Line source emission modelling S.M.S. Nagendra, Mukesh Khare*
Department of Civil Engineering, Indian Institute of Technology, Hauz Khas, New Delhi 110016, India
Received 6 September 2001; accepted 22 January 2002
Abstract
Line source emission modelling is an important tool in control and management of vehicular exhaust emissions (VEEs) in urban environment. The US Environmental Protection Agency and many other research institutes have developed a number of line source models (LSMs) to describe temporal and spatial distribution of VEEs on roadways. Most of these models are either deterministic and/or statistical in nature. This paper presents a review of LSMs used in carrying out dispersion studies of VEEs, based on deterministic, numerical, statistical and artificial neural network techniques. The limitations associated with deterministic and statistical approach are also discussed.
Keywords: Vehicular pollutant; Deterministic models; Numerical models; Statistical models; Multilayer perceptron; Model limitations
Contents
2083 1. Introduction 2083 2. Theoretical approaches of LSEM 2084 3. Line source deterministic models 2086 4. Line source numerical models 2089 5. Line source stochastic models 2090 6. ANN-based LSMs 2091 7. Limitations of LSMs 2091 8. Conclusion 2093 References 2093
1. Introduction transportation demand (Sharma and Khare 2001a; Mayer, 1999). Line source emission modelling (LSEM) In recent years, in most of the countries, the air is an important tool in screening of VEEs and helps in pollution from industrial and domestic sources has control and management of urban air quality. The US markedly decreased due to passage of various acts by Environmental Protection Agency (EPA) and many different governments. However, there has been a other research institutes have developed a number of line substantial increase of air pollution caused by the source models (LSMs) for estimating vehicular pollutant vehicular exhaust emissions (VEEs) due to addition of concentrations. All these models involve deterministic more and more vehicles on roadways to meet increase in and/or stochastic approach which, at present, are most widely used. In the recent past, artificial neural network (ANN) and in particular, multilayer perceptron (MLP) has also been applied in modelling the line source dispersion phenomena. Sharma and Khare (2001a) 2084 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 described various VEEs modelling studies in the domain Considering the above assumption, the basic ap- of analytical modelling techniques—deterministic, nu- proach to develop deterministic LSM is the coordinate merical and statistical. The present review is aimed at transformation between the wind coordinate system readers with little or no understanding of LSM (X1 ; Y1;Z1) and line source coordinate system (X; Y; Z). techniques. It is designed to act as a guide through the Let the length of the roadway be 'L', which makes an literature so that the readers may be able to appreciate angle 'y' with the wind vector. The middle point of the these techniques. The review is divided into several line source can be assumed as the origin for both sections, beginning with a brief introduction to the LSM coordinate systems, which also have the same Z-axis. approaches and followed by relevant LSM studies based The line source is along ' Y-axis' and the wind vector is in on deterministic, numerical, statistical and ANN tech- the X1 direction. In the line source coordinate system, all niques. Some of the common practical problems and parameters, viz., X; Y; Z and L are known from the limitations associated with deterministic, numerical, road geometry. A hypothetical line source is assumed to statistical and ANN techniques have also been dis- exist along Y1 that makes the wind direction perpendi- cussed. cular to it (Fig. 1). The concentration at the receptor is given by Csanday (1972):
\{Z-H 2. Theoretical approaches of LSEM c=-. "2
The deterministic mathematical models (DMM) H calculate the pollutant concentrations from emission +exp
C=- exp ifZ-h
Wind coordinate system +exp sinc?(L/2- F)-Xc erf I lay Line source sinyðL=2þ YÞ þ Xcosy Line source coordinate system þerf (2) laY where y is the angle between roadway and wind Receptor direction, ue — u sin y Þ U0; U0 is the wind speed correc- (x,y) tion due to traffic wake (Chock, 1978), h0 = H þ Hp; Hp is the plume rise. Numerical LSMs also come under the deterministic Fig. 1. Orientation of line source and wind direction coordinate modelling technique which are based on numerical system. approximation of partial differential equations S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 2085 representing atmospheric dispersion phenomena. First- interconnected adaptive processing units. These net- order closure models, also called K-models, have their works are fine-grained parallel implementation of non- common roots in the atmospheric diffusion equation linear static or dynamic systems. The important feature derived by using a K-theory approximation for the of these networks is their adaptive nature, where closure of the turbulent diffusion equation. These 'learning by example replaces programming' in solving models are time dependent and are applied through problems. This feature makes such computational computer software; eulerian grid, lagrangian trajectory, models very appealing in application domains where hybrid of eulerian-lagrangian and random walk particle one has little or incomplete understanding of the trajectory approaches are the commonly used techni- problem to be solved but where training data are readily ques. The detailed methodology of development of available. Using readily available inputs, the neural- numerical LSMs is available elsewhere (Rezler, 1989). network model automatically develops its own internal In contrast to deterministic modelling, the statistical model and subsequently predicts the output. For LSEM, models calculate concentrations by statistical methods the MLP structure of the neural networks seems to be from meteorological and traffic parameters after an the most suitable for predicting VEEs (Moseholm et al., ‘ appropriate statistical relationship has been obtained 1996). ‘ The multilayer perceptron consists of a system empirically from measured concentrations. Regression, of layered interconnected 'neurons' or 'nodes', as multiple regression and time-series technique are some illustrated in Fig. 2. The neural network model, shown key methods in statistical modelling. The time-series in Fig. 2, represents a nonlinear mapping between an ’ analysis techniques (Box-Jenkins (B-J) models) have input vector and output vector ’ (Gardner and Dorling, been widely used to describe the dispersion of VEEs at 1998). The 'nodes' are arranged to form an input layer, trafficked intersection and at busy roads. Auto-regres- one or more 'hidden' layers, and an output layer with sive integrated moving averages (ARIMA), ARIMA nodes in each layer connected to all nodes in neighbour- with exogenous inputs (ARMAX) and transfer function ing layers (Comrie, 1997). The input layer 'neurons' noise (TFN) algorithm have been adopted in LSM serve as buffer that distribute input signals to the next studies (Sharma, 1998). The Box and Jenkins (1976, layer, which is a hidden layer. Each 'neuron' in the 1970) models are empirical models created from the hidden layer sums its input signal, processes it with historical data and it is important that the iterative simple nonlinear transfer or 'activation function' (e.g., model building process proposed by Box and Jenkins is logistic and hyperbolic tangent) and distributes the always followed. Further, ARIMA models do not result to the output layer. The 'neurons' in the output specifically distinguish the physical causes of dispersion layer compute their output signal in a similar manner. phenomena (e.g. meteorological variables, emission The output signals from each neuron in an MLP rates of the sources, etc.) in their input. Such models propagate in forward direction; therefore MLP is also represent the 'Black Box' approach. All possible called as 'feed-forward' neural network (Haykin, 2001). uncertainties of the model are taken into account by a By selecting a suitable set of connecting weights and 'noise' variable with assigned statistical properties (Juda, transfer functions, it has been shown that an MLP can 1986). The TFN model is a dynamic model describing approximate any smooth, measurable function between the dependent variable as a response to the 'impulses' of the independent variables, with the latter playing the role of time-dependent forcing functions in an ordinary penden linear differential equation. The characteristics of the iffer response are described by the impulse response func- Input lser tions. The technique is quite general and useful in Output handling multivariate time series. It specifically builds layer into the model dynamics of impulse and response that is capable of describing a wide range of physical phenom- Input '2 ena (in linear regime). One important feature of TFN values Output modelling is that, because of its ability to control eatureo statistically the past variations that are common in two time series, it is a very good way to avoid spurious correlations and true causality in time series (Milionis and Davis, 1994a). ANN is a kind of statistical modelling technique offering several advantages over traditional phenomen- i=[ii,i2,ijl = input vector o = [o1,o2] = output vector ological or semi-empirical models, since they require 2 = •#• = transfer function or activation function known input data set without any assumptions. ANNs Fig. 2. A MLP with a hidden layer (Gardner and Dorling, are parallel computational models comprised of densely 1998). 2086 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 input and output vectors (Hornik et al., 1989; roadside receptor increases marginally as wind direction Gardner and Dorling, 1998). MLP has the ability becomes parallel to the highway. Dabberdt et al. (1973) to learn through training. A supervised back-propaga- presented a practical multipurpose urban diffusion tion algorithm is commonly employed in the training model (APRAC-1A) for predicting inert vehicular of MLP (Haykin, 2001). 'Training requires a set of pollutant concentration. The model requires routinely data, which consists of a series of input and associated available meteorological and traffic parameters for output vectors. During training, the multilayer percep- prediction of concentration isopleths, sequential hourly tron is repeatedly presented with training data and values and frequency distributions. A model for the the weights in the network are adjusted until the diffusion of pollutants from a line source in an urban desired input output mapping occurs. During training, atmosphere was also developed by Sharma and Myrup the output from the multilayer perceptron is compared (1975). This study revealed that wind shear (variation of with the desired output. If the network output is not wind with height) was responsible for turbulent diffu- matched with desired output, an error signal is sion in lower atmosphere. Based on the finite length propagated back through the network. Training approximation, Stukel et al. (1975) formulated a line uses the magnitude of these error signals to adjust the source dispersion model for estimating particulate/ weights and this process continues till the network gaseous pollutant concentration in urban roadways. In output matches the desired output' (Gardner and another study, Nicholson (1975) presented a scalar Dorling, 1998). budget box diffusion model for prediction of CO concentration in street canyons. Fay and King (1975) formulated a Gaussian model, considering vehicle- 3. Line source deterministic models induced effects on dispersion of pollutants. This model assumed that near the road, vehicle wake-induced Historically, as far as modelling of VEE is concerned, turbulence dominated over atmospheric turbulence. the work of Sutton (1932) may be regarded as the first of Therefore, dispersion of pollutants was assumed to be its kind. One of the early studies on deterministic independent of atmospheric parameters except wind vehicular pollution modelling was reported in Waller speed and dependent upon the drag characteristics of et al. (1965). The analytical method for estimating the passing vehicles. pollution levels from motor vehicles in the vicinity of The General Motor (GM) corporation experimental highways of common geometric configuration was data reported by Cadle et al. (1976) were the earliest field developed by Chen and March (1971). The preliminary experimental data used for understanding traffic influ- computational examples indicated that this method was ences on adjacent roadways. These data were taken over capable of representing, in a realistic manner, of all the a simulated test track of four lane free way of 5 km long physical variations accounted in the derivation. Dilley at the GM proving grounds in Milford, Michigan, USA. and Yen (1971) derived an analytical solution to a two- Chock (1977a) conducted a number of experiments to dimensional transport and diffusion equation that evaluate the influence of traffic on dispersion of described the downwind pollutant concentration from pollutants near urban roadways. He observed the an infinite crosswind line source. Both large scale and variations in upwind dispersion due to crossroad wind, mesoscale winds were included in their model. Further, which occurred within a few meters of the road. The US the analysis showed that mesoscale winds are not EPA developed a number of air pollution models for significant in reducing pollutant concentration. Peters highway which included CALINE (Beaton et al., 1972), and Klinzing (1971) described two separate equations EGAMA (Egan et al., 1973), and HIWAY (Zimmerman for ground level as well as elevated line source and and Thompson, 1975). The popular HIWAY model was analysed the effects of diffusion coefficient in line source based on the Gaussian equation with the assumption of dispersion. Using the diffusion equation, Lamb and a series of finite line sources. CALINE model is also a Neiburger (1971) came out with a model for computing Gaussian-based LSM, but it has got separate equations pollutant concentrations resulting from both point and for calculating pollutant concentration under crosswind line sources. Later, this model was tested with respect to and parallel wind conditions. Chock (1977b) and Noll its diffusion characteristics by computing the hourly CO et al. (1978), evaluated these models and found that the concentrations on a particular day at 760 locations in EPA-HIWAY model overestimates pollutant concentra- the Los Angles basin. The model results showed tions adjacent to the highway. This model avoids the reasonably good agreement with observed values. Csan- cumbersome integration necessary for the conventional day (1972) developed a hypothetical model for a finite Gaussian model that makes point source assumption; line source and it was applicable only when the wind was instead it uses an infinite line source approach and perpendicular to the roadway. Calder (1973) studied the specifies one dispersion parameter as a function of wind effect of oblique wind on line source pollution dispersion road orientation from the source. Later, a series of near roadways. He showed that the concentration at improved versions of CALINE model, viz. CALINE-2, S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 2087 CALINE-3 and CALINE-4 were developed by Ward to handle pollutant dispersion under low wind condi- et al. (1977) and Benson (1979, 1989). Middleton et al. tions. HIWAY-4 is another version of HIWAY model (1979) developed a dispersion model for estimating the developed by incorporating modified dispersion curves concentration of inert gaseous pollutants from the and an aerodynamic drag factor to the original HIWAY curved circular and straight sections of a complex road model. Chang (1980) reviewed various urban air quality interchange. For small wind angles, pollutant concen- simulation models. These models were based on either tration predicted from a finite road length well matched Gaussian principles or the conservation of mass with the concentration estimates obtained from an principle with the gradient-transport approximation. infinite LSM. For complex roadway geometry, Colwill The EPA rollback model (EPARM) and a generalized et al. (1979) conducted experiments to observe the rollback model (GRM) were evaluated by Chang et al. change in pollutant concentration over a short distance (1980a). Both models showed similar predictions when at a site downwind of an isolated motorway and within a identical inputs were used for estimation. Data from road complex. DeTar (1979) came out with a model GM dispersion experiments were utilized by Sedefian which estimated the concentration of pollutants from and Rao (1981) to assess the characteristics of traffic- line sources. This model was applicable at various generated turbulence and its effects on the dispersion receptor heights, distances, wind speed and direction. process near roadways. They found that the dispersion Eskridge et al. (1979) presented a finite-difference next to the highway was dominated by the traffic and its highway model, which was based on the surface layer influence decreases considerably at further downwind similarity theory and vehicle wake theory to determine distances and at higher elevations. However, at low wind the atmospheric dispersion along a roadway. The model speeds and perpendicular cases, the traffic contribution simulation results were compared with the GM experi- to the total diffusivity was still about 50% at a ment data and showed that they are closer to the downwind distance of 30 m. The use of analytical observation than the other Gaussian-formulated EPA models for estimating the vehicular pollution dispersion highway model (HIWAY). In another study, Green et al. on Indian urban roadways having heterogeneous traffic (1979) found that the actual ground level concentrations was reported by Munshi and Patil (1981). The authors might decrease with decreasing wind speed particularly applied the atmospheric turbulent diffusion laboratory when it dropped below some critical value. Data from (ATDL) model to estimate the pollutant concentration an ongoing model validation program for dispersion of for Bombay city, which is one of the most populated and pollutants from roadways in Texas showed that trafficked cities in India. Hickman and Colwill (1982) pollutant concentrations did not increase as rapidly described a simple and effective method of estimating with decreasing wind speed as predicted by most models pollutant concentrations around highways, which used which were based on Gaussian formulation. Rao et al. the Gaussian dispersion theory with empirical modifica- (1979) studied the impact of traffic-induced turbulence tions so that it accurately represented the roadside on the near roadway dispersion of air pollutants. The situation. This model was calibrated and validated with study concluded that there is a noticeable augmentation the measured CO concentration at three different of turbulent kinetic energy due to wake-generated locations in UK. Using Texas and GM data, Rodden moving traffic. Later, Rao et al. (1980) evaluated four et al. (1982) evaluated 5-line source dispersion models Gaussian models, namely, GM (Chock, 1978), HIWAY namely CALINE-3, CALINE-2, AIRPOL-4, HIWAY (Zimmerman and Thompson, 1975), AIRPOL-4 (Car- and TRAPS-IIM (Bullian et al., 1980). All the four penter and Clemena, 1975), CALINE-2 (Ward et al., models showed poor performance when compared with 1977) and three numerical models—DANARD (Da- observed data. The time resolution and vertical spacing nard, 1972), MROAD-2 (Krisch and Mason, 1975) and that were necessary to resolve vehicle wake turbulence ROADS (Pitter, 1976). Their comparative results and the role of pseudo-turbulence in modelling pollutant showed that GM model simulations were more precise diffusion near the roadway was discussed by Eskridge than any other model. Peterson (1980) presented an and Rao (1983). The study revealed that velocity updated version of HIWAY model, i.e. HIWAY-2, variances measured in GM experiment data were released by EPA in May 1980. The difference between dominated by the wake-passing effect (time variation the original and improved version was that the latter in the wind velocity as a vehicles wake passes the model gave more realistic concentration estimates as it observation point) and were inadequate to resolve the used an updated dispersion algorithm. Rao and Keenan wake-turbulence effect. The Texas Instrumentation model (TEXIN) for predicating air quality near road- (1980) modified the Pasquill-Grfford dispersion curves way intersection was developed by Nelli et al. (1983). built in the EPA-HIWAY model and found that the The TEXIN model predictions were also compared with modified HIWAY model (HIWAY-3) had better simu- three intersection models—Inter section Midblock lation of the physics of the near-roadway dispersion Model, program MICRO and Indirect Source Guide- compared to the original HIWAY model. Further, an lines. Their comparative results showed that the TEXIN empirical aerodynamic drag factor was also developed 2088 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 model predictions were better than other three models. TEXIN-2 and CALINE-4. The study suggested an Using the Gaussian equation, Segal (1983) presented a alternative emission analysis procedure, which could Graphical Input Microcomputer model (GIMM) for be used in standard LSMs to estimate air quality predicting CO concentration from various types of line conditions at the intersection. Luhar and Patil (1989) sources. This model appeared to be suitable for screen- presented a GFLSM, based on the Gaussian diffusion ing purposes as it required less computational time. equation, pertaining to heterogeneous traffic conditions, Hickman and Waterfield (1984) developed a computer at two traffic junctions in Bombay city. The GFLSM code for predicting air pollutant concentrations for predictions were later compared with GM, CALINE-3 roadway traffic. The code provided a wide range of and HIWAY-2 model predictions, which showed that information required for air quality assessment, such as, GFLSM performed with reasonable accuracy for Indian concentration of pollutants corresponding to traffic traffic conditions. Using historical meteorological and flows, weather conditions and exposure periods. Cohn vehicular data, Cooper (1989) derived meteorological and Gaddipati (1984) developed an interactive graphics persistence factor (MPF) and vehicular persistence method for highway air pollution analysis. This factor (VPF) for Florida city. Further, a worst case approach was developed to resolve problems associated 'total persistence factor' was also derived which, is equal with the use of coordinates in CALINE-3 and HIWAY- to the product of the mean annual second-highest MPF 2 models. A digital computer model simulation of traffic and the mean VPF defined (TPF). Kono and Ito (1990a) flow has been developed by Beiruti and Al-Omishy developed a microscale dispersion model—the OMG (1985). Later, this model was used to predict NO2 and volume source model. Later, the OMG volume source, HC concentrations at three busy traffic roads in JEA model, Tokyo model and EPA-HIWAY-2 model Baghdad, Iraq. It showed a good agreement with the results were compared with the measured SF6 concen- measured concentrations. Cooper (1987) reviewed var- tration and it was found that OMG volume source ious models used for estimating the impact of indirect model predictions were superior to the other two models sources on CO and air quality. A methodology for (Kono and Ito, 1990b). Singh et al. (1990) developed an predicting the 8-h concentration by using 1-h CO analytical dispersion model (IITCO) for computing CO concentrations was also discussed. Hlavinka et al. concentration for heterogeneous traffic. The perfor- (1987) presented an improved version of TEXIN model, mance of the IITCO model was compared with the i.e., TEXIN-2. This model used the critical movement pollution episodic model (PEM) and intersection mid- analysis (CMA) procedure for estimating traffic flow block model (IMM). The results showed better perfor- parameters, MOBILE-3 to determine free flowing traffic mance of IITCO model when compared with PEM and cruise emissions and CALINE-3 to model the pollutant IMM. In another study, the IITCO model was applied distribution downwind of an intersection. Kunler et al. to Kuwait traffic. The results were compared with the (1988) discussed the applicability of various air quality US operational model, namely the IMM (Zanaidi et al., models for describing the dispersion of car exhaust 1991). Among the simulations obtained from both the emissions. The impact of VEEs, namely, CO and NOx models, the performance of IMM was found to be better on forest cover has also been discussed. Hoydysh et al. than IITCO. Miles et al. (1991) developed a hybrid approach for assessing air quality implications of urban (1987) used four distinct approaches, i.e., solution of the planning. This hybrid approach was a combination of Navier-Stokes equations, two-dimensional semi-empiri- both deterministic and statistical models. It was a cal models, zero-dimensional semi-empirical models and function of vehicular traffic and basic meteorology. empirical adaptation of Gaussian line source dispersion Using the theory of large eddy simulation, Nieuwstadt to model air flow and mass dispersion in street canyons. (1992a) studied the dispersion characteristics of a The results indicated that none of the approaches gave passive line source pollutant in the convective atmo- accurate distribution of pollutant concentrations in spheric boundary layer. Further, he also studied the street canyons. In another study, a chemical mass dynamics of line source by considering it into two balance (CMB) model was formulated by Khalil and parts—part-I (governed by the internal buoyancy) and Rasmussen (1988) which was applied for CO apportion- part-II (governed by the ambient turbulence). For the ment among residential wood burning sources and latter part, he developed a simple integral plume rise automobile sources in Olympia, Washington. Gronskei model (Nieuwstadt, 1992b). Benson (1992) studied (1988) studied the influence of car speed on dispersion of recent versions of CALINE models, namely, CALINE- exhaust emissions. He also pointed out that vertical 3 and CALINE-4. He evaluated the predictive capability diffusion of exhaust gas tends to be larger from high of CALINE-4 and found its performance to be better speed driving cars than low speed driving cars. His than CALINE-3. In another study, Alcxopolos et al. experimental results, when compared with HIWAY-2 (1993) came out with a model for spatial and temporal predictions, showed that HIWAY-2 under-predicted the evaluation of traffic emissions in metropolitan areas. pollutant concentrations. Sculley (1989) reviewed four The model was found to be useful where raw traffic representative approaches namely IIM, MICRO-2, S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 2089 data, network and number of trip data were difficult to and modified Gaussian equation to identify street generate. Qin and Kot (1993) carried out dispersion canyon and vehicle wake effects on transport of air studies in low wind conditions for three streets in pollution from urban road microenvironments. The Guangzhou city. Using the observed data, a simple computer model simulated and analysed the wind flow operational model was proposed to simulate the and their components in the street canyon considering a dispersion of vehicular emissions in street canyons. two-dimensional street canyon flow pattern. Vehicles Burden et al. (1994) used CALINE-4 and DMRB wake turbulence was also estimated in microenviron- models for predicting NO2 concentration at roadway ments. Subsequently, the turbulent parameters were intersections, in Bristol, UK. The two models provided integrated in Gaussian equations to estimate CO and satisfactory predictions for flexible traffic volume. NOx concentrations. Later, Karim (1999) developed a Akeredoiu et al. (1994) used the CALINE-4 model for traffic pollution inventory and modelled dispersion of forecasting CO at a roadway intersection. Chan et al. vehicular pollutants in an urban environment. Buckland (1995) tested the applicability of four simple dispersion and Middleton (1999) presented nomograms for screen- models, namely, APRAC, GZE, CALINE-4 and ing of vehicular pollution in congested street canyons. PWILG. These models were evaluated by comparing Sivacoumar and Thanasekaran (2001) evaluated four the predicted CO and NOx concentrations with mea- Gaussian dispersion models, namely, GM, CALINE-3, sured values at street canyons in Guangzhou city. The PAL-2 and ISCST-2 for Indian traffic conditions. The models were found to be accurate in predicting study revealed that GM model proved to be the best maximum ground level concentrations. Derwent et al. among four models, followed by CALINE-3, ISCST-2, (1995) used 1-year air quality data collected at one of and PAL-2. A detailed review on analytical modelling the urban roadside locations in central London to techniques, including deterministic and statistical mod- evaluate Gaussian and box models. The predicted results elling approach in the area of VEEs can be found in were later used for a comprehensive validation of the Sharma and Khare (2001a). Further, model perfor- published emission inventory estimates of London city. mance evaluation and comparative assessment were also Further, a relationship between hourly mean NO2 and discussed in this review. NOx concentrations was also developed for local air quality management. Esplin (1995) presented approx- imate explicit solutions to the general line source 4. Line source numerical models problem that could be used down to angles of 151 between the line source and the wind vector, for angles Danard (1972) developed a two-dimensional Eularian below 151, he presented a point source approximation model named as DANARD. It solved the mass solution. Clifford et al. (1995) studied the mechanisms conservation equation based on numerical methods involved in the dispersion of pollutants around slow outlined by Dufort and Frankel (1953). Using the moving vehicles. The spatial distribution of tracer gas boundary conditions imposed in DANARD, Ragland along and across the vehicles showed that a significant and Pierce (1975) derived the continuity equation for level of pollution was received by a commuter in a slow parallel and non-parallel diffusivity classes by an moving vehicle from the automobile immediately in efficient matrix inversion technique. The model pre- front. Using traffic counts and fleet composition, Yu dicted concentrations for oblique and perpendicular et al. (1996) developed a mathematical model for cases by ignoring lateral diffusion. For parallel cases, the predicting trends in CO emissions. The model results model solved the equation in three dimensions including were later used for examination of long-term trends in lateral diffusion. Krisch and Mason (1975) developed human exposure to CO. Chock and Winkler (1997) the MROAD-2 model, which was also a Eulerian two- compared the impact on air quality predictions using a dimensional grid model. It numerically solved the mass fixed layer-depth and a varying layer-depth structure in conservation equation. The size of the grid can be the urban airshed model (UAM). The analysis showed specified by the user and model allowed the existence of that the fixed layer-depth approach yielded substantially several line sources (all assumed to be perpendicular to higher concentrations of CO, NO and VOC in the lower the plane of the road), including elevated roadways. layer of atmosphere in isolated areas in early morning Pitter (1976) described ROADS model, which was a than the varying layer depth approach. Khare and two-dimensional conservation model. The model deter- Sharma (1999) presented a deterministic model for Delhi mined the steady-state concentrations of pollutants by traffic conditions (heterogeneous in nature), i.e. Delhi numerically solving the equations using the Lax- Finite Line Source model (DFLSM). This model showed Wendroff finite difference scheme. Chock (1978) for- better prediction accuracy for CO compared to GFLSM mulated a numerical model to solve advection diffusion (Luhar and Patil, 1989). Karim and Matsui (1998) and equation for a line source. In this model traffic effects Karim et al. (1998) developed a computer model were considered as additive components of the eddy consisting of wind distributions, emission dispersion diffusivity tensor (Kij) over that of atmospheric effects in 2090 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 the following form: CAR-FMI (Harkonen et al., 1995, 1996a, b) is a road network dispersion model, developed by the Finnish Meteorological Institute. The Norwegian Institute of Air where Kf- is the ambient/atmospheric eddy diffusivity Research (NILU) also developed ROADAIR and CON- tensor and Kijt is traffic-induced component of eddy TILENK models for open roads and NERI OSRM for diffusivity tensor. Later, this model was validated for street canyons, respectively (NILU, 1996). A recent study GM experiment data and was reported to be valid by Karppinen et al. (1997, 1998) described the application within 710% accuracy limits. Eskridge et al. (1979) of CAR-FMI model in estimating the contribution from presented a finite difference highway model. The model mobile sources in predicting the emission, dispersion and used surface layer similarity theory and vehicle wake chemical transformation of NOx in an urban area. theory of Eskridge and Hunt (1979) to determine the atmospheric structure along the roadway. The model results were compared with HIWAY model and found 5. Line source stochastic models closure to the observed GM experimental data. Eskridge and Thompson (1982) developed the ROADWAY McGuire and Noll (1971) studied the relationship model. It was a finite difference model, which predicted between maximum concentration and average time for pollutant concentration near a roadway. It assumed a CO, NOx and NO2 pollutants collected at 17 monitoring surface layer describable by surface layer similarity stations in California city. From the past studies on air theory with the superposition of the effects of vehicle pollution modelling, there existed substantial evidence wakes. This model could also predict velocity and that the series of pollution concentration and meteor- turbulence along the road. ROADCHEM model ological data were highly auto-correlated irrespective of (Eskridge and Thompson, 1982) was a version of time (Merz et al., 1972). McCollister and Wilson (1975) ROADWAY which incorporated the chemical reactions used the B-J type models for short-term forecast of involving NO, NO2 and O3 as well as advection and oxidant and CO in Los Angeles basin. The model was dispersion phenomena. In this version, surface layer one-dimensional and showed poor predictions for similarity theory was used to produce vertical angle extreme events. Tiao et al. (1975) studied the effects of turbulence profiles. Maddukuri (1982) presented a intervention caused by a new highway on the oxidant numerical model for the estimation of vehicular exhaust time series in the Los Angeles basin. A univariate (CO) dispersion. The model was based upon the semi- analysis of the weekly averages of the daily maxima of empirical equation of turbulent diffusion equation. oxidant, CO, NO2, and total HC was done by Chock Eskridge and Rao (1986) modified the ROADWAY et al. (1975) for Riverside, CA. The relationships model by using experimentally determined eddy diffu- between the weekly averages of the daily maxima of sion coefficients. The revised version of ROADWAY the oxidant and the weekly averages of meteorological was completely independent of the GM sulphate parameters were also investigated. Based on least-square experiment data, whereas the initial version used the technique, Aron and Aron (1978) presented a stochastic GM data to determine the diffusion coefficients needed model for forecasting daily maximum CO concentra- in the wake theory. Later version of ROADWAY model tions at Los Angles basin. The analysis showed that CO predictions were closer to GM data than the initial concentration of the preceding days, pressure differences version. Thompson and Eskridge (1987) experimentally between nearby station, surface temperature, day of the studied the influence of vortex pair in turbulent diffusion week, length of daylight, solar radiation and inversion behind vehicles. These experimental results were incor- height were the most significant variables in model porated into ROADWAY model for improving its development. Hirtzel and Quon (1979) used auto- predicting efficiency. The model physics was based correlation function to model hourly and average primarily on the vehicle speed, turbulence and diffusion eight-hourly CO concentrations measured at the con- of tracer. The calculation of air pollution from road tinuous air monitoring program station in Chicago. In traffic (CAR) model developed by the Dutch National another study, Ledolter and Tiao (1979) presented a Institute of Environmental Health (RIVM, 1991) and statistical model that predicted CO concentration on Dutch Institute of Applied Scientific Research (Van den both sides of the freeway in Los Angles. Regression Hout et al., 1989) was evaluated for a Dutch city by analysis technique was used by Chang et al. (1980b) to Eerens et al. (1993). The calibration of the model was determine the relative impact of mobile and stationary done by using data from the Dutch National Air Quality sources on high NO2 concentrations. The analysis Monitoring Network (Elskamp, 1989; Heida et al., showed that hourly NO2 concentrations observed in 1989) and wind tunnel experiments (Van den Hout et al., Los Angles basin arise largely from vehicle emissions 1989). The CAR model satisfactorily estimated the and support the assumptions used in generalised roll- air pollutant concentrations in urban streets and back model (Chang et al., 1980a). Lincoln and Rubin was comparable with Dutch air quality standards. (1980) applied multiple regression analysis to correlate S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 2091 CO with daily average haze coefficient and total another work, Dorzdowicz et al. (1997) developed a suspended particulate (TSP) in Downtown urban area line source neural network model for estimating hourly of Los Angles. Zamurs and Piracci (1982) developed a mean concentrations of CO in the urban area of Rosario multiple linear regression model to predict CO concen- city. Eleven inputs, e.g. vehicular flux-vehicles/h (cars, tration at selected intersections. Using a statistical taxis, median vehicles, trucks and buses), wind speed theory, a simple formula for calculating dispersion from and direction, solar radiation, humidity, pressure, rain a continuous finite line source was presented by intensity and temperature were used for development of Mikkelsen et al. (1982). Jakeman et al. (1991) used a the ANN-based model. This model was later validated hybrid (deterministic þ stochastic) model to predict for each type of network (i.e., considering different seasonal extremes of 1-h average CO concentration. number variable sets), and using approximately a set of Bardeschi et al. (1991) noted the importance of time 100 patterns. Using hourly NOx and NO2 and meteor- series of concentration emission and meteorological ological data of Central London, Gardner and Dorling conditions during the hours prior to the high CO (1999) developed MLP neural network models. The concentrations. In the recent past, a number of studies predicted results showed better performance when have been carried out using multivariate time series compared with previously developed regression models analysis in which various B-J modelling techniques have (Shi and Harrison, 1997) for the same location. been applied for homogeneous traffic conditions (Hsu, 1992; Hernandez et al., 1992; Manteiga et al., 1993; Trier and Firinguetti, 1994; Milionis and Davis, 1994a, b). Liu 7. Limitations of LSMs et al. (1994) used Monte Carlo simulation method to predict personal exposure levels to CO in Taipei. The Deterministic LSEM approach is the most logical and skewness and kurtosis methods were used by Zhang et al. traditional approach for the prediction of air pollution (1994) to investigate the statistical distribution of CO concentrations, yet it is not free from limitations. The and hydrocarbon (HC) emissions on a road in Denver, prediction capability of deterministic models depends on US. Glen et al., (1996) developed an empirical model of the condition fulfilling the simplifying assumptions, monthly CO for long-term trend assessment. Comrie which are made in the model formulation. For instance, and Diem (1999) examined the relationships between when unit time interval is short (i.e., p1 day) and meteorology, traffic patterns and CO concentration at 'steady state' assumptions required for the application seasonal, weekly and diurnal time scales in Phoenix, AZ. of Gaussian type models are not met, deterministic Sharma et al. (1999) applied extreme value theory to models do not give satisfactory results (Fingi and know the expected number of violations of the National Tebaldi, 1982). The Gaussian dispersion equation has Ambient Air Quality Standards (NAAQS) to hourly and a singularity at zero wind speeds. Therefore, all eight-hourly average CO concentrations for an air Gaussian-based models perform poorly when wind quality control region comprising of an urban road speeds are o 1 m/s. intersection followed by the development of an inter- In general, Gaussian-based model predictions are vention analysis model (IAM) for the same AQCR reasonably accurate for long-term average concentra- (Sharma and Khare, 1999). Further, the authors tions and for the frequency distribution up to 90 (Sharma and Khare, 2000) used B-J modelling techni- percentiles (Kretzschmar et al., 1976). However, the ques to provide short-term and real-time forecast of the predictions become inaccurate when the frequency ambient air pollution levels due to vehicular sources at distribution is 98 percentiles (Nieuwstadt, 1980). an urban intersection. In another study, Sharma and Further, deterministic models are not suitable for Khare (2001b), developed models for Delhi traffic extreme value predictions (North et al., 1984). However, conditions which were stochastic in nature and per- they are most suitable for long-term planning decisions formed with reasonable prediction accuracy for CO (Benarie, 1980; Juda, 1986; Zanneti, 1990). Table 1 concentration in heterogeneous traffic conditions. summarises the limitations of selected line source models. Numerical models have common limitations arising 6. ANN-based LSMs from employing the K-theory for the closure of diffusion equation. The K-theory diffusion equation is valid only Literature on application of ANN in line source if the size of the 'plume' or 'puff of pollutants is greater modelling is found to be very scanty (Nagendra and than the size of the dominant turbulent eddies. The K- Khare, 1999). Moseholm et al. (1996) studied the model assumption is also not valid for the convective usefulness of neural network in understanding the boundary layer under strong instability. The other relationships between traffic parameters and CO con- limitations of numerical models are large computational centrations measured near an intersection, which was costs in terms of time and storage of data. It also sheltered from wind by multi-storied buildings. In requires large amounts of input data. The solutions 2092 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 Table 1 Applicability and limitations of selected line source models Sl. No. Model Applicability Limitations Pollutant type Receptor location and traffic type California line source CO, NOx, SPM Roadside Tendency to predict high pollutant model (Beaton et al., concentration for parallel wind case 1972) Homogeneous No treatment of plume rise due to hot exhaust of vehicles HIWAY-1 CO Roadside Predicts poorly for low winds (Zimmerman and Thompson, 1975) Homogeneous Overestimates pollutant concentration for stable atmospheric condition and parallel wind case No treatment of plume rise due to hot exhaust of vehicles CALINE-2(Wardetal., CO, NOx, SPM Roadside Predicts poorly for unstable and 1977) neutral stability conditions Homogeneous Over-predicts the pollutant concentration for parallel wind cases and under-predicts for oblique wind conditions GM model (Chock, CO Roadside Tendency to over-predict the 1978) concentration under parallel wind conditions Homogeneous Predicts poorly for low winds CALINE-3 (Benson, CO, NOx, SPM Roadside Tendency to predict high for parallel 1979) wind condition Homogeneous No proper treatment for mechanical and thermal turbulence created by vehicle exhaust HIWAY-2 (Peterson, CO Roadside Inadequate dispersion parameters 1980) Homogeneous No treatment of plume rise due to hot exhaust of vehicles HIWAY-3 (Rao et al., CO Roadside Predicts poorly for low winds 1980) Homogeneous Tendency to predict high for parallel wind condition No treatment of plume rise due to hot exhaust of vehicles HIWAY-4 (Rao et al., CO Roadside Tendency to predict high for parallel 1980) wind condition Homogeneous No treatment of plume rise due to hot exhaust of vehicles CALINE-4 (Benson, CO, NOx, Aerosol Roadside Tendency to predict high for parallel 1989) wind condition Homogeneous S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 2093 Table 1 (continued) Sl. No. Model Applicability Limitations Pollutant type Receptor location and traffic type 10. ISCST-2 (EPA, 1992) CO, NOx, SPM Roadside Tendency to predict high for parallel wind condition Homogeneous No treatment for turbulence caused by heated exhaust 11. GFLSM (Luhar and CO, SPM Roadside Predicts poorly for low winds Patil, 1989) Heterogeneous 12. DFLSM (Khare and CO Roadside Predicts poorly for low winds Sharma, 1999) Heterogeneous obtained from numerical model are approximate one. the problem being addressed and the degree to which the Past studies revealed that the performance of numerical problem is understood. Deterministic LSEM approach models namely DANARD (Danard, 1972), MROAD-2 seems to be the most logical and traditional approach (Krisch and Mason, 1975), ROADS (Pitter, 1976), for modelling air pollution concentrations. The predic- ROADWAY (Eskridge et al., 1979) was not better than tion capability of deterministic models depends on the the Gaussian-based deterministic LSMs (Rao et al., condition fulfilling the simplifying assumptions, which 1978, 1980, 1986). are made in the model formulation. The Gaussian model Limitations of statistical models include the require- is generally accepted for prediction of long-term average ments of long historical data sets and lack of physical concentrations. Numerical models are most desirable interpretation. Another limitation of statistical model is solution, if adequate data, computational resources and that they cannot provide information about how other theoretical understanding of dispersion phenom- pollutant levels would respond to emission controls, ena are available. Statistical models are site specific and though statistical distribution modelling has been do underperform when modelled with highly nonlinear reported to be used in developing simple roll-back data. In recent years, ANN approach, particularly formulae for determining a desirable level of source MLPs are particularly apparent in applications where emission control to meet with the objectives of air a full theoretical (deterministic and statistical) models quality management (Georgopoulos and Seinfeld, 1982). cannot be constructed, and especially when dealing with The statistical models are therefore site specific. complex conditions (Gardner and Dorling, 1998). The The main reasons often cited for not using the MLP, existing modelling studies relevant to temporal and in vehicular exhaust modelling is that they are difficult spatial dispersion of VEEs reveals that most of the to implement. The other problem faced when training LSEM studies are made on homogeneous traffic MLP is deciding upon the network architecture (i.e., conditions (Sharma and Khare, 2000,2001b). For number of hidden layers, number of nodes in hidden heterogeneous traffic conditions, very few studies were layers and their interconnection, Boznar et al., 1993). At carried out. The commonly used LSMs (Table 1) for air present, no procedure has been established for selecting quality assessment performs poorly on parallel wind a proper network architecture. Apart from these, no cases (Sivacoumar and Thanasekaran, 2001). A limited thumb rules exist in selection of data set for training, number of studies were reported on time series and testing and validation of neural network based model. ANN approaches in modelling of VEEs (Sharma, 1998; Nagendra and Khare, 2002). 8. Conclusion References The LSEM has been shown to be useful tool for prediction of urban air quality. Analytical modelling Akeredoiu, F.A., Oluwole, A.F., Betika, E.A., Ogunsola, O.J., approaches including deterministic and statistical tech- 1994. Modelling of carbon monoxide concentration from niques, are commonly used for LSEM. Choosing the motor vehicles travelling near roadway intersections in most suitable approach, depends on the complexity of Lugos, Nigeria. In: Baldanano, J.M., Brebbia, C.A., Power, 2094 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 H., Zannetti, P. (Eds.), Air Pollution II. Pollution Control Experiments. General Motors Research Publication, GMR- and Monitoring, Vol. 2. Computational Mechanics Pub- 2107. General Motor Corporation, Warren. lications, Southampton, Boston, pp. 149-157. Calder, K.L., 1973. On estimating air pollution concentrations Alcxopolos, A., Assimacopoulos, D., Mitsoulis, E., 1993. from a highway in an oblique wind. Atmospheric Environ- Model for traffic emissions estimation. Atmospheric En- ment 7, 863-868. vironment 27B (4), 435-446. Carpenter, W.A., Clemena, G.G., 1975. Analysis and compara- Aron, R.H., Aron, I.M., 1978. Statistical forecasting models: tive evaluation of AIRPOL-4. Virginia Highway and carbon monoxide concentrations in the Los Angles basin. Transportation Research Council Report, VHTRC75-R55. Journal of Air Pollution Control Association 28 (7), Charlottesville, VA. 681-684. Chan, L.Y., Hung, W.T., Qin, Y., 1995. Assessment of Bardeschi, A., Colucci, A., Granelle, V., Gangnetti, M., vehicular emission dispersion models applied in street Tamponi, M., Tebaldi, G., 1991. Analysis of the impact canyons in Guangzhou, PRC. Journal of Environmental on air quality of motor vehicle traffic in the Milan urban International 21 (1), 39-46. area. Atmospheric Environment 25B, 415^128. Chang, T.Y., 1980. Current concepts and applications of air Beaton, J.L., Ranzieri, A.J., Shirley, E.C., Skog, J.B., 1972. quality simulation models. Atmospheric Environment 3, Mathematical approach to estimating highway impact on 337-351. air quality. Fed. Highway Admin. Report. No. FHWARD- Chang, T.Y., Norbeck, J.M., Weinstock, B., 1980a. Urban 72-36, Washington, DC. center CO air quality projections. Journal of Air Pollution Beiruti, A.A.R., Al-Omishy, H.K., 1985. Traffic atmospheric Control Association 30 (9), 1022-1025. diffusion model. Atmospheric Environment 19 (9), Chang, T.Y., Norbeck, J.M., Weinstock, B., 1980b. NO2 air 1519-1524. quality precursor relationship: an ambient air quality Benarie, M.M., 1980. Urban air pollution modelling. The MIT evaluation in the Los Angles basin. Journal of Air Pollution Press, London. Control Association 30 (2), 157-162. Benson, P.E., 1979. CALINE-3. A versatile dispersion model Chen, T.C., March, F., 1971. Effect of highway configurations for predicting air pollutant levels near highway and arterial on environmental problems dynamics of highway associated roads. Final Report, FHWA/CA/TL.-79/23, California air pollution. In: Englund, H.M., Berry, W.T. (Eds.), Department of Transportation, Sacramento, CA. Proceedings of Second International Clear Air Congress. Benson, P.E., 1989. CALINE-4. A dispersion model for Academic Press, New York, pp. 35-40. predicting air pollutant concentrations near roadway. Final Chock, D.P., 1977a. General Motors sulphate dispersion Report, FHWA/CA/TL.-84/15. California Department of experiment: an overview of the wind temperature and Transportation, Sacramento, CA. concentration fields. Atmospheric Environment 11, Benson, P.E., 1992. A review of the development and 553-559. application of the CALINE-3 and CALINE-4 models. Chock, D.P., 1977b. General Motors sulphate dispersion Atmospheric Environment 26B (3), 379-390. experiment: assessment of the EPA HIWAY model. Journal Box, G.E.P., Jenkins, G.M., 1970. Time Series Analysis of Air Pollution Control Association 27, 39-45. Forecasting and Control. Holenday, San Franciso. Chock, D.P., 1978. A simple line source model for Box, G.E.P., Jenkins, G.M., 1976. Time Series Analysis dispersion near roadways. Atmospheric Environment 12, Forecasting and Control, 2nd Edition. Holenday, San 823-829. Francisco. Chock, D.P., Winkler, S.L., 1997. Air quality prediction Boznar, M., Lesjak, M., Malker, P., 1993. A neural network using a fixed layer depth vertical structure in the urban based method for short-term predictions of ambient SO2 airshed model. Environmental Science and Technology 31, concentrations in highly polluted industrial areas of 359-370. complex terrain. Atmospheric Environment 27B (2), Chock, D.P., Terrel, T.R., Levitt, S.B., 1975. Time series 221-230. analysis of riverside California air quality data. Atmo- Buckland, A.T., Middleton, D.R., 1999. Nomograms for spheric Environment 9, 978-989. calculating pollution within street canyons. Atmospheric Clifford, M.J., Clarke, R., Riffat, S.B., 1995. Local aspects of Environment 33, 1017-1036. vehicular pollution. Atmospheric Environment 31 (2), Bullian, J.A., Polasek, J.C., Green, N.J., 1980. Analytical and 271-276. experimental assessment of highway impact on air quality: Cohn, L.F., Gaddipati, S.R., 1984. An interactive graphics data analysis and model evaluation. Texas Transportation method for use in highway air quality analysis. Journal of Institute Research Report, 218-5F. Chemical Engineering Air Pollution Control Association 34 (11), 1137-1139. Department, Texas, A&M University, College station, TX. Colwill, D.M., Middleton, D.R., Bulter, J.D., 1979. A Gaussian Burden, N.C., Whitwell, I., Longhurst, J.W.S., 1994. Testing plume dispersion model applicable to a complex motorway the tools of an air quality management: a comparison of interchange. TRRL Report: Supplimentary Report 505. predictions of nitrogen dioxide concentrations by CALINE- Crowthorne, Berkshire. 4 and DMRB models with monitored levels at the M4/M5 Comrie, A.C., 1997. Comparing neural networks and regression interchange in Bristol, UK. In: Power, H., Tirabassi, T., model for ozone forecasting. Journal of Air and Waste Brebbia, C.A. (Eds.), Air Pollution V. Computational Management Association 47, 653-663. Mechanics Publication, Southampton, Boston, pp. 335-344. Comrie, A.C., Diem, J.E., 1999. Climatology and forecast Cadle, S.H., Chock, D.P., Heuss, J.M., Monson, P.R., modelling of ambient carbon monoxide in Phoenix, 1976. Result of the General Motors Sulfate Dispersion Arizona. Atmospheric Environment 33, 5023-5036. S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 2095 Cooper, D.C., 1987. Indirect source impact analysis carbon Eskridge, R.E., Binkowski, F.S., Hunt, J.C.R., Clark, T.L., monoxide modelling. Journal of Air Pollution Control Demerjain, K.C., 1979. Highway modelling—II advection Association 37 (11), 1308-1313. and diffusion of SF6 tracer gas. Journal of Applied Cooper, D.C., 1989. Persistence factor for mobile sources Meteorology 18, 401-412. (roadway) carbon monoxide modelling. Journal of Air Esplin, G.L., 1995. Approximate explicit solution to the general Pollution Control Association 39, 714-720. line source problem. Atmospheric Environment 29 (12), Csanday, G.T., 1972. Crosswind shear effects on atmospheric 1459-1463. diffusion. Atmospheric Environment 6, 221-232. Fay, T.A., King, D.E., 1975. Wake induced dispersion of Dabberdt, W.F., Ludwing, F.L., Jhonson Jr., W.B., 1973. automobile exhaust pollutants. Journal of Air Pollution Validation and application of an urban diffusion model for Control Association 1, 44-75. vehicular pollutants. Atmospheric Environment 7, 603-618. Fingi, G., Tebaldi, G., 1982. A mathematical model for air Danard, M.B., 1972. Numerical modelling of carbon monoxide pollution forecast and alarm in an urban area. Atmospheric concentration near a highway. Journal of Applied Meteor- Environment 16, 2055-2059. ology 11, 947-957. Gardner, M.W., Dorling, S.R., 1998. Artificial neural networks: Derwent, R.G., Middleton, D.R., Field, R.A., Goldstone, the multilayer perceptron—a review of applications in M.E., Lester, J.N., Perry, R., 1995. Analysis and inter- atmospheric sciences. Atmospheric Environment 32 (14/ pretation of air quality data from an urban roadside 15), 2627-2636. location in central London over the period from July 1991 Gardner, M.W., Dorling, S.R., 1999. Neural network model- to July 1992. Atmospheric Environment 29 (8), 923-946. ling and prediction of hourly NOx and NO2 concentrations DeTar, D.F., 1979. A new model for estimating concentrations in urban air in London. Atmospheric Environment 33, of substances emitted from a line sources. Journal of the Air 709-719. Pollution Control Association 29 (2), 138-141. Georgopoulos, P.G., Seinfeld, J.H., 1982. Statistical distribu- Dilley, J.F., Yen, K.T., 1971. Effect of mesoscale type wind on tion of air pollutant concentrations. Environmental Science the pollutant distribution from a line source. Atmospheric and Technology 16, 401-416A. Environment 5, 843-851. Glen, W.G., Zelenka, M.P., Graham, R.C., 1996. Relating Dorzdowicz, B., Benz, S.J., Sonta, A.S.M., Scenna, N.J., 1997. meteorological variables and trend in motor vehicle emis- A neural network based model for the analysis of carbon sions to monthly urban carbon monoxide concentrations. monoxide concentration in the urban area of Rosario. In: Atmospheric Environment 30, 4225^1232. Power, H., Tirabassis, T., Brebbia, C.A. (Eds.), Air Green, N.J., Bullin, J.A, Polasek, J.C., 1979. Dispersion of Pollution V. Computational Mechanics Publications, carbon monoxide from roadways at low wind speeds. Southampton, Boston, pp. 677-685. Journal of Air Pollution Control Association 29 (10), Dufort, E.C., Frankel, S.P., 1953. Stability condition in the 1057-1061. numerical treatment of parabolic differential equation. Gronskei, K.E., 1988. The influence of car speed on dispersion Mathematical Tables National Research Council Washing- of exhaust gases. Atmospheric Environment 22 (2), ton 7, 135pp. 273-281. Eerens, H.C., Sliggers, C.J., Hout, K.D.V., 1993. The car model Harkonen, T., Valkonen, E., Kukkonen, J., Rantakrans, E., the Dutch method to determine city street air quality. Jalkanen, L., Lahtinen, K., 1995. An operational dis- Atmospheric Environment 27B (4), 389-399. persion model for predicting pollution from a road. Egan, B.A., Epstein, B.A., Keefe, M., League, J., Lavery, T.C., International Journal of Environmental and Pollution 4-6, 1973. Development of procedure to simulate motor vehicle 602-610. pollution levels. Environment Research and Technology, Harkonen, J., Walden, J., Kukkonen, J., 1996a. Comparison of Inc., Report No. ERT-P-343F. Lexington, MA. model predictions and measurements near a major road in Elskamp, H.J., 1989. Normal air quality monitoring network— an urban area. In: Kretzchgman, J., Cosemans, G. (Eds.), technical description. RIVM Report-228702017. Bithoven, Proceedings of the Fourth Workshop on Harmonisation The Netherlands. within Atmospheric Dispersion Modelling for Regulatory EPA, 1992. User's Guide for the Industrial Source Complex Purposes, Vol. 2, Oostende, Belgium, 1996. Vloamse. Dispersion Models. Office of air quality planning and Instelling Voor Technologish Onderzock, Mol. Belgium, standards, Washington, DC. pp. 453-460. Eskridge, R.E., Hunt, J.C.R., 1979. Highway modelling—I Harkonen, T., Valkonen, E., Kukkonen, J., Rantakrans, E., prediction of velocity and turbulence fields in the wake of Lahtinen, K., Karppinen, A., Jalkanen, L., 1996b. A Model vehicles. Journal of Applied Meteorology 18, 387-400. for the Dispersion of Pollution from a Road Network, Vol. Eskridge, R.E., Rao, S.T., 1983. Measurement and prediction 23. Finnish Meteorological Institute, Publication on Air of traffic induced and velocity fields near roadways. Journal Quality, Helsinki, p. 34. of Climate and Applied Meteorology 22, 1431-1443. Haykin, S., 2001. Neural networks—A Comprehensive Foun- Eskridge, R.E., Rao, S.T., 1986. Turbulent diffusion behind dation, 2nd edition. Pearson Education Inc., New Delhi, vehicle: experimentally determined mixing parameters. India. Atmospheric Environment 20 (5), 851-860. Heida, H., Jang, A.L., Huygen, C., 1989. In: Brasser, Eskridge, P.E., Thompson, R.S., 1982. Experimental and L.J., Mulder, W.C. (Eds.), Model calculation of street theoretical study of the wake of a block-shaped vehicle in air concentrations for carbon monoxide, Vol. 3. a shear-free boundary flow. Atmospheric Environment 16, Elsevier, Amsterdam, The Hague, The Netherlands, 2821-2836. pp. 233-238. 2096 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 Hernandez, E., Martin, F., Valero, F., 1992. State-space source apportionment. Journal of Air Pollution Control modeling for atmospheric pollution. Journal of Applied Association 38 (7), 901-906. Meteorology 30, 793-811. Khare, M., Sharma, P., 1999. Performance evaluation general Hickman, A.J., Colwill, D.M., 1982. The estimation of air finite line source model for Delhi traffic conditions. pollution concentration from road traffic. TRRL Supple- Transportation Research: Part-D 4, 65-70. mentary Report No. 1052. Crowthorne, Berkshire. Kono, H., Ito, S., 1990a. A micro scale dispersion model Hickman, A.J., Waterfield, V.H., 1984. A user's guide to the for motor vehicle exhaust gas in urban areas—OMG computer programs for predicting air pollution from road volume source model. Atmospheric Environment 24B (2), traffic. TRRL Supplementary Report No. 806, Crowthorne, 243-252. Berkshire. Kono, H., Ito, S., 1990b. A comparison of concentration Hirtzel, C.S., Quon, J.E., 1979. Statistical dependence of hourly estimates by the OMG volume-source dispersion model carbon monoxide measurements. Journal of Air Pollution with three line source dispersion models. Atmospheric Control Association 29 (2), 161-163. Environment 24B (2), 253-260. Hlavinka, M.W., Korpics, J.J., Bullin, J.A., 1987. TEXIN-2. A Kretzschmar, J.G., de Baere, G., Vandervee, J., 1976. Valida- versatile model for predicting carbon monoxide concentra- tion of the emission frequency distribution model in the tions near intersections. Journal of Air Pollution Control region of Antwerpen, Belgium. In: Proceedings of the Association 37, 819-822. Seventh International Technical Meetings on Air Pollution Hornik, K., Stinchcombe, M., White, H., 1989. Multi layer feed Modelling and its Application, Airlie House, Virginia. forward networks are universal approximators. Neural NATO/CCMS Publication No. 51, pp. 235-270. Networks 2, 359-366. Krisch, J.W., Mason, B.F., 1975. Mathematical models for air Hoydysh, W., Orentlicher, M., Dabberlt, W., 1987. Air pollution studies involving the Oregon I205 Highway movement and vehicular pollution in urban street canyons. Project. Systems Science and Software Report, SSSR-76- In: Sterrett, F.S. (Ed.), Environmental Sciences, Vol. 2. The 2744, LaJolla, CA. New York Academy of Sciences, New York. Kunler, M., Kraft, J., Koch, W., Windt, H., 1988. Dispersion of Hsu, K.J., 1992. Time-series analysis of the interdependence car emissions in the vicinity of a highway. In: Grefen, K., among air pollutants. Atmospheric Environment 26, Lobel, J. (Eds.), Environmental Meteorology. Kluwer 491-503. Academic Publishers, London. Jakeman, A.J., Bai, J., Miles, G.H., 1991. Prediction of Lamb, R.G., Neiburger, M., 1971. An interim version of a seasonal extremes of 1-hour average urban CO concentra- generalized urban air pollution model. Atmospheric Envir- tions. Atmospheric Environment 25B, 219-229. onment 5, 239-264. Juda, K., 1986. Modelling of the air pollution in the Cracow Ledolter, J., Tiao, G.C., 1979. Statistical models for ambient air area. Atmospheric Environment 20, 2449-2456. pollution, with spatial reference to the Los Angeles catalyst Karim, M.M., 1999. Traffic pollution inventories and modelling study (LSCS) data. Environmental Science and Technology in metropolitan Dhaka, Bangladesh. Transportation Re- 13 (10), 1233-1240. search D4, 291-312. Lincoln, D.R., Rubin, E.S., 1980. Contribution of mobile Karim, M.M., Matsui, H., 1998. A mathematical model of wind sources to ambient particulate concentrations in a Down- flow, vehicle wake and pollution concentration in urban town urban area. Journal of Air Pollution Control road microenvironments—Part I: Model description. Trans- Association 30 (7), 777-780. portation Research Part D 3, 81-92. Liu, J.J., Chan, C.C., Jeng, F.T., 1994. Predicting personal Karim, M.M., Matsui, H., Guensler, R., 1998. A mathe- exposure levels to carbon monoxide (CO) in Taipei, based matical model of wind flow, vehicle wake and pollution on actual CO measurements in microenvironments and a concentration in urban road microenvironments— Monte Carlo simulation method. Atmospheric Environ- Part II: Model results. Transportation Research Part D 3, ment 28, 2361-2368. 171-191. Luhar, A.K., Patil, R.S., 1989. A general finite line source Karppinen, A., Kukkonen, J., Konttinen, M., Rantakrans, E., model for vehicular pollution prediction. Atmospheric Valkonen, E., Harkonen, J., Koskentalo, T., Elolahde, T., Environment 23 (3), 555-562. 1997. Comparison of dispersion model prediction model Maddukuri, C.S., 1982. A numerical model of diffusion of predictions and the results from an urban air quality carbon monoxide near highways. Journal of Air Pollution measurement network. In: Power, H., Tirabassis, T., Control Association 32 (8), 834-836. Brebbia, C.A. (Eds.), Air Pollution V, Fifth International Manteiga, W.G., Sanchez, J.M.P., Cao, R., Jurado, I.G., 1993. Conference on Air Pollution Modelling, Monitoring and Time series analysis for ambient concentrations. Atmo- Management. Computational Mechanics Publications, spheric Environment 27, 153-158. Southampton, Boston, pp. 405—411. Mayer, H., 1999. Air pollution in cities. Atmospheric Environ- Karppinen, A., Kukkonen, J., Konttinen, M., Harkonen, J., ment 33, 4029^1037. Valkonen, E., Rantakrans, E., Koskentalo, T., Elolahde, T., McCollister, G.W., Wilson, K.R., 1975. Linear stochastic 1998. The emission dispersion and chemical transformation model for forecasting daily maxima and hourly concentra- of traffic-originated nitrogen oxides at the Helsink metro- tions of air pollutants. Atmospheric Environment 9, politan area. International Journal of Vehicle Design 20 417^123. (1-4), 131-136. McGuire, T., Noll, K.E., 1971. Relationship between concen- Khalil, M.A.K., Rasmussen, R.A., 1988. Carbon monoxide in trations of atmospheric pollutants and average time. an urban environment: application of a receptor model for Atmospheric Environment 5, 291-298. S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 2097 Merz, P.H., Painter, H.J., Payson, P.R., 1972. Aerometric data Noll, K.E., Miller, T.L., Calggett, M., 1978. A comparison of analysis —time series analysis and forecast and an atmo- three highway line source dispersion models. Atmospheric spheric smog diagram. Atmospheric Environment 6, Environment 12, 1323-1329. 319-342. North, M., Harnandez, E., Garcia, R., 1984. Frequency Middleton, D.R., Butler, J.D., Colwill, D.M., 1979. Gaussian analysis of high CO concentrations in Madrid by stochastic plume dispersion model applicability to a complex motor- process modelling. Atmospheric Environment 18, way interchange. Atmospheric Environment 13, 1039-1049. 2049-2054. Mikkelsen, T., Troen, I., Larsen, S.E., 1982. On the finite line Peters, L.K., Klinzing, G.E., 1971. The effect of variable source problem in diffusion theory. Atmospheric Environ- diffusion coefficients and velocity on the dispersion of ment 16, 2591-2594. pollutants. Atmospheric Environment 5, 497-504. Miles, G.H., Jakeman, A.J., Bai, J., 1991. A method for Peterson, W.B., 1980. Users Guide for Highway Air Pollution predicting the frequency distribution of air pollution from Model. EPA-60018-80-018. vehicle traffic, basic meteorology, and historical concentra- Pitter, R.L., 1976. Users Manual ROADS, PSMG, VISI. tions to assist urban planning. Journal of Environmental Oregon Graduate Center, Beaverton, Oregon. International 17, 575-580. Qin, Y., Kot, S.C., 1993. Dispersion of vehicular emission in Milionis, A.E., Davis, T.D., 1994a. Regression and stochastic street canyons, Guangzhou, city, south China (PRC). models for air pollution—I: review comments and sugges- Atmospheric Environment 27B (3), 283-291. tions. Atmospheric Environment 28, 2801-2810. Ragland, K.W., Pierce, J.J., 1975. Boundary layer model for air Milionis, A.E., Davis, T.D., 1994b. Regression and stochastic pollutant accumulations due to highway traffic. Journal of models for air pollution—II: application to stochastic Air Pollution Control Association 25, 48-51. models to examine the links between ground level smoke Rao, S.T., Keenan, M.T., 1980. Suggestion for improvement of concentrations and temperature inversion. Atmospheric EPA-HIWAY model. Journal of Air Pollution Control Environment 28, 2811-2822. Association 30 (3), 247-256. Moseholm, L., Silva, J., Larson, T.C., 1996. Forecasting carbon Rao, S.T., Chen, M., Keenan, M., Sistla, G., Peddada, R., monoxide concentration near a sheltered intersection using Wotzak, G., Kolak, N., 1978. Dispersion of pollutants near video traffic surveillance and neural networks. Transporta- highways: experimental design and data acquetion proce- tion Research Part D 1 (1), 15-28. dures. EPA-600-4-78-037. NTIS. PB-284-866. US Environ- Munshi, U., Patil, R.S., 1981. Application of ATDL type mental Protection Agency, Research Triangle Park, NC. models to Bombay. Journal of Air Pollution Control Rao, S.T., Sedefian, L., Czapski, U.H., 1979. Characteristics of Association 31, 998-1001. turbulence and dispersion of pollutants near major road- Nagendra, S.M.S., Khare, M., 1999. Artificial neural network ways. Journal of Applied Meteorology 18, 283-293. technique in short-term air pollution modelling. In: Rao, S.T., Sistal, G., Keenan, M.T., Wilson, J.S., 1980. An Proceedings of the National Seminar on Wind Engineering, evaluation of some commonly used highway dispersion Department of Aerospace Engineering, Indian Institute of models. Journal of Air Pollution Control Association 30 (3), Technology, Kharagpur, December 2-3, 1999, pp. 96-110. 239-246. Nagendra, S.M.S., Khare, M., 2002. Artificial neural network Rao, S.T., Sistla, G., Eskridge, R.E., Petersen, W.B., 1986. based line source emission modelling: a review. In: Turbulent diffusion behind vehicles: evaluation of roadway Bandyopadhyay, J.N., Nagesh Kumar, D. (Eds.), Proceed- models. Atmospheric Environment 20, 1095-1103. ings of International Conference on Advances in Civil Rezler, K.J., 1989. Air pollution modelling. In: Cheremisinoff, Engineering: Water Resources and Environmental Engi- P.N. (Ed.), Encyclopedia of Environmental Control Tech- neering, Volume 1, Department of Civil Engineering, Indian nology, Air Pollution Control, Vol. 2. Gulf Publishing Institute of Technology, Kharagpur, January 3-5, 2002. Company Book Division, London. Allied Publishers, New Delhi, India, pp. 663-667. RIVM, 1991. National Environmental Survey 2: 1990-2010 Nelli, J.P., Messina, A.D., Bullin, J.A., 1983. Analysis and Publication of the National Institute of Public Health and modelling of air quality at street intersections. Journal of Environmental Protection (RIVM), P.O. 1, 3720, BA, Air Pollution Control Association 33, 760-764. Bilthorn, The Netherlands. Nicholson, E., 1975. A pollution model for street-level air. Rodden, J.B., Green, N.J., Messina, A.D., Bullian, J.A., 1982. Atmospheric Environment 9, 19-31. Comparison of roadway pollutant dispersion models using Nieuwstadt, F.T.M., 1980. Prediction of air pollution frequency the Texas data. Journal of Air Pollution Control Associa- distribution—III, the Gaussian plume model. Atmospheric tion 32 (12), 1226-1228. Environment 14, 259-265. Sculley, R.D., 1989. Vehicle emissions rate analysis for carbon Nieuwstadt, F.T.M., 1992a. A large-eddy simulation of a line monoxide hot spot modelling. Journal of Air Pollution source in a convective atmospheric boundary layer—I, Control Association 31 (10), 911-924. dispersion characteristics. Atmospheric Environment 26A Sedefian, L., Rao, T., 1981. Effects of traffic generated (3), 485-495. turbulence on near field dispersion. Atmospheric Environ- Nieuwstadt, F.T.M., 1992b. A large-eddy simulation of a line ment 15, 527-536. source in a convective atmospheric boundary layer—II, Segal, H., 1983. Microcomputer graphics in atmospheric dynamics of a buoyant line source. Atmospheric Environ- dispersion modelling. Journal of Air Pollution Control ment 26A (3), 497-503. Association 33 (6), 247-256. NILU, 1996. ROAD AIR 3.11 Users Manual. Norwegian Sharma, P., 1998. Air quality modelling for an urban Institute for Air Research, Kjeller, Norway. intersection of Delhi City. Ph. D. Thesis, Department of 2098 S.M.S. Nagendra, M. Khare / Atmospheric Environment 36 (2002) 2083-2098 Civil Engineering, Indian Institute of Technology, New Thompson, R.S., Eskridge, R.E., 1987. Turbulent diffusion Delhi, India. behind vehicles: experimentally determined influence of Sharma, P., Khare, M., 1999. Application of intervention vertex pair in vehicle wake. Atmospheric Environment 21 analysis for assessing the effectiveness of CO pollution (10), 2091-2097. control legislation in India. Transportation Research D 4, Tiao, G.C., Box, G.E.P., Harmming, W.I., 1975. Analysis of 427^132. Los Angeles photochemical smog data: a statistical over- Sharma, P., Khare, M., 2000. Real-time prediction of extreme view. Atmospheric Environment 25, 260pp. ambient carbon monoxide concentrations due to vehicular Trier, A., Firinguetti, L., 1994. A time series investigation of emissions using univariate linear stochastic models. Trans- visibility in an urban atmosphere—I. Atmospheric Environ- portation Research D 5, 59-69. ment 28, 991-996. Sharma, P., Khare, M., 2001a. Modelling of vehicular Van den Hout, K.D., Baars, H.P., Duijn, N.J., 1989. Effects of exhaust—a review. Transportation Research D 6, 179-198. buildings and street on air pollution by road traffic. In: Sharma, P., Khare, M., 2001b. Short-time, real-time prediction Basser, L.J., Mulder, W.C. (Eds.), Proceedings of the Eighth of extreme ambient carbon monoxide concentrations due to World Clean Air Congress, Vol. 4. Elsevier, The Hague, The vehicular exhaust emissions using transfer function noise Netherlands, Amsterdam. models. Transportation Research Part D6, 141-146. Waller, R.E., Commins, B.T., Lawther, P.J., 1965. Air Sharma, P., Khare, M., Chakrabarti, S.P., 1999. Application of pollution in a city street. British Journal of Industrial extreme value theory for predicting violations of air quality Medicine 22, 128-138. standards for an urban road intersection. Transportation Ward, C.E., Rangier, A.J., Shirley, E.C., 1977. CALINE 2—an Research D 4, 201-216. improved micro-scale model for the dispersion of air Sharma, V., Myrup, O.L., 1975. Diffusion from a line source in pollutants from a line source. Federal Highway Adminis- an urban atmosphere. Atmospheric Environment 9, tration Report, FHWA-RD-77-74, Washington, DC. 907-922. Yu, L.E., Hildemann, L.M., Ott, W.R., 1996. A mathematical, Shi, J.P., Harrison, R.M., 1997. Regression modelling of hourly model for predicting trends in carbon monoxide emissions NOx and NO2 concentration in urban air in London. and exposure on urban arterial highways. Journal of Air Atmospheric Environment 31 (24), 4081^1094. and Waste Management Association 46, 430-440. Singh, M.P., Goyal, P., Basu, S., Agarwal, P., Nigam, S., Zamurs, J., Piracci, R.J., 1982. Modelling of carbon monoxide Kumari, M., Panwar, T.S., 1990. Predicted and measured hot spots. Journal of Air Pollution Control Association 32 concentrations of traffic carbon monoxide over Delhi. (9), 947-954. Atmospheric Environment 24A (4), 801-810. Zanaidi, M.A., Singh, M.P., Karim, M., 1991. Traffic CO Sivacoumar, R., Thanasekaran, K., 2001. Comparison and dispersion pattern in Kuwait. Atmospheric Environment performance evaluation of models used for vehicular 25A (5/6), 909-914. pollution prediction. Journal of Environmental Engineer- Zanneti, P., 1990. Air pollution modelling, theories, computa- ing, ASCE 127 (6), 524-530. tional methods available software. Van Nostrand Remhold, Stukel, J.J., Soloman, R.L., Hudson, J.L., 1975. A model for New York. the dispersion of particulate or gaseous pollutants from a Zhang, Y., Bishop, G.A., Stedman, D.H., 1994. Automobile network of streets and highways. Atmospheric Environment emissions are statistically '/ distributed. Environmental 9, 990-999. Science and Technology 28 (7), 1370-1374. Sutton, O.G., 1932. A theory of eddy diffusion in the Zimmerman, J.P., Thompson, R.S., 1975. User's Guide for atmosphere. Proceedings of the Royal Society of London HIWAY. A Highway Air Pollution Model, EPA-650/4-74- A 135 (826), 143. 008.