ARTICLE IN PRESS
Atmospheric Environment 37 (2003) 5165–5175
Dispersion from road tunnel portals: comparison of two different modelling approaches Dietmar Oettla,*, Peter Sturma, Raimund Almbauera, Shin’ichi Okamotob, Kenji Horiuchic a
Institute for Internal Combustion Engines and Thermodynamics, Graz University of Technology, Inffeldgasse 25, Graz 8010, Austria b Tokyo University of Information Sciences, Yatocho, Wakaba-ku Chiba, Japan c Chiyoda Engineering Consultants Co., Ltd., Japan Received 11 November 2002; received in revised form 8 January 2003; accepted 20 January 2003
Abstract Two different models for the dispersion of tunnel exhausts are compared with each other. The first model was developed in Japan by several research institutions, and the second model was developed in Austria at the Institute of Internal Combustion Engines and Thermodynamics. To the best of the authors’ knowledge, these are the only dispersion models developed for that specific purpose. Significant differences in the modelling approach concern the basic physical assumptions as well as the formulation regarding the dispersion process. While the Japanese model incorporates the jet flow from the tunnel portal in a mass-consistent wind field model, the Austrian model incorporates the jet flow in the dispersion process itself. Further, the Austrian model assumes a ‘‘bending jet flow’’ in dependence on the ambient wind field, while the Japanese model does not. Since the ambient wind field fluctuates around a mean value, the position of the jet flow does also. This is an additional dispersion process treated in the Austrian model too. For the evaluation study, three tracer experiments performed at the Ninomiya, the Hitachi, and the Enrei tunnel in Japan were used. Comparison with observed concentrations reveals the ability of both models to simulate the dispersion of pollutants in the vicinity of road tunnel portals, although the Austrian model performs slightly better as indicated by e.g. the correlation coefficient, or the linear regression analysis. r 2003 Elsevier Ltd. All rights reserved. Keywords: Lagrangian model; Taylor–Galerkin method; Tunnel portal; Road traffic; Air quality; Jet flow
1. Introduction Roadway tunnels become ever more the solution to keep away the increasing traffic from residential areas (especially in urban areas). In many cases, the thrust of the vehicles driving in a tunnel is sufficient for the ventilation to meet imposed air quality standards inside a tunnel. Nevertheless, in the vicinity of roadway portals problems concerning the air quality may arise due to the concentrated release of polluted air there. Hence, it is *Corresponding author. Tel.: +43-316-873-8081; fax: +43316-873-8080. E-mail address:
[email protected] (D. Oettl).
important to have suitable tools to assess the dispersion from roadway tunnels for planning purposes. At first sight, the most appealing approach to simulate the dispersion from tunnel exhausts would be the usage of a three-dimensional microscale-model solving the Reynolds averaged Navier–Stokes equations (i.e. RANS-models). As already pointed out in Oettl et al. (2002) such models suffer from certain disadvantages: (i) high CPU-requirements; (ii) complicated to handle; and (iii) tendency to overestimate concentrations. The latter is a result of neglecting the influence of the ambient wind fluctuations on the position of the jet flow from the portal, since such models are usually used to compute steady-state flow fields.
1352-2310/$ - see front matter r 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2003.09.003
ARTICLE IN PRESS 5166
D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
In this paper, two substantially different modelling approaches are evaluated and compared: The Japanese Highway Public Corporation air quality dispersion model for tunnel portals (JH-model, Okamoto et al., 1998) and the dispersion model proposed by Oettl et al. (2002, GRAL=Grazer Lagrange model). Comparisons were made for three tracer experiments taken by the Japanese Highway Public Corporation. All tunnel portals in consideration are situated in highly complex terrain. To the best of the authors knowledge, the JHmodel and the GRAL model are the only dispersion models for roadway tunnel portals developed for regulatory purposes which appeared in literature, apart from the mentioned RANS-models. Other approaches, which are all empirical, such as the German Guideline (MLuS-92, 1998), may be considered as ‘‘rule of thumb’’ models intended to get a rough estimate of the concentration distribution around a portal.
2. Modelling approaches 2.1. Japanese highway public corporation model A basic description of the Japanese highway public corporation model (JH-model) can be found in Okamoto et al. (1998). The model has been revised recently, and it can now be applied to buoyant jet flows from tunnel portals, too (Okamoto et al., 2001). The model consists of a wind field and a diffusion module. The wind fields are calculated with the MASCON variational wind field model, developed by Kimura et al. (1996). The wind field model was chosen from a number of existing models, after a comparative study on the prediction accuracy and the required CPU times. The interpolated wind field from observational data is combined with the empirically determined tunnel jet flow from a tunnel portal. The latter is assumed to follow a Gaussian distribution according to: ! ! y2 z2 U ¼ U0 expðkxÞexp 2 exp 2 ; ð1Þ 2sy 2sy where U is the wind speed at a certain grid point, U0 is the exit velocity at the portal, k is called the distance decrement coefficient, which varies by ambient wind direction, wind velocity, and traffic conditions (Ueyama, 1985), sy and sz are the standard deviations of the horizontal and vertical wind components determined from a scale model experiment (Koso and Ohashi, 1978), x is the space coordinate along the jet flow, y is the coordinate perpendicular to it, and z is in the vertical direction. An initial evaluation study (Matsumoto et al., 1998) indicated that the predictive performance was poor for stable atmospheric conditions. This was reasoned with
the neglecting of buoyancy effects of the exhaust plume in the model. The problem has been overcome by presuming a constant rise of the jet flow in the vicinity of the tunnel portal, which is calculated as a function of the temperature difference between ambient air and the tunnel jet flow. After the calculation of a massconsistent wind field, the diffusion is computed using the Taylor–Galerkin–Forester-filter method, which was found to perform better than the Taylor–Galerkin technique, since it does not produce computational ripples. In the JH-model, the jet flow is assumed to align along the street without changing its direction. The horizontal eddy diffusivities are calculated as follows: Ky ¼
us0y 2 ; 2x
ð2Þ
where u is assumed to be 3 m s1, x is the downwind distance in (m), and s0y is taken from the Pasquill– Gifford (PG) diagram for a downwind distance of 100 m. The vertical eddy diffusivity is set equal to 2.0 m2 s1 according to Eskridge et al. (1979) at the lowest two levels of grid cells within the roadway. Anywhere else the vertical diffusion coefficients are calculated according to the method described by Shir and Shieh (1974). 2.2. Model of Oettl et al. (2002) (GRAL=Grazer Lagrange model) The concept of the GRAL model is quite different to that of the JH-model. Its basic idea is a bending jet flow as a function of the ambient wind field. The bending jet flow is modelled by the tracks of numerous particles released from a portal. It is assumed, that the jet flow changes its direction according to two forces: (i) frictional forces, which cause the jet flow to slow down, and (ii) pressure forces perpendicular to the jet flow, due to the ambient wind field, which finally change the direction of the tunnel jet flow. Most important to notice is a dispersion effect, which is not accounted for in the JH-model: Since the ambient wind field varies instantaneously, the position of the jet flow does also. This is kind of a diffusion effect, which is assumed to be of higher importance than the diffusion due to entrainment of ambient air in the tunnel jet flow. Especially for low wind speeds (uo2 m s1), where large wind direction fluctuations can be observed (e.g. Etling, 1990), the position of the jet flow varies significantly. Such variations may be of the order of tens of metres, while the eddies involved in the entrainment of ambient air in the tunnel exhausts are of the order of metres. When simply comparing these length-scales it may be concluded that this effect enhances the dispersion of the tunnel exhausts by one order of magnitude compared to the dispersion by entrainment only.
ARTICLE IN PRESS D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
Simulations with a RANS-model, not taking into account meandering effects, confirmed that hypothesis (Oettl et al., 2002). Simulated concentrations were found to be about one order of magnitude larger than those observed or simulated with the GRAL model for a tracer experiment taken near a highway tunnel portal in Austria. For convenience, the described diffusion effect will be abbreviated by the acronym ADAPT (=Additional diffusion by the influence of a fluctuating ambient wind field on the position of the jet flow). Buoyancy is taken into account in the GRAL model by making use of a modified version of Van Dop’s model for plume rise for stacks (Van Dop, 1992). The model is described in detail in Oettl et al. (2002). Some empirical parameters in the GRAL model were determined as to obtain good results for the tracer experiments in Austria mentioned above. In the following evaluation of the JH-model and the GRAL model none of such empirical parameters were changed. In the course of the evaluation study it became apparent, that the wakes induced by the vehicles may not be negligible in the very vicinity of the tunnel portal and the road, respectively. In all of the three Japanese tracer experiments, many samplers were placed at the edges of the roads or close to the portals. Hence, to obtain good model performance for those receptor points, it was necessary to incorporate the effect of wakes in the GRAL model. This was done by assuming that a certain fraction of the tunnel exhausts (represented by the same fraction of particles released from a portal in the model) moves with the wakes of the vehicles, and does only experience small changes in flow direction. The position of the jet flow centre line is assumed to be governed by a friction force along the jet flow: dUp q2 Up ¼ K dt qy2
ð3Þ
and a pressure force perpendicular to the jet flow dUnS 1 2 ¼ bUnA dt 2
ð4Þ
which causes the jet flow to slow down and to bend according to the ambient wind direction. In Eqs. (3) and (4) K is the turbulent exchange coefficient (m2 s1), UP is the jet flow velocity (m s1) defining the x-axis, UnS is the jet flow velocity in y-direction (m s1), UnA is the ambient wind speed in y-direction of the jet flow, and b (m1) is calculated using: b ¼ gð1 þ tÞ;
ð5Þ
where g (m1 s) is an empirical constant. The coordinate system is re-orientated after each time-step, such that x is always orientated along the jet flow, and y is perpendicular to it. The turbulent exchange coefficient is taken to be time-dependent (Oettl et al., 2002) and
5167
reads: K ¼ að1 þ tÞ;
ð6Þ
2 2
where a (m s ) is another empirical constant. a and g are taken to be 1 for all simulations of the bending jet flow. For the calculation of the vehicle-induced effects, a and g were set equal to 0.5. The fraction of material (i.e. particles in the model) captured in the wakes is taken dependent on the number of vehicles driving actually on the highway. It varied between 10% and 35% for the three tracer experiments described in the next section. The ambient wind field was analysed by means of a mass-consistent wind field model (Oettl et al., 2000). A major weakness of the GRAL model is the limitation of using only one single ambient wind vector in the calculation of the jet flow. Here, the ambient wind vector at the location of the tunnel portal is used, which is provided by the wind field analysis. The reason, why only one ambient wind vector can be used is that only the horizontal components of the ambient wind are used to calculate the geometry of the jet flow. In 3-D flows, there might exist regions with converging flows leading to significant vertical winds. In such regions, the computed jet flow according to Eqs. (3)–(6) would not satisfy mass conservation, if the actual horizontal ambient wind field is taken at each particles location. Clearly, errors in the geometry of the modelled jet flow result in cases of a strongly inhomogeneous ambient wind field at the portal region. The simulation of the ADAPT in the GRAL model requires knowledge about the standard deviations of the horizontal wind component fluctuations. Since they were not available for the experiments, the following parameterisations were utilized (e.g. listed in Zannetti, 1990): Neutral conditions : z su ¼ sv ¼ 1:3u 1 ; zi Convective conditions : z 1=3 ; su ¼ sv ¼ u 12 þ 0:5 L
ð7Þ
ð8Þ
Stable conditions : su ¼ sv ¼ 1:3u e2fz=u ;
ð9Þ
Surface layer : su ¼ sv ¼ 2:5u :
ð10Þ
In Eqs. (7)–(10), L is the Monin–Obukhov length, u is the friction velocity, zi is the mixing height, f is the Coriolis parameter, and z is the height above ground level. Eqs. (7)–(10) are not valid under low wind speed conditions. Usually, using Eqs. (7)–(10) in such conditions leads to an underestimation of the standard deviations. Hanna (1990) suggested to set a lower limit
ARTICLE IN PRESS 5168
D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
for the standard deviations at 0.5 m s1, which was also done in the GRAL model. The friction velocity u and the Monin–Obukhov length L were estimated from the stability class, the roughness length, and the wind speed by means of the relationships given in Venkatram (1996) and Golder (1972). It should be emphasized, that the relations given in Eqs. (7)–(10) introduce additional uncertainty as they were not derived under the conditions prevailing in this study (i.e. complex terrain, proximity of traffic).
3. Experimental and model set-up Three tracer experiments taken by the Japanese Highway Public Corporation in the years 1994 and 1995 were available for comparison purposes. Table 1 lists some relevant information regarding the three tunnels and the tracer experiments. Fig. 1 shows the locations of the samplers, the tunnel portals and the terrain, where the tracer tests were performed. The portals lie all in rather steep topography. About 10 meteorological monitoring stations were placed in the surroundings of each of the portals, to provide an input for the wind field models. The varying meteorological conditions (wind speed, wind direction, and stability) during the SF6 tracer tests (Table 2) allowed for a critical evaluation of the models. SF6 was released inside the tunnels. It is assumed that the tracer gas was immediately mixed with the ambient air in the tunnel as a consequence of the high turbulence introduced by the moving vehicles there. The horizontal grid size used by the JH-model was 5 m 5 m for the central 20 grid points and 30 m for the most outer grid cells. The cell heights increased from 3 m near the surface up to 200 m at the top of the computational domain. The time increment Dt was set as to maintain the numerical scheme stable, that was, 0.25–0.5 s depending on the Courant number. Since the areas considered cover several hundred square metres, stationary concentration fields were obtained usually after 360-time steps (i.e. 180 s) for wind speeds between 1 and 3 m s1. The grid size used in the GRAL model was
taken to be 4 m 4 m 0.5 m, while for the wind field model the horizontal grid size was 16 m 16 m, and for the vertical direction variable cell heights between 5 and 100 m between the lowest and highest levels were used. It should be mentioned that a higher resolution in the GRAL model does not result in increasing computation times, but does only require more memory space. In the computations 105 particles were released in each of the runs, and typical CPU times were around 3 min on a conventional PC. Concentration fields were obtained by counting the particles and weighted with the time spent in each cell. Sensitivity analysis revealed that 105 particles were enough to prevent statistical uncertainty in the concentration fields. Measured temperature differences between the ambient air and the air inside the tunnels were not available for the simulations but were estimated, such that around noon and the early afternoon temperature differences were assumed to be equal to zero, while at the other times, positive temperature differences of about 5 K were supposed (tunnel air warmer than ambient air). In case of the Ninomiya tunnel, temperature differences were assumed to be zero, due to the small length of the tunnel.
4. Model evaluation 4.1. Overall statistics In a first step the overall performance of the models was evaluated for each of the three tunnels. As statistical measures the linear regression, coefficient of determination R2 ; fractional bias FB, normalized mean square error NMSE, and the fraction of data, where the ratio between computed concentrations and observations laid in the interval 0:5pCcalc =Cobs p2:0 were utilized. It should be mentioned, that all statistical measures are very sensitive regarding the highest concentrations, mainly occurring in the very vicinity of the portal (distance o10 m) and at the edges of the roads out of the tunnels. For practical applications, those locations are of minor interest, since that areas
Table 1 Some relevant information concerning the tracer experiments at the three tunnel sites in Japan
Length and ventilation system Highway Traffic volume Experiment date No. of sampling sites: SF6 No. of runs Tracer release period (h)
Ninomiya tunnel
Hitachi tunnel
Enrei tunnel
445 m () Odawara-Atsugi road 30,000 veh./day 20 Jan.–1 Feb. 1994 64 21 144
2439 m (jet fan) Joban expressway 24,000 veh./day 3–9 Feb. 1995 85 18 159
1800 m (jet fan) Chuo expressway 32,000 veh./day 23–29 Nov. 1995 86 17 168
ARTICLE IN PRESS D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
5169
Fig. 1. The experimental sites: dots indicate the sampling points for the SF6 tracer tests (contour line is 5-m interval in height).
are not used for residential purposes. In other words, the statistical analysis presented here reflects mainly the performance of the models very close to the roads and the tunnel portals. Nevertheless, for the analysis presented, all samplers were considered with exception of the one inside the tunnel and the one(s) placed directly at each portal. The sampling point inside the tunnel represents the dilution of the vehicle emissions in the tunnel due to the ventilation, which is by definition
calculated exactly by both models, because the measured source strengths and jet flow velocities were used as input to the models. The sampling points placed directly at the corners of the portals cause problems in the comparison, as very sharp concentration gradients exist there, which cannot be resolved by the models accurately. Scatter plots for each tunnel are shown in Fig. 2, and the corresponding statistical measures are listed in
5170
Table 2 Meteorological and traffic conditions during the SF6 tracer tests Tunnel
Wind
Stability
Tunnel
Hitachi tunnel (1995)
WD
WS (m s1)
R1-1
WSW
5.7
C–D
5.5
R1-2 R1-3 R1-4 R1-5 R1-6 R1-7 R1-8 R1-9 R1-10 R1-11 R1-12 R1-13 R1-14 R1-15 R1-16 R1-17 R1-18 R1-19 R1-20 R1-21
WSW NE NE NE NE NE NE ENE NNE WNW WNW NE E SE SE SSE SSE SW SW W
4.9 3.6 3.4 3.7 3.8 3.5 4.9 2.0 2.6 3.8 3.1 1.2 1.2 2.0 2.6 2.1 2.2 2.0 2.2 2.9
D D D D D C D F F B–C B–C A–B A–B B A–B B B C F D
5.2 5.3 5.1 4.8 4.8 4.6 4.2 4.2 4.2 4.3 4.1 4.8 4.9 4.8 4.7 4.9 4.9 5.0 5.0 5.0
Run
Wind
Stability
Tunnel jet flow (m s1)
Tunnel
Enrei tunnel (1995)
WD
WS (m s1)
R2-1
SE
1.0
D
4.8
R2-2 R2-3 R2-4 R2-5 R2-6 R2-7 R2-8 R2-9 R2-10 R2-11 R2-12 R2-13 R2-14 R2-15 R2-16 R2-17 R2-18
WNW SE E NNW NNE ESE NE W WNW NW E WNW W SW NE N E
2.7 1.1 1.0 2.4 2.1 1.0 1.3 0.8 2.6 3.8 1.0 2.5 2.7 1.1 1.6 0.5 0.4
D D D D D D G D B D G D B D G G D
4.5 4.0 4.3 3.8 4.1 4.9 4.2 5.6 4.8 4.8 4.5 5.4 5.2 4.9 4.3 3.6 4.4
Run
Wind
Stability
Tunnel jet flow (m s1)
WD
WS (m s1)
R3-1
W
3.8
B-C
6.3
R3-2 R3-3 R3-4 R3-5 R3-6 R3-7 R3-8 R3-9 R3-10 R3-11 R3-12 R3-13 R3-14 R3-15 R3-16 R3-17
SW SW WNW ESE ESE SSW NW NW WNW WNW WNW WNW E W WNW WNW
3.5 3.5 2.0 3.9 3.9 3.3 4.0 3.7 4.3 4.8 6.2 4.4 0.9 2.6 4.6 3.6
D B-C F B-C B-C B-C C-D C C-D C-D D D G B C-D E
6.0 6.0 4.3 5.5 5.7 5.3 5.8 6.1 6.3 5.8 5.4 5.6 6.5 6.0 5.6 5.8
ARTICLE IN PRESS
Tunnel jet flow (m s1)
D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
Niniomiya Tunnel (1994)
Run
ARTICLE IN PRESS D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
5171
Fig. 2. Scatter plots of hourly tracer concentrations for the JH-model (left) and the GRAL model (right), respectively. The solid line marks the linear regression (top: Ninomiya tunnel; middle: Hitachi tunnel; bottom: Enrei tunnel).
Table 3. Regarding the coefficient of determination, both models performed less good for the Hitachi tunnel, although the coefficients of determination of 0.50 for the JH-model and 0.68 for the GRAL model are still in an acceptable range. The statistical measures indicate the ability of both models to simulate the dispersion from road tunnel exhausts. However, the performance of the GRAL model is somewhat better than the one of the JH-model.
In particular, the coefficient of determination could serve as an indicator, that the dispersion process is modelled more realistic by the ‘‘bending plume’’ approach, while the slope b in the linear regression analysis may give a hint, that the ADAPT incorporated in the GRAL model does also improve the results. The ‘‘bending plume’’ approach is also supported by simulations made previously with a microscale model based on the RANS equations, which showed also
ARTICLE IN PRESS D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
5172
Table 3 Statistical measures for the overall performance of the models for the three tunnels Model
Tunnel
No. of data
R2
Regression Cobs ¼ a þ bCcalc Constant a
Slope b
Factor 2
NMSE
FB
JH-model
Ninomiya Hitachi Enrei
1117 1466 1232
0.72 0.50 0.53
0.55 0.39 2.29
0.71 0.60 0.71
0.21 0.32 0.55
2.09 2.32 3.02
0.11 0.25 0.23
GRAL model
Ninomiya Hitachi Enrei
1117 1466 1232
0.81 0.68 0.69
1.17 0.41 0.18
0.90 0.86 1.15
0.40 0.40 0.51
1.47 1.57 2.85
0.33 0.12 0.27
Table 4 Statistical measures for specific meteorological conditions obtained with the JH-model Tunnel
Condition
No. of case
No. of data
R2
NMSE
Regression Cobs ¼ a þ bCcalc Constant a
Slope b
Factor 2
FB
Ninomiya
Stability
Unstable Neutral Stable
12 4 5
669 231 217
0.69 0.85 0.80
2.4 1.2 2.0
0.81 0.65 0.58
0.84 0.76 0.56
0.20 0.22 0.26
0.25 0.12 0.16
Hitachi
Stability
Unstable Neutral Stable
2 13 3
154 1066 246
0.57 0.50 0.55
2.2 2.2 3.2
0.07 0.49 0.08
0.88 0.60 0.52
0.39 0.34 0.26
0.07 0.21 0.56
Enrei
Stability
Unstable Neutral Stable
9 5 3
678 356 198
0.57 0.73 0.56
2.4 2.4 6.2
1.83 1.90 2.74
0.86 1.03 0.39
0.55 0.64 0.38
0.10 0.10 0.98
bending jet flows according to the ambient wind field (Oettl et al., 2002). The empirical constants a and g; which were determined on the basis of the tracer experiments at the Ehrentalerbergtunnel in Austria (Oettl et al., 2002) need not be changed raising the hope that they will also result in good fits of the plume spread for other tunnel portals. The tendency of the JH-model to overestimate the observed concentrations is expressed by the slope b of the linear regression analysis, which is in all cases significantly lower than 1. On the other side, the linear regression concerning the results obtained with GRAL shows values for the slope b closer to 1. The relatively large scatter in all cases (Fig. 2) leads to the conclusion, that the modelled pollutant dispersion on a single case basis is always associated with large uncertainty. Hence, model predictions of peak concentrations should always be considered of being merely more than rough estimates. This is of importance, if air quality standards define threshold values for peak concentrations (e.g. in Austria there exists a threshold value for NO2 based on mean half-hourly concentrations).
4.2. Statistics separated according to meteorological conditions The second step of the analysis comprises the calculation of the same statistical measures as in step one, but separated according to specific meteorological conditions. Results for both models are listed in Tables 4 and 5. The JH-model seems to perform better in unstable and neutral conditions, while for the stable cases the model has a stronger tendency to overestimate concentrations as indicated by the slope b of the linear regression analysis and the FB. The GRAL model has a rather clear tendency to underestimate concentrations in unstable conditions and a slight tendency to overestimate concentrations in stable conditions. However, the overestimations for the stable cases are much less pronounced compared with the JH-model. Again, the incorporation of the ADAPT seems to enhance the simulation results for stable conditions. This outcome is not unexpected as the effect of the ADAPT should be largest during low wind speeds, which often go with stable atmospheric conditions, as horizontal wind
ARTICLE IN PRESS D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
5173
Table 5 Statistical measures for specific meteorological conditions obtained with the GRAL model Tunnel
Condition
No. of case
No. of data
R2
NMSE
Regression Cobs ¼ a þ bCcalc Constant a
Slope b
Factor 2
FB
Ninomiya
Stability
Unstable Neutral Stable
12 4 5
669 231 217
0.81 0.79 0.79
1.6 1.3 1.4
1.06 0.96 1.68
0.92 0.87 0.87
0.40 0.46 0.31
0.34 0.23 0.41
Hitachi
Stability
Unstable Neutral Stable
2 13 3
154 1066 246
0.61 0.70 0.63
3.3 1.4 2.0
0.03 0.50 0.03
1.10 0.86 0.77
0.36 0.41 0.40
0.38 0.15 0.23
Enrei
Stability
Unstable Neutral Stable
9 5 3
678 356 198
0.69 0.81 0.59
3.0 2.5 2.9
0.32 0.03 0.01
1.20 1.31 0.76
0.40 0.59 0.43
0.40 0.30 0.28
direction fluctuations generally increase with decreasing wind speeds. Relatively low wind speeds occurred also during the neutral cases of the Hitachi experiments (Table 2). The JH-model again tends to overestimate the observed concentrations for these cases, while the GRAL model performs better here. On the contrary, the neutral cases during the experiments at the Enrei tunnel were accompanied by relatively strong winds around 4.5 m s1. These conditions can be modelled by the JH-model particularly well, while the GRAL model has the largest underestimations of observed concentrations for high wind speed situations as indicated by the slope b of the linear regression analysis. 4.3. Evaluation of average concentration distributions Of particular interest is the ability of the models to simulate mean concentrations, since most air quality guide lines are based on limiting values, for e.g. annual mean concentrations. Generally, dispersion models should be able to simulate average concentrations better than hourly concentration fields, since models are based on average statistics of turbulence quantities. Further, if a model produces no systematic deviations from observations, model uncertainties should be reduced by taking the average over the concentration fields. Results of calculated mean concentration fields are shown in Fig. 3 for both models. As expected, the performance of the models increases for averaged concentration fields. Coefficients of determination are high and read 0.96, 0.88, and 0.92 for the Ninomiya, Hitachi, and Enrei tunnel. The fraction of concentrations, that could be modelled within to a factor of two, was 0.44 (Ninomiya), 0.66 (Hitachi), and 0.77 (Enrei). The corresponding results for the JH-model are 0.90, 0.78, and 0.74 regarding the coefficients of determination for the Ninomiya, Hitachi, and Enrei tunnel. The
fraction of concentrations, that could be modelled within to a factor of two, was 0.56 (Niniomiya), 0.40 (Hitachi), and 0.51 (Enrei). When using meteorological statistics in environmental assessment studies to compute concentration distributions, several hundreds of cases have to be simulated. It can be assumed that the performance of the models regarding average concentrations will further improve parallel with the number of cases calculated.
5. Conclusions The comparison of the calculated concentrations by the JH-model and the GRAL-model versus observed concentration distributions shows the ability of both models to be used for regulatory purposes. Some statistical measures, such as the coefficients of determination, or the linear regression analysis, point in the direction, that the ‘‘bending plume’’ approach of the GRAL model in combination with the ADAPT results in slightly more realistic concentration distributions around a portal, than does the JHmodelling approach, where a Gaussian shape of the jet flow is assumed. This is especially the case for stable dispersion conditions, where wind speeds are generally found to be low, and thus, wind direction fluctuations are largest. It was also found, that very steep gradients in the concentrations exist very close to the portals, which cannot be resolved by the models exactly. For instance, concentrations varied about two orders in magnitude over a distance of only 10 m. When it is important, to know the concentration distribution along the road coming out of the tunnel, the effect of the wakes induced by the vehicles on the road on the dispersion may not be neglected.
ARTICLE IN PRESS 5174
D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
Fig. 3. Scatter plots of mean concentrations for the JH-model (left) and the GRAL model (right), respectively. The solid line marks the linear regression (top: Ninomiya tunnel; middle: Hitachi tunnel; bottom: Enrei tunnel).
RANS models seem not to be a proper choice for modelling dispersion from tunnel portals, since they do not account for the ADAPT, which generally results in narrow plumes and overestimations of concentrations along the jet flow centre line (Oettl et al., 2002). Although the ADAPT is not exclusively incorporated in the JH-model, it seems as it is partly absorbed in the empirically determined horizontal eddy diffusivities. In comparison with RANS models, the computational times for both models are quite low. Approximately 3 min are needed by both models to simulate one single
case on a conventional personnel computer. Thus, the application to meteorological statistics, which are necessary to compute long-term concentration values (e.g. annual means), is generally possible with both models.
Acknowledgements The provision of the tracer data by the Japanese Highway Public Corporation is greatly acknowledged.
ARTICLE IN PRESS D. Oettl et al. / Atmospheric Environment 37 (2003) 5165–5175
The joint-project was partly funded by the Austrian science fund (No. 14075-TEC). Many thanks also to the referees for their very helpful comments.
References Eskridge, R.E., Binkowski, F.S., Hunt, J.C.R., Clark, T.L., Demerjian, K.L., 1979. Highway modelling Part II: advection and diffusion of SF6 tracer gas. Journal of Applied Meteorology 18, 401–412. Etling, D., 1990. On plume meandering under stable stratification. Atmospheric Environment 24, 1979–1985. Golder, D., 1972. Relations among stability parameters in the surface layer. Boundary-Layer Meteorology 3, 47–58. Hanna, S.R., 1990. Lateral dispersion in light-wind stable conditions. Il Nuovo Cimento 13, 889–894. Kimura, F., Sakai, K., Matsumoto, K., Horiuchi, K., Kobayashi, K., Okamoto, S., 1996. Comparative simulation of wind over complex terrain using the mass conservation method (MASCON) and non-linear models based on the fluid dynamic equations. Preprints of the Fifth International Conference on Atmospheric Science and Air Quality, Seattle, USA. Koso, T., Ohashi, H., 1978. Disturbed flow diffusion at grounded semicircular jets. Transactions of the Japan Society of Mechanical Engineers 44, 1266–1274. Matsumoto, K., Sakai, K., Horiuchi, K., Takahashi, H., Okamoto, S., 1998. Advanced air quality simulation model for road tunnel portals in complex terrain. International Journal of Vehicle Design 20, 71–78. MLuS-92, 1998. Merkblatt uber . Luftverunreinigungen an StraXen; Teil: StraXen ohne oder mit lockerer Randbebauung; Stand 1998, p. 41 (available from: Bundesanstalt
5175
fur . StraXenwesen, BruderstraXe . 53, 51427 Bergisch Gladbach, Germany). Oettl, D., Almbauer, R.A., Sturm, P.J., Piringer, M., Baumann, K., 2000. Analysing the nocturnal wind field in the city of Graz. Atmospheric Environment 35, 379–387. Oettl, D., Sturm, P.J., Bacher, M., Pretterhofer, G., Almbauer, R.A., 2002. A simple model for the dispersion of pollutants from a road tunnel portal. Atmospheric Environment 36, 2943–2953. Okamoto, S., Sakai, K., Matsumoto, K., Horiuchi, K., Kobayashi, K., 1998. Development and application of a three-dimensional Taylor–Galerkin numerical model for air quality simulation near roadway tunnel portals. Journal of Applied Meteorology 37, 1010–1025. Okamoto, S., Hada, Y., Konno, Y., Kobayashi, K., Horiuchi, K., 2001. Evaluation of the JH air quality simulation model for tunnel portals. Proceedings of the Seventh International Conference on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes, 28–31 May, 2001, Belgirate, Italy, pp. 229–233. Shir, C., Shieh, L.J., 1974. A generalized urban air pollution model and its application to the study of SO2 distributions in the St. Louis metropolitan area. Journal of Applied Meteorology 13, 185–204. Ueyama, S., 1985. Dispersion estimates for automobile exhaust gas from a tunnel portal. Japan Highway Public Corporation Technical Note 79, pp. 38–45 (in Japanese). Van Dop, H., 1992. Buoyant plume rise in a Lagrangian framework. Atmospheric Environment 26A, 1335–1346. Venkatram, A., 1996. An examination of the Pasquill–Gifford– Turner dispersion scheme. Atmospheric Environment 30, 1283–1290. Zannetti, P., 1990. Air Pollution Modelling. Theories, Computational Methods and Available Software. Computational Mechanics Publications, Southhampton, Boston, p. 444.
