Evaluating a Canadian regional air quality model using ground-based observations in north-eastern Canada and United States

Abstract

The simulated concentrations from a numerical 3-dimensional regional air quality model (MC2AQ) are compared to those of ground-based observations in north-eastern Canada and the United States. The model has oxidant chemistry for both inorganic and organic species and deposition routines driven online by a mesoscale compressible community meteorological model (MC2). A standard emission inventory of anthropogenic, natural and biogenic sources for the year 1990 for 21 atmospheric trace species was used in the simulation. The model was run for July 1999, because of the occurrence of a high ozone episode and the availability of the monitoring data for surface O₃, SO₂, NO, NO₂ and NO_x. The comparisons during the episode show that the model performs quite well for predicting concentrations and diurnal variations of the surface ozone. The predictions for other gaseous species show some discrepancies with observations, but they are consistent with the results from other models evaluated in the literature. The uncertainties in the emission inventory for these species might be the main causes of the discrepancies. Further studies are needed to improve the predictability of SO₂ and NO_x, especially as the model is developed to include particulate matter formation as a result of these gaseous precursors.

---

Introduction

The impact of ground level pollution on the environment and human health is of concern to the local and national governments of Canada and the United States, as well as many other countries worldwide.¹ High concentrations of ozone in the lower troposphere are considered an important issue of air quality and tropospheric chemistry.² Surface ozone concentrations during summertime in many regions of north-eastern Canada and the United States have been measured to exceed standard levels.³

Usually it is difficult to quantify the direct contribution of precursor species to the pollutant concentrations. Chemical species, after being directly emitted into the atmosphere from anthropogenic and natural sources, undergo complex non-linear physical-chemical transformations, in which new species can be formed from their precursors.⁴ Moreover, meteorology can also play important roles in the pollution formation processes. Biswas and Rao⁵ and Hogrefe et al.⁶ pointed out that photochemical models can be very sensitive to meteorological inputs after they examined the variability in the predictions of the Urban Airshed Model (UAM-V) when the meteorological input was derived from two different meteorological models, namely, RAMS (or RAMS3b) and MM5. Hogrefe et al.⁶ found that such a variability can be decomposed into intra-day, diurnal, synoptic and long-term meteorological effects. The observed concentrations at a given site are therefore the comprehensive result of emissions, chemical transformations, microphysical processes and meteorological actions.

Many regional air quality models have been developed so far to study ozone, and more recently particulate matter (PM). Jacobson et al.⁷ developed the model GATOR for modelling ozone and PM in the Los Angeles metropolitan area.^8–10 Other air quality models include CIT,¹¹ URM¹² UAM-V,¹³ CAMx,¹⁴ SAQM,¹⁵ MAQSIP,^16,17 SMRAQ¹⁸ and Models-3.¹⁹ Comparison studies for many air quality models can be found in Russell and Dennis.²⁰ In Canada, two air quality models have been developed: ADOM (Acid Deposition and Oxidant Model)²¹ and CHRONOS (Canadian Hemispheric and Regional Ozone and NO_x system).²²

Until now, pre-calculated meteorological fields were used to drive the transport of chemical species in most of the regional air quality models mentioned above. Such “off-line” models often include differences between the vertical and horizontal grids and sub-grid scale parameterizations can be inconsistent for temporal resolution.²³ An online chemistry modeling system, in which the meteorological fields are computed concurrently with the chemistry, will have significant advantages. In this paper, we present comparisons of results obtained from the online regional air quality modeling system MC2AQ to observation data from the routine ground-based monitoring network in Canada and the United States.

Model description

The air quality model MC2AQ was described by Plummer in his Ph.D. thesis,²⁴ and in Plummer et al.²⁵ as well as McConnell et al.²⁶ Here only some of the important features and new developments of the model are summarized.

Meteorological model

MC2AQ is based on the non-hydrostatic 3-dimensional Mesoscale Compressible Community model (MC2), developed by the Meteorological Service of Canada (MSC) in collaboration with the University of Quebec in Montreal.²⁷ It uses a semi-implicit treatment for gravity and acoustic waves, and a semi-Lagrangian treatment for advection. This combination enables long simulation timesteps in comparison with the Eulerian formulation and results in efficient and accurate solutions of the primitive equations.²⁸

Chemistry module

The governing equation for chemical trace species can be written as where X is the species mixing ratio, V is the wind velocity vector, K is the eddy diffusivity, X_p, X_l, X_e and X_d are the chemical production rate, chemical loss rate, time dependent emission rate and deposition rate, respectively.

Forty-seven inorganic and organic chemical species are involved in 98 chemical reactions and 16 photolysis reactions. The gas phase chemistry mechanism is that of Version II of ADOM.²¹ An operator-splitting approach is used to solve the system. Initial and boundary conditions for the chemical fields in the simulation were derived from global Chemical Transport Model (CTM) results.²⁴ Those species, which are only middle products of chemical reactions and do not appear in the CTM chemistry, were assigned near-zero values for their initial and boundary conditions.

Emissions

For this study, the 1990 emission inventory for 21 atmospheric trace species was used.²⁹ Anthropogenic emissions and emissions from natural sources such as forest fires were taken from standard pre-processed emission files developed for the ADOM speciation and the MC2 grid scheme (21.2 × 21.2 km² on a polar stereographic projection).²⁹ Biogenic emissions use BEIS2 in which emissions from 120 surface vegetation types^24,25,26 are driven by the model meteorology, whereas the anthropogenic emissions are prescribed. Annual emissions are partitioned by seasons and further divided into hourly emission rates for each of three day-of-the-week categories: weekday, Saturday and Sunday.

The emissions of 1985 and 1990 have been extensively tested and proven to be a reliable dataset for regional modelling studies within the Canadian modelling community.²⁵ Whilst the new emission inventory is now under construction, this 1990 inventory is up to now the best available data for this application.

Dry deposition

The dry deposition rates of species depend not only on meteorological conditions in the boundary layer and chemical properties of the species, but also on land surface types. Eight land use categories are used in MC2AQ: inland waters, deciduous broadleaf trees, evergreen needle-leaf trees, wetland with plants, crops and mixed farming, grass, urban and desert. The land surface type at each grid is composed of these eight categories. Further details on the treatment of deposition in the model can be found in Plummer.²⁴

Observation data

Data from the standard ground-based monitoring networks in Canada and the United States were used. The observation data for July 1999 were selected for this model scenario, because an extensive Canadian field campaign in southern Ontario took place during this time.³⁰ The goal of the field campaign was to obtain detailed hourly data for the comparison with PM estimated by regional models. The measured pollutants include O₃, SO₂, NO, NO₂, NO_x, PM_2.5 and PM₁₀. The current work focuses on the gas phase measurements. The Canadian data were obtained from the Analysis and Air Quality Division of Environment Canada. The hourly meteorological data for Ontario and Quebec were obtained from the Ontario Climate Centre of Environment Canada. The measured hourly pollutant concentrations and the meteorology data for the north-east United States in the model domain were extracted from the AIRS database in the United States Environmental Protection Agency's (US EPA). The number of monitoring sites for the domain in both Canada and the United States are given in Table 1, and their geographical locations are shown in Fig. 5.

Table 1 Number of monitoring sites within model domain

	United States	Canada
O3	194	70
NO2	62	46
NO	35	46
NOx	43	28
SO2	148	38
Wind	40	78

---

Another reason we chose observation data for July 1999 is that an occurrence of high ozone concentration at ground level is found in this month. Ozone episode formation is strongly influenced by meteorology.¹ Light surface winds, a strong subsidence inversion, sunny and warm weather associated with a high pressure system will minimize the dispersion of pollutants and provide favourable conditions for the photochemical generation of ozone.¹ Ozone episodes in southern Ontario occur predominately when a stationary high-pressure ridge lies to the east of the Great Lakes region. Under this condition, synoptic flow is from the south or southwest direction, placing southern Ontario downwind from regions with significant anthropogenic emissions in the American Midwest and Ohio valley. The ozone episode case in July 1999 was formed under similar conditions.

Modelling scenario

The model domain is an 83 × 83 horizontal grid with 21.2 km horizontal resolution and 25 vertical layers up to 20 km. The model has previously been run at a horizontal resolution of 5.3 km, 21 km, and 42 km over the domain covering most of eastern North America for resolution sensitivity tests, and the results can be found in Plummer.²⁴ Initial and boundary conditions for the chemical fields for 21 km resolution simulations were derived from the global CTM developed by McConnell et al.²⁶ More details about chemical and meteorological initialization can be found in Plummer.²⁴

The model was run for the entire month of July 1999 during a high ozone episode in this month, with concentrations above the national standards in both Canada and the United States. The results for the period from July 10 to 22 are presented to cover three four-days states: ‘before’, ‘during’ and ‘after’ the high ozone episode. The simulated ozone concentration was obtained by using bilinear interpolation. The comparison between observations and simulations in this period will help evaluate the model performance under pre-episodic, episodic and post-episodic conditions.

Comparisons and discussions

Our primary focus is on a comparison of simulated ozone with ground-based measurements since ozone is one of the main species of interest.² For the industrialized areas in the model domain for this work, anthropogenic emissions dominate in the emission inventories, especially for organic species that greatly contribute to ozone production. So it can be expected that the anthropogenic emissions have a major contribution to ozone concentrations.

Fig. 1 shows simulated ozone mixing ratios (in ppbv) at the peak of the episode, which corresponds to 2 pm Eastern Daylight Time (EDT) in eastern North America on July 17, 1999. The factors driving this ozone episode have been mentioned above. The maximum ozone level can be found along the eastern seaboard, in western Pennsylvania and in West Virginia. Ozone concentrations in the areas east of Sarnia and north of Lake Ontario exceed the 82 ppbv national standard in Canada. Fig. 2 shows the wind direction over the model domain corresponding to the conditions for Fig. 1. The prevailing winds are from the west over most of the model domain, from the southwest in southern Ontario and along the Saint Lawrence River, and from the south over the eastern seaboard. A stable high-pressure system off the eastern coast was responsible for this episode. These winds bring ozone and other chemical species into far-distance transport, which can contribute to the ‘base’ loadings of these species in downwind areas as shown in Fig. 3.

Fig. 1 Surface ozone mixing ratios (in ppbv) predicted by the model at 2 pm (EDT) on July 17, 1999.

Fig. 2 Simulated surface wind vectors (in Knots) at 2 pm (EDT) on July 17, 1999.

Fig. 3 Time series of ozone mixing ratios from the ground-based observations (black curve) and simulations with emissions of 1985 (red curve) and 1990 (blue curve) for selected sites in Canada and the United States in July 1999 (EDT). The vertical lines correspond to midnight every second day of the simulation.

Fig. 3 shows the time series of ozone mixing ratios predicted with the 1985 and 1990 emission inventories and observed for eight sites among all those in the model domain. Four of them are in Canada. They are London, Toronto, Peterborough and Montreal, extending from the south-west to the north-east, representing the airshed upwind, within and downwind of major metropolitan areas. The other four sites are in the United States: New York City, Ohio Warren, Pennsylvania Centre and Pennsylvania Lawrence. These sites were selected due to their spread-out geographical extent and their high ozone levels. The comparison between the predicted and the measured mixing ratios shows that the qualitative agreements in all cases are quite good. During afternoons, ozone reaches its maximum concentration because of high photochemical production resulting from the emissions of NO_x and volatile organic compounds (VOCs) in urban air. The titration process of ozone by NO_x results in the decrease of ozone concentration within the areas of high NO_x emissions. This is illustrated by the lower ozone concentrations in the urban regions, as shown in Fig. 1. During nights, high O₃ is seldom observed because of low ozone production, net deposition, and the titration of ozone in the urban plume. The diurnal variations are well captured by the model, which suggests the ozone production and removal processes in the boundary layer are well characterized by the model.

By comparing two model predictions associated with two emission inventories, we can find that ozone levels predicted by using the emission inventory of 1985 (red curves) are in slightly better agreement with observations than those of the 1990 emission inventory (blue curve) at Canadian monitoring sites. While at monitoring sites in the United States, the 1990 emission inventory predicts better results than the 1985 emission. The old emissions lead to higher ozone predictions at high ozone levels. It should be noted that different emission processing methods are used in Canada and the United States. Although results with the two inventories underestimate ozone at peak levels, the general predicted geographical distributions and patterns for ozone agree well.

As mentioned above, mesoscale long-range transport can play an important role in determining the level of surface ozone.^5,6 Such transport effects are clearly indicated at the four Canadian sites, as shown in Fig. 3. During nights, high ozone is seldom observed because of few ozone productions, net deposition, and the titration of ozone in urban plume. The only contribution to high ozone at night is from transport. Ozone levels don’t drop to low levels at night in Canadian sites compared with the four sites in the United States, especially for the period between July 16 and July 18. It is also evident that ozone transport effects at Montreal and Peterborough are greater than at Toronto and London. The north-eastward wind, as shown in Fig. 2, brings pollutants from high concentration areas in the American Midwest and Ohio valley, and therefore, increases the ‘base’ level in downwind areas during this period.

Fig. 4 shows the correlation coefficients between the simulated and the observed ozone on an-hourly basis for the eight sites of Fig. 3. The coefficients were computed for four days ‘before’, ‘during’ and ‘after’ the episode, as well as for the total period, at each location. The overall correlations are quite good for the eight sites, with total period correlations between 0.85 and 0.92 for the United States and between 0.79 and 0.86 for Canada (Fig. 4). In the United States, the best correlation seems to be during the peak in the episode when the maximum ozone concentration occurs. The correlations for the sites in the United States are in general better than those of the four Canadian locations. There is also a trend of decreasing correlation as time progresses ‘before’, ‘during’ and ‘after’ the episode in Canada. Fig. 5 shows the correlation coefficient for each site in the model domain over the total period. Most of the sites have satisfied values larger than 0.70, and many even larger than 0.8, which suggests the model has dynamically and chemically captured the main features of the oxidant chemistry. It is also interesting to note that the further a site is from the source region, for example in Quebec at the north-eastern region of the model domain, the worse the correlation. This is probably due to the meteorological uncertainties in the long-range transport of trace species as indicated by Biswas and Rao⁵ and Hogrefe et al.⁶ The wind and other meteorological fields are usually not predicted as well in the surface layer as in upper layers of the model.

Fig. 4 Correlation coefficients calculated during four-day periods ‘before’, ‘during’ and ‘after’ the ozone episode, and for the total period in July 1999 for the eight selected sites of Fig. 3.

Fig. 5 Correlation coefficient for O₃ from July 10 to 21, 1999.

In addition to the correlation coefficient, the index of agreement is also introduced complementarily to evaluate the model performance.³⁴ It is expressed as:

where N is the number of observations and P and O indicate the predicted and observed concentrations, and Ō is the mean value of the observations at each site. The index of agreement has a theoretical range of 0–1, the latter score suggesting perfect agreement. Our results show that 80% of the monitoring sites have the index of agreement larger than 0.8 for ozone, moreover, 11% of the monitoring sites have the index of agreement larger than 0.9.

To statistically evaluate model performance, two other parameters are calculated: the normalized gross error G and the normalized bias B. They are defined as

Fig. 6 shows the geographical distribution of G in the entire domain for ozone evaluated with a cut-off of 60 ppbv. Observation–prediction pairs are often excluded if the observations are below a cut-off, typically chosen to be 60 ppbv for ozone²⁰ based on the US EPA-recommended criteria for evaluating ozone performance.³³ All sites but one have values smaller than 35%. When the cut-off is reduced to 30 ppbv, which is much lower than usually selected, there are only 9 sites with G larger than 35%, but still smaller than 40%.

Fig. 6 Mean normalized gross error G for O₃ from July 10 to 21, 1999.

Fig. 7 gives the normalized bias B for ozone over the model domain. Most sites have B within ±15%. Russell and Dennis²⁰ presented overall bias and gross error from seven air quality models. The biases ranged from −24% to +7.4%, and the gross errors ranged from 15% to 36%. The California Air Resources Board³¹ and the US EPA have established acceptable ranges for G and B for model performance evaluation, which are <35% for G and within ±15% for B. Over the entire MC2AQ domain, the values of 18.5% for G and −10.9% for B are within the acceptable ranges. It is clear that the MC2AQ performs quite well for surface ozone, especially in the regions of southern Ontario, Ohio, Pennsylvania and the New York City area.

Fig. 7 Mean normalized bias B for O₃ from July 10 to 21, 1999.

While the primary focus of this paper is to investigate the model’s performance with respect to ozone, it is interesting to look at the predictions of other gaseous concentrations. Fig. 8 presents a summary of the correlation coefficients for NO, NO₂, NO_x, and SO₂ over the model domain. NO_x was evaluated separately from NO and NO₂, because certain monitoring locations only record total NO_x, and because there are possible uncertainties in the ratio of NO/NO₂ in the emissions from each source. The correlation extends up to 0.8 for NO, 0.65 for NO₂, 0.7 for NO_x and 0.6 for SO₂. The best correlations occur in the Greater Toronto Area, northern Virginia and Pennsylvania for NO₂, in eastern Pennsylvania and central New York for SO₂, in Toronto and New York City for NO.

Fig. 8 Histograms of correlation coefficients for NO, NO₂, NO_x, SO₂, wind speed and wind angle agreement within 30 degrees between the observations and the model results from July 10 to 21, 1999.

While the correlations for these gases seem not as good compared to those for ozone, they nevertheless are within the ranges obtained by others. For example, Biswas et al.³² obtained a correlation of about 0.2 with a maximum of 0.4 for NO_x in their unfiltered data in the RAMS/UAM-V model evaluation. Russell and Dennis²⁰ showed a systematic under-prediction of NO_x by the models and indicated that there seems to be a tendency of the models to convert NO_x too rapidly to NO_y. The correlation value of this work is quantitatively similar to their results at most of the sites, except for the over-predictions of NO_x for the sites around New York City. Compared with the 1985 emission inventory, predictions of NO_x have been improved around the Great Toronto Area, and the SO₂ predictions have been improved around Montreal.

The reason that modelled NO_x deviates from observations is that measurements are made near the sources where the model emission schemes will never be able to reproduce small-scale fluctuations observed. Diurnal temporal variations of NO_x and SO₂ emissions can also account for the poor predication of these species.³² Sub-grid scale variability in emissions will have a major impact on the comparison between the model and observations. When only observations between 2 and 6 pm are considered, the correlation coefficients for NO_x improve, with 18 stations having values between 0.5 and 0.7, compared to only 8 when the entire day is analyzed. This is a reflection of the fact that in the afternoon, the lower atmosphere is well mixed, which minimises the uncertainties in sub-grid scale variability. Further work to improve the predication for these species is necessary in order to model accurately the secondary formation of nitrate and sulfate with the regional air quality model. If the precursor emissions are not adequately represented in the model, or the precursor chemistry is not correctly simulated, the predictive capability of the model for PM will be reduced significantly.

The meteorological prediction of the model has been extensively studied by the original MC2 group.²⁷ The purpose here is to provide a level of confidence in the predictions of wind speed and direction as they relate to the gas phase concentrations predicted by the model. Thus Fig. 8 presents a histogram of the correlation coefficient of the wind speed over the entire domain. The correlation extends to 0.75, with a maximum between 0.2 and 0.6. Fig. 8 also presents a histogram of the fraction of hours at each site where the difference between the observed wind direction and that predicted by the model is less than 30 degrees. The poorest correlations for winds occur around the Great Lakes. This is likely due to poorly resolved lake breeze circulation in that region, which would require much higher horizontal spatial resolution (e.g. Plummer²⁴). The study of the effects of the lake breeze circulation at high resolution with MC2AQ is underway.

Conclusions

An air quality modelling system MC2AQ with a coupled online gas chemistry module and tracer transport algorithms, integrated with an hourly-based emission inventory, was evaluated. The simulation results were statistically and graphically compared against observations for surface O₃, NO, NO₂, NO_x and SO₂ as well as wind speed and direction, in eastern Canada and the United States. The time series comparisons for ozone on the surface demonstrate that the simulated results provide a good representation of the diurnal variation and peak concentration. The mean normalized gross error and the normalized bias for the surface ozone are within 35% and ±15%, respectively, which are acceptable for air quality models. The study indicates that the model can capture many main features of ozone incidents. The estimations for other species are not as good as ozone, but are in good agreement with other air quality models evaluated.²⁰ Although further improvements of precursor predictions are necessary, the present analysis shows that MC2AQ can be applied to the studies of surface ozone and other pollutants. The model can also serve as a basis for further developments to include the formation, microphysical evolution and transport of PM so that it can be used for regulatory purposes in Canada.

Acknowledgements

The authors would like to thank Dr. Virginia Ambrose of US EPA’s AIRS for providing the valuable monitoring datasets in the north-east US domain, Dr. Tom Dann of the Analysis and Air Quality Division of Environment Canada for the air quality datasets in Canada, and Dr. Bryan Smith of Ontario Climate Centre of Environment Canada for the meteorology datasets. The authors also acknowledge the Centre for Research in the Earth and Space Technology (CRESTech), the Natural Science and Engineering Research Council (NSERC), and the Canadian Petroleum Products Institute (CPPI) for their generous funding of this research.

References

1. National Research Council, Rethinking the ozone problem in urban and regional air pollution, National Academy Press, Washington DC, 1991.
2. P. J. Crutzen, in Tropospheric Ozone, ed. I. S. A. Isaksen, Reidel Publishing Co., 1988.
3. Fed. Regist., 1996, 61 (241), Dec. 13, 65763.
4. J. H. Seinfeld and S. N. Pandis, Atmospheric Chemistry and Physics: From Air Pollution to Climate Change, J. Wiley and Sons, New York, 1998.
5. J. Biswas and S. T. Rao, J. Appl. Meteorol., 2001, 40, 117.
6. C. Hogrefe, S. T. Rao, P. Kasibhatla, G. Kallos, C. T. Tremback, W. Hao, D. Olerud, A. Xiu, J. McHenry and K. Alapaty, Atmos. Environ., 2001, 35, 4159.
7. M. Z. Jacobson, R. Lu, R. P. Turco and O. B. Toon, Atmos. Environ., 1996, 30, 1939.
8. M. Z. Jacobson, J. Geophys. Res., 2001, 106, 5385.
9. R. Lu, R. P. Turco and M. Z. Jacobson, J. Geophys. Res., 1997, 102, 6063.
10. R. Lu, R. P. Turco and M. Z. Jacobson, J. Geophys. Res., 1997, 102, 6081.
11. Z. Meng, J. Geophys. Res., 1998, 103, 3419.
12. N. Kumar, M. T. Odman and A. G. Russell, J. Geophys. Res., 1994, 99, 5385.
13. Systems Applications International (SAI), Users Guide to the variable Grid Urban Airshed Model (UAM-V), Systems Applications International, San Rafael, CA, 1995p. 131.
14. ENVIRON, User's Guide to the Comprehensive Air 'Quality Model with Extensions (CAMx), ENVIRON International Corporation, Novato, CA, 1997.
15. J. S. Chang, S. Jin, Y. Li, M. Beauharnois, C. H. Lu, H. C. Huang, S. Tanrikulu and J. DaMassa, The SARMAP Air Quality Model, California Environmental Protection Agency, Sacramento, CA, Final Report, 1997.
16. M. T. Odman and C. L. Ingram, Multiscale Air Quality Simulation Platform (MAQSIP): source code documentation and validation, Research Triangle Park, 1996, Technical Report, ENV-96TR002-v1.0. MCNC.
17. N. Wheeler, Proceedings of the Air and Waste Management Association’s 1998 Annual Meeting, Paper 98-A739, A&WMA, Pittsburgh, PA.
18. P. Kasibhatla and W. L. Chameides, Geophys. Res. Lett., 2000, 27, 1415.
19. D. W. Byun and J. K. S. Ching, Science Algorithms of EPA models-3 Community Multiscale Air Quality (CMAQ) Modeling System, 1999, EPA/600/R-99/030.
20. A. Russell and R. Dennis, Atmos. Environ., 2000, 34, 2283.
21. A. Venkatram, P. Karamchanchandani and P.K. Misra, Atmos. Environ., 1988, 24, 737.
22. J. A. Pudykiewicz, A. Kallaur and P. K. Smolarkiewicz, Tellus Ser. B, 1997, 49, 231.
23. R. L. Dennis, D. W. Byun, J. H. Novak, K. J. Galluppi, C. J. Coats and M. A. Vouk, Atmos. Environ., 1996, 30, 1925.
24. D. A. Plummer, PhD Thesis, York University, 1999.
25. D. A. Plummer, J. C. McConnell, L. Neary, J. Kaminski, R. Benoit, J. Drummond, J. Narayan, V. Young and D. R. Hastie, Atmos. Environ., 2001, 35, 6453.
26. J. C. McConnell, D. A. Plummer, J. W. Kaminski and L. Neary, The Ozone Monitoring Network in SW Ontario: Assessment using MC2AQ, Report to the Ontario Ministry of Environment, Sept. 28, 1999.
27. R. Benoit, M. Desgagne, P. Pellerin, S. Pellerin, Y. Chartier and S. Desjardins, Mon. Weather Rev., 1997, 125, 2382.
28. M. Tanguay, A. Robert and R. Laprise, Mon. Weather Rev., 1990, 118, 1970.
29. T. Scholtz, A. Taylor and A. Ivanoff, Preparation of 1990 North American emissions inventory modelling files for AES regional air quality models, Report for Environment Canada, Jan. 21, 2000.
30. CRESTech, Application of the Models-3 Framework for a Canadian City Using an Advanced Aerosol Module and Ambient Monitoring Techniques, Report to CRESTech, Aug. 22, 2001.
31. CARB (California Air Resources Board), Technical guidance document: Photochemical modeling, Sacramento, 1992.
32. J. Biswas, C. Hogrefe, S. T. Rao, W. Hao and G. Sistla, Atmos. Environ., 2001, 35, 6129.
33. US Environmental Protection Agency, Guideline for Regulatory Applications of the Urban Airshed Model, Research Triangle Park, NC 27711, 1991, EPA-450/4-91-013.
34. C. Willmott, Bull. Am. Meteorol. Soc., 1982, 63, 1309.

---