

(s)cC(s)w 1+

−++ = [4]

The problem formulation, using equations [3] and [4], is the same for both ordinary and universal kriging. The difference between the two methods is in the assumptions made about the mean value (global trend) of the variable under study. Ordinary kriging assumes that the mean is known or that the data have been detrended. Universal kriging takes account of spatial variability in the means by way of some polynomial function of the spatial coordinates. The method is more applicable in the presence of a strong spatial trend in the measurements (Krivoruchko and Gotway, 2004). This difference requires different definition of the matrices in equation [4]. For ordinary kriging, the weight matrix w+(s) constitutes a column vector with n+1 elements. The first n elements are the weights to be estimated, while the (n+1)th element is a Lagrange multiplier that enters, because of the minimization of the mean square error (Kanaroglou et al., 2002). c+(s) denotes also a column vector of n+1 elements. The first n of these elements are the covariances C(s, si) between the prediction point s and each of the n sample sites (i.e., CALINE-4 estimates), while the last element equals to 1. Finally, C+ is the augmented variance-covariance matrix of the observations:

=+

0111)(),(

1),()(

1

11

LL

MMOML

nn

n

sVssC

ssCsV

C

In contrast, the matrices for universal kriging are augmented by the independent variables (i.e., spatial coordinates) that are used to estimate the global trend. For both cases the inverse matrix of C+ is obtained, as required by equation (4), and it can be used for all prediction points. In general, ordinary kriging is preferred when there is no global trend. In cases where exploratory analysis of the data indicates a strong trend, universal kriging is more likely applicable. Exploratory analysis of the estimated CO concentrations from



-11-

CALINE-4, indicated a strong trend along the northeast direction, making necessary the use of universal kriging. The notation in equation [4] indicates that the estimation of the weight matrix is dependent on covariances. Estimates of those covariances are obtained with the help of a semi-variogram function (e.g., spherical, gaussian) (Kanaroglou et al., 2002 ; Bailey and Gatrell, 1995). The semi-variogram (or variogram, for short) is one of the basic requirements for kriging and it is advised that its estimation should include the whole study area. Good kriging estimates are sensitive to the choice of an appropriate variogram and one needs to experiment thoroughly before the selection of a final variogram. One of the major advantages of kriging is the statistical evaluation of the results and the estimation of confidence intervals around the predicted values. For this purpose, the mean square prediction error or kriging variance is used:

)s(cC)s(c 1T22e +

−++−σ=σ [5]

where the variance σ2 is derived from the known covariance structure. The mean square prediction error is essentially the minimized mean prediction error when the estimated weight values are substituted in it. Also, cross-validation is another method of evaluating the results of kriging. This works by eliminating a given observation and estimating a predicted value at that location by applying the same kriging methodology to the rest of the observations. This process is repeated for all observations in turn. The analysis of the resulting differences between observed and predicted values (errors) provides an opportunity for the evaluation of the interpolation results (Kanaroglou et al., 2002).

3.4 Study Area The Hamilton CMA, located on the west shore of Lake Ontario approximately 100 km southwest from the city of Toronto, has a population of roughly 600,000 mostly concentrated on the southern shore of Hamilton Harbour (figure 5).

Figure 5. The Hamilton Census Metropolitan Area



-12-

For travel demand modelling purposes, the metropolitan area is divided into 151 census tracts. The road network consists of highways such as QEW, freeways (403) and arterial roads with at least two lanes. It is worth mentioning that more than 51% of the road network has more than two lanes. The vector model of the road network used in this study includes approximately 1500 links and 1100 nodes.

4. Results and Discussion Through the graphical user interface of IMULATE, the user can specify and run simulations in five-year intervals, starting from the base year 1986. Policy simulations can be undertaken by altering land-use and transportation parameters, such as population and residential density, employment location as well as network characteristics and performance. For the purpose of illustration, a base-case scenario is developed for the year 2006. A base-case scenario projects past trends into the future. To develop such a scenario, IMULATE uses two types of exogenous information: (a) The Hamilton CMA population projections for each simulation time-interval, and (b) Estimates of the potential of census tracts for residential development in terms of housing units. Under the base-case scenario for the year 2006, Figure 6 shows CO emission estimates on a link-by-link basis during the morning peak-period (Anderson et al., 1996b).

Figure 6. Carbon Monoxide emission estimates on a link-by-link basis in the year 2006

QEW Expressway

Highway 403

Highway 20



-13-

Maximum emissions of CO occur along the routes that lead to highway 403 and QEW expressway. Specifically, the estimated emissions on QEW expressway, highway 403 and their connections exceed the 800 Kgr CO per Km, during the peak-hour period. The majority of trips appear to be generated from Ancaster and the “Hamilton Mountain” areas, which, based on present trends, experience high rates of residential development.

4.1 Estimation of CO concentration CALINE-4 is capable of processing a maximum of twenty links and twenty receptor-points on each run. To overcome this limitation and to conduct analysis for the greater part of the Hamilton CMA, we make use of a generalized form of the road network compared to the IMULATE output. The basic rule of this generalization is the unification of links that belong on the same arterial road and form a straight line. In addition to that, the estimation procedure excludes links with emissions rates lower than 10 Kgr/Km. These links can be omitted because they have emission rates lower than one or two orders of magnitude, compared to other links on the network. Following that, the average values of traffic flows and emission factors are assigned to each link of the generalized network. Also, the study area is divided into five zones on the basis of development characteristics. This way, in addition to including more links in the analysis, the runs of CALINE-4 become consistent with the specifications of the dispersion model. Thus, we can define explicitly the Central Business District (CBD) and suburban areas. Table 2 summarizes zonal divisions and the basic characteristics of each CALINE-4 run. Also, figure 7 shows the generalized model of the road network set-up on CALINE-4 and the location of the receptors over the Hamilton CMA.

[Table 2, about here]

Figure 7. Receptor locations and CALINE-4 zones over the Hamilton CMA

Zone 1

Zone 2

Zone 3

Zone 4

Zone 5



-14-

Within CALINE-4, differences in the landscape are reflected by the Aerodynamic Roughness Coefficient, which determines the amount of local air turbulence affecting the plume spreading. The coefficient takes the value of 100cm for suburban areas and 400cm for CBD’s (Coe et al., 1998). The Hamilton CMA is generally characterized as a flat area, where street canyons and complex topography are absent. An exception to that is the area "Hamilton Mountain" (i.e., Zone 2, see figure 7), which is part of the Niagara escarpment, with a maximum height of 250 meters. These observations allow us to make the assumption of homogenous wind and temperature field under all runs (Table 3). Table 3. Caline-4 meteorological and run conditions

Parameter Value

Wind Speed (m/s) 3 Wind Direction (degrees) 0 Wind Direction Std. Dev. (degrees) 15 Atmospheric Stability Class (1-7) 7 Mixing Height (m) 40 Ambient Temperature (degrees oC) (average in January)

-5

Ambient Pollutant Concentration (ppm) 0 The output of the dispersion model is one-hour average CO concentration at the receptors. Assuming that the non-vehicle-originated ambient (background) concentration of CO is 0 ppm, the estimation of CO concentration at the receptors corresponds to vehicle-generated emissions. Based on the previous considerations, five separate runs, one per zone, were conducted using CALINE-4. The proportional map in Figure 8 shows the estimated values of CO concentration (in ppm) at the ninety-seven receptor locations.

Figure 8. CO estimates (ppm) at receptors’ locations

QEW Expressway

Highway 403

Highway



-15-

As expected, the concentration of CO is proportional to the emission-rate on the road network. Maximum CO concentrations occur on the connection of QEW and highway 403 in Burlington, along the QEW, the “Hamilton Mountain” and the city of Hamilton. Also, a pattern of high levels of CO appears along the highway 20.

4.2 Surface Map of CO concentration The estimation of the CO concentration, using universal kriging, requires the calibration of a model variogram (e.g., spherical, exponential or Gaussian). Figure 9(a) shows that the omni-directional variogram increases with distance, failing to reach an upper level or sill, as it is theoretically expected (Bailey and Gatrell, 1995). This indicates that first order effects are present, requiring the removal of the general trend. To remove the general trend, we fit a second degree trend surface model. Through transformation and adjustment procedures, a spherical variogram approximates better the underlying process and it is fitted to the empirical variogram in figure 9(b).

Figure 9. (a) Empirical omni-directional and (b) fitted-directional variogram model The general form of a spherical model variogram is given by the following form (Bailey and Gatrell, 1995):

σ

=

≤<

−σ+α

=γ

otherwise0hfor0

rh0forr2

hr2h3

)h(2

3

32

[6]

where, h is the distance between two point-observations, σ2 is the sill, r is the range and α is the nugget effect. The estimated theoretical variogram of the CO concentration data has the following form:

0 2000 4000 6000 8000 10000distance

0.0

0.2

0.4

0.6

0.8

gam

ma

objective = 0.0457

0 5000 10000 15000 20000distance

0.0

0.2

0.4

0.6

0.8

1.0

1.2

gam

ma

(a) (b)



-16-

=

≤<

−+

=γ

otherwise77.0

0hfor0

rh0for)7.9924(*2

h7.9924*2

h377.040.0

)h(

3

3

[7]

Equation [7] is applied to equation [4] in order to estimate the weights w+(s) and consequently the CO concentration from equation [3]. Figure 10 displays CO concentration over the Hamilton CMA using the universal kriging geostatistical interpolation.

Figure 10. Carbon monoxide concentration (ppm) using universal kriging

Three main “hot-spot” areas of maximum concentration can be identified in the CO pollution map (figure 10). The first two occur at the northern and eastern exists of the city of Hamilton providing access to highway 403 and the connection of highway 403 and QEW, respectively. These highways constitute main routes to the city of Toronto, a major employment centre for a significant portion of the Hamilton’s residents. The other “hot spot” is the area around highway 20, which is a major by-pass for drivers travelling from the southern part of the city to QEW highway towards Toronto. From a technical point of view, the interpolation of a finite number of point estimates (approximately 100), using universal kriging, is a significantly faster approach compared to numerical estimation. Using an IBM Xeon IntelliStation (512MB RAM) the estimation of the variogram and the generation of the CO concentration surface take no longer than 2 minutes of computational time.



-17-

Table 4 summarizes the prediction errors of the universal kriging estimates, derived with the help of equation (5). Table 4. Prediction errors of universal kriging

Prediction Error Value

Mean -0.037 Root-Mean-Square 1.502 Average Standard Error 2.175

Mean Standardized -0.099

Root-Mean-Square Standardized 0.937 The value of mean prediction error (-0.037) being close to zero, indicates that the predicted values are unbiased. Similar information is provided by the mean standardized prediction error (-0.099). Also, the average standard error (2.175) is greater than the root-mean-square of

prediction errors ( 2eσ ) (1.502). This shows that our model slightly over-estimates the

variability of CO concentration. The root-mean square prediction error (or kriging standard deviation) is a measure of the error that occurs when predicting data from point observations and provides the means for deriving confidence intervals for the predictions (Kanaroglou et al., 2002). Finally, the root-mean-square standardized (0.937) prediction error is very close to one, and thus corresponds to a very good fit between the point estimates of CALINE-4 and the geostatistical model using universal kriging.

5. Conclusions In both developed and developing countries traffic emissions are major contributors to poor urban air quality (Fenger, 1999). A number of recent applications have been specifically developed to assess the contribution of traffic to air pollution. The common approach has been to consider traffic as exogenous information that has loose or no connection with land-use activities. However, it has been recognized that assessment of the problem requires an integrated approach to account for the interaction between land-use and transportation (Badoe and Miller, 2000). Integrated urban models are sophisticated decision support systems that account for this interaction. Recent theoretical and technological advances have enabled the development of modules (e.g., estimation of vehicle emissions) within sophisticated software applications and user-friendly computer environments. An example is the operational integrated urban model IMULATE, calibrated for the city of Hamilton CMA (Buliung et al., 2004). IMULATE simulates land-use and transportation interactions and forecasts traffic flows, average speeds and vehicle emissions in five-year intervals during a typical morning-peak period for all the links of a simplified transport network. This paper describes a method that



-18-

provides the first step towards extending this system to include a module for vehicular air pollution estimation (i.e., carbon monoxide) and mapping using the CALINE-4 dispersion model and spatial data analysis. Forecasted traffic volumes and emission factors are used as input to CALINE-4, which estimates hourly average CO concentrations at predefined point locations. Subsequently, universal kriging is used to estimate and map the concentration of the pollutant over the study area. For illustration purposes, vehicle-generated carbon monoxide concentration is estimated and mapped under a base-case scenario for the year 2006. The approach proposed in this study provides an integrated environment, which not only simulates land-use and transportation interactions and estimates vehicle emissions, but also assesses the contribution of traffic to urban air quality. Several planning and policy scenarios can be developed and "hot-spots" of traffic -originated air pollution can be identified and visualized within a GIS framework. This application shows that line source dispersion models in conjunction with geostatistical models allow for rigorous calculations, lowering the demand for computer power and computational time. A next step in this research is to assess the performance of the developed system under various scenario results for which have been produced by IMULATE. A further area of research would be the automation of the whole process. This would include the extension of IMULATE to estimate and automatically map the concentration of a pollutant over the study area. This automation will require at the least the development of a variogram that represents accurately the process of pollutant dispersion. Using this variogram, an algorithm would perform concentration calculations using the procedure developed in this paper. In this context, IMULATE would have the capability to assess the environmental impacts of a series of transportation infrastructure and land-use change scenarios. Furthermore, such a tool would be suitable for the development and assessment of air quality guidelines and regulations. A potential extension to the system is the estimation of concentration and mapping of other vehicle-originated pollutants such as PM and NOx. Finally, in parallel to this extension, the estimates of the proposed approach would be compared with the results of other air pollution models.

Acknowledgements The first author wishes to thank the “Alexander S. Onassis” benefit foundation for its financial support. The second author gratefully acknowledges the support of the Social Sciences and Humanities Research Council (SSHRC) and the Canada Research Chair (CRC) program.



-19-

References Anderson, W. P., Kanaroglou, P. S. and Miller, E. J. (1994) Integrated Land Use and

Transportation Model for Energy and Environmental Analysis: A Report on Design and Implementation, McMaster University, Canada, Hamilton.

Anderson, W. P., Kanaroglou, P. S. and Miller, E. J. (1996a) Urban form, energy and the environment: A review of issues, evidence and policy Urban Studies, 33, 7-35.

Anderson, W. P., Kanaroglou, P. S., Miller, E. J. and Buliung, R. N. (1996b) Simulating automobile emissions in an integrated urban model Transportation Research Record, 1520, 71-80.

Badoe, D. A. and Miller, E. J. (2000) Transportation-land use interaction: empirical findings in North America, and their implications for modeling Transportation Research Part D, 5, 235-263.

Bailey, T. C. and Gatrell, A. C. (1995) Interactive Spatial Data Analysis, Longman Group Limited, Essex, England.

Bates, J. (2000) History of demand modelling In Handbook of Transport Modelling (Eds., Hensher, D.A. and Button, K.J.) Elsevier Science Ltd., pp. 11-33.

Benson, P. (1989) CALINE4 - A Dispersion Model for Predicting Air Pollutant Concentrations Near Roadways, California Department of Transportation, Sacramento, California.

Buliung, R. N., Kanaroglou, P. and Maoh, H. (2004) GIS, objects and integrated urban models In The behavioural foundations of integrated land-use and transportation models: Assumptions and new conceptual frameworks (Eds, .Lee-Gosselin, M. and Doherty, S.T.).

Coe, D. L., Eisinger, D. S., Prouty, J. D. and Kear, T. (1998) User's Guide for CL4: A User-Friendly Interface for the CALINE4 Model for Transportation Project Impact Assessments, Caltrans-U.C. Davis Air Quality Project, Sacramento, California.

Colvile, R. N., Hutchinson, E. J., Mindell, J. S. and Warren, R. F. (2001) The transport sector as a source of air pollution Atmospheric Environment, 35, 1537-1565.

E.P.A. http://www.epa.gov/otaq/m5.htm ESRI (1996) Building application with Map Objects, Environmental Systems Research

Institute, Redlands CA. European Commission (1998) System for Planning and Research in Towns and Cities for Urban

Sustainability (SPARTACUS), EC Environment and Climate Research Programme (1994 - 1998): Research Theme 4, EC DGXII.

Faiz, A. (1993) Automotive emissions in developing countries - relative implications for global warming, acidification and urban air quality Transportation Research A, 27, 167-186.

Fenger, J. (1999) Urban air quality Atmospheric Environment, 33, 4877-4900. Gualtieri, G. and Tartaglia, A. (1998) Predicting urban traffic air pollution: A GIS

framework Transportation Research Part D, 3, 329-336. Jensen, S. S., Berkovicz, R., Hansen, H. S. and Hertel, O. (2001) A Danish decision-support

GIS tool for management of urban air quality and human exposures Transportation Research Part D, 6, 229-241.

Johnston, K., Ver Hoef, J. M., Krivoruchko, K. and Lucas, N. (2001) Using ArcGis Geostatistical Analyst, Environmental Systems Research Institute, Redlands CA.



-20-

Kanaroglou, P. S. and Scott, D. M. (2002) Integrated urban transportation and land-use models for policy analysis In Governing Cities on the Move: Functional and Management Perspectives on Transformations of Urban Infrastructure in European Agglomerations. (Ed, Dijst, M.) Ashgate Publishing Ltd., pp. 43-75.

Kanaroglou, P. S., Soulakellis, N. A. and Sifakis, N. I. (2002) Improvement of satellite derived pollution maps with the use of a geostatistical interpolation method Journal of Geographical Systems, 4, 193-208.

Kanaroglou, P. S. and South, R. (1999) Can Urban Form Affect Transportation Energy Use and Emissions? An Analysis of Potential Growth Patterns for the Hamilton Census Metropolitan Area Energy Studies Review, 9, 22-40.

Karppinen, A., Kukkonen, J., Elolahde, T., Konttinen, M. and Koskentalo, T. (2000a) A modelling system for predicting air pollution: comparison of model predictions with the data of an urban measurement network in Helsinki Atmospheric Environment, 34, 3735-3743.

Karppinen, A., Kukkonen, J., Elolahde, T., Konttinen, M., Koskentalo, T. and Rantakrans, E. (2000b) A modelling system for predicting urban air pollution: model description and applications in the Helsinki metropolitan area Atmospheric Environment, 34, 3723-3733.

Kousa, A., Kukkonen, J., Karppinen, A., Aarnio, P. and Koskentalo, T. (2001) Statistical and diagnostic evaluation of a new-generation urban dispersion modelling system against an extensive dataset in the Helsinki area Atmospheric Environment, 35, 4617-4628.

Kousa, A., Kukkonen, J., Karppinen, A., Aarnio, P. and Koskentalo, T. (2002) A model for evaluating the population exposure to ambient air pollution in an urban area Atmospheric Environment, 36, 2109-2119.

Krivoruchko, K. and Gotway, C. A. (2004) Creating exposure maps using kriging Public Health GIS News and Information, 56, 11-15.

Kukkonen, J., Harkonen, J., Walden, J., Karppinen, A. and Lusa, K. (2001) Evaluation of the CAR-FMI model against measurements near a major road Atmospheric Environment, 35, 949-960.

Maoh, H., Kanaroglou, P. S. and Buliung, R. N. (2002) Modelling the location of firms within an Integrated Transport and Land-use model for Hamilton, Ontario In International Colloquium 1: Behavioural Foundations of Integrated Land-use and Transportation Models: Assumptions and New Conceptual FrameworksQuebec City, 16-19 June 2002.

McHugh, C. A., Carruthers, D. J. and Edmunds, H. A. (1997a) ADMS and ADMS-Urban International Journal of Environment and Pollution, 8, 437-440.

McHugh, C. A., Carruthers, D. J. and Edmunds, H. A. (1997b) ADMS-Urban: an air quality management system for traffic, domestic and industrial pollution International Journal of Environment and Pollution, Vol. 8, 666-675.

Namdeo, A., Mitchell, G. and Dixon, R. (2002) TEMMS: an integrated package for modelling and mapping urban traffic emissions and air quality Environmental Modelling and Software, 17, 179-190.



-21-

Sawyer, R. F., Harley, R. A., Cadle, S. H., Norbeck, J. M., Slott, R. and Bravo, H. A. (2000) Mobile sources critical review: 1998 NARSTO assessment Atmospheric Environment, 34, 2161-2181.

Sheffi, Y. (1985) Urban Transportation Networks: Equilibrium Analysis with Mathematical Programming Methods, Prentice Hall, Inc., New Jersey.

Southworth, F. (1995) A technical review of urban land use - transportation models as tools for evaluating vehicle travel reduction strategies, Center for Transportation Analysis, Energy Division.

van den Hout, D. and Larsen, S. (2003) Chapter 9: Integrated Urban Air Quality In SATURN: Studying Atmospheric Pollution in Urban Areas (Ed, Moussiopoulos, N.).

World Business Council for Sustainable Development (WBCSD) (2001) Mobility 2001 - World Mobility at the end of the Twentieth Century and its Sustainability, Geneva, Switzerland.

Date post:	29-Nov-2023
Category:	Documents
Upload:	independent
View:	0 times
Download:	0 times

Carbon Monoxide Emissions from Passenger Vehicles

Documents