• Baklanov, A., and Coauthors, 2007: Integrated systems for forecasting urban meteorology, air pollution and population exposure. Atmos. Chem. Phys., 7 , 855874.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bessagnet, B., Hodzic A. , Blanchard O. , Lattuati M. , Le Bihan O. , and Marfaing H. , 2005: Origin of particulate matter pollution episodes in wintertime over the Paris Basin. Atmos. Environ., 39 , 61596174.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Brücher, W., Kessler C. , Kerschgens M. , and Ebel A. , 2000: Simulation of traffic-induced air pollution on regional to local scales. Atmos. Environ., 34 , 46754681.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ching, J., Herwehe J. , and Swall J. , 2006: On joint deterministic grid modeling and sub-grid variability conceptual framework for model evaluation. Atmos. Environ., 40 , 49354945.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1993: A nonhydrostatic version of the Penn State/NCAR mesoscale model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev., 121 , 14931513.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebel, A., Memmesheimer M. , and Jacobs J. , 2007: Chemical perturbations in the planetary boundary layer and their relevance for chemistry transport modelling. Bound.-Layer Meteor., 125 , 265278.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Galmarini, S., Vinuesa J-F. , and Martilli A. , 2007: Modeling the impact of sub-grid scale emission variability on upper-air concentration. Atmos. Chem. Phys., 8 , 141158.

    • Search Google Scholar
    • Export Citation
  • Ginoux, P., Chin M. , Tegen I. , Prospero J. M. , Holben B. , Dubovik O. , and Lin S-J. , 2001: Sources and distributions of dust aerosols simulated with the GOCART model. J. Geophys. Res., 106 , 2025520273.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hanna, S., Chang J. , and Fernau M. , 1998: Monte Carlo estimates of uncertainties in predictions by a photochemical grid model (UAM-IV) due to uncertainties in input variables. Atmos. Environ., 32 , 36193628.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hanna, S., Lu Z. , Frey H. , Wheeler N. , Vukovitch J. , Arunachalam S. , Fernau M. , and Hansen D. , 2001: Uncertainties in predicted ozone concentrations due to input uncertainties for the UAM-V photochemical grid model applied to the July 1995 OTAG domain. Atmos. Environ., 35 , 891903.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hauglustaine, D. A., Hourdin F. , Jourdain L. , Filiberti M-A. , Walters S. , Lamarque J-F. , and Holland E. A. , 2004: Interactive chemistry in the Laboratoire de Météorologie Dynamique general circulation model: Description and background tropospheric chemistry evaluation. J. Geophys. Res., 109 .D04314, doi:10.1029/2003JD003957.

    • Search Google Scholar
    • Export Citation
  • Hourdin, F., and Armengaud A. , 1999: The use of finite-volume methods for atmospheric advection of trace species. Part I: Test of various formulations in a general circulation model. Mon. Wea. Rev., 127 , 822837.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kousa, A., Kukkonen J. , Karppinen A. , Aarnio P. , and Koskentalo T. , 2002: A model for evaluating the population exposure to ambient air pollution in an urban area. Atmos. Environ., 36 , 21092119.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kruize, H., Hänninen O. , Breugelmans O. , Lebret E. , and Jantunen M. , 2003: Description and demonstration of the EXPOLIS simulation model: Two examples of modeling population exposure to particulate matter. J. Exposure Anal. Environ. Epidemiol., 13 , 8799.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mass, C., Ovens D. , Westrick K. , and Colle B. , 2002: Does increasing horizontal resolution produce more skillful forecasts? Bull. Amer. Meteor. Soc., 83 , 407430.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Masson, V., 2000: A physically-based scheme for the urban energy budget in atmospheric models. Bound.-Layer Meteor., 94 , 357397.

  • Menut, L., 2003: Adjoint modelling for atmospheric pollution processes sensitivity at regional scale during the ESQUIF IOP2. J. Geophys. Res., 108 .8562, doi:10.1029/2002JD002549.

    • Search Google Scholar
    • Export Citation
  • Mestayer, P. G., and Coauthors, 2005: The urban boundary-layer field campaign in marseille (ubl/cluescompte): Set-up and first results. Bound.-Layer Meteor., 114 , 315365.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Miao, J-F., Chen D. , and Wyser K. , 2006: Modelling subgrid scale dry deposition velocity of O3 over the Swedish west coast with MM5-PX model. Atmos. Environ., 40 , 415429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Michou, M., Laville P. , Serca D. , Fotiadi A. , Bouchou P. , and Peuch V-H. , 2005: Measured and modeled dry deposition velocities over the ESCOMPTE area. Atmos. Res., 74 , 89116.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Moeng, C., and Wyngaard J. , 1989: Evaluation of turbulent transport and dissipation closures in second-order modeling. J. Atmos. Sci., 46 , 23112330.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ott, W., 1985: Total human exposure: An emerging science focuses on humans as receptors of environmental pollution. Environ. Sci. Technol., 19 , 880886.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rotach, M., 1993a: Turbulence close to a rough urban surface. Part I: Reynolds stress. Bound.-Layer Meteor., 65 , 128.

  • Rotach, M., 1993b: Turbulence close to a rough urban surface. Part II: Variances and gradients. Bound.-Layer Meteor., 66 , 7592.

  • Schmidt, H., Derognat C. , Vautard R. , and Beekmann M. , 2001: A comparison of simulated and observed ozone mixing ratios for the summer of 1998 in Western Europe. Atmos. Environ., 35 , 62776297.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sillman, S., Vautard R. , Menut L. , and Kley D. , 2003: O3-NOX-VOC sensitivity and NOX-VOC indicators in Paris: Results from models and Atompheric Pollution over the Paris Area (ESQUIF) measurements. J. Geophys. Res., 108 .8563, doi:10.1029/2002JD001561.

    • Search Google Scholar
    • Export Citation
  • Stockwell, W., 1995: Effects of turbulence on gas-phase atmospheric chemistry: Calculation of the relationship between time scales for diffusion and chemical reaction. Meteor. Atmos. Phys., 57 , 159171.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tetzlaff, G., Dlugi R. , Fridrich K. , Gross G. , Hinneburg D. , Pahl U. , Zegler M. , and Molders N. , 2002: On modeling dry deposition of long-lived and chemically reactive species over heterogeneous terrain. J. Atmos. Chem., 42 , 123155.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117 , 17791800.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Van Leer, B., 1979: Towards the ultimate conservative difference scheme. V. A second order sequel to Godunov’s method. J. Comput. Phys., 32 , 101136.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • van Loon, R., and Coauthors, 2007: Evaluation of long-term ozone simulations from seven regional air quality models and their ensemble. Atmos. Environ., 41 , 20832097.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vardoulakis, S., Fisher B. E. A. , Pericleous K. , and Gonzalez-Flesca N. , 2003: Modelling air quality in street canyons: A review. Atmos. Environ., 37 , 155182.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vardoulakis, S., Gonzalez-Flesca N. , Fisher B. , and Pericleous K. , 2005: Spatial variability of air pollution in the vicinity of a permanent monitoring station in central Paris. Atmos. Environ., 39 , 27252736.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vautard, R., Beekmann M. , Roux J. , and Gombert D. , 2001: Validation of a hybrid forecasting system for the ozone concentrations over the Paris area. Atmos. Environ., 35 , 24492461.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vautard, R., and Coauthors, 2003a: Paris emission inventory diagnostics from ESQUIF airborne measurements and a chemistry transport model. J. Geophys. Res., 108 .8564, doi:10.1029/2002JD002797.

    • Search Google Scholar
    • Export Citation
  • Vautard, R., and Coauthors, 2003b: A synthesis of the air pollution over the Paris region (ESQUIF) field campaign. J. Geophys. Res., 108 .8558, doi:10.1029/2003JD003380.

    • Search Google Scholar
    • Export Citation
  • Vautard, R., Honore C. , Beekman M. , and Rouil L. , 2005: Simulation of ozone during the August 2003 heat wave and emission control scenarios. Atmos. Environ., 39 , 29572967.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vautard, R., and Coauthors, 2007: Evaluation and intercomparison of ozone and PM10 simulations by several chemistry transport models over four European cities within the CityDelta project. Atmos. Environ., 41 , 173188.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vinuesa, J-F., and de Arellano J. V-G. , 2005: Introducing effective reaction rates to account for inefficient mixing in the convective boundary layer. Atmos. Environ., 39 , 445461.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wesely, M., 1989: Parameterization of surface resistances to gaseous dry deposition in regional-scale numerical models. Atmos. Environ., 23 , 12931304.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • WHO, 2000: Air Quality Guidelines for Europe. 2nd ed. WHO European Series, Vol. 91, WHO Regional Publications, 273 pp.

  • View in gallery

    IdF domain. The contours represent the surface emission flux of total NOx (NO + NO2) at 0800 1 Aug 2003 (10−10 gm−2 s−1). AIRPARIF monitoring stations with red stand for urban, orange for periurban, and black for rural stations. The exact location of each station is marked with a blue triangle.

  • View in gallery

    Surface O3 concentrations (ppb) modeled with the CHIMERE model for (top) 1, (middle) 2, and (bottom) 3 Aug 2003 at 1700 UTC for region IdF. The model configurations are (left) FINE, (middle) AVER, and (right) COARSE. FINE runs use meteorological input computed by MM5 at 5-km2 horizontal resolution and emission fluxes are provided by the AIRPARIF inventory (1-km2 resolution). COARSE runs use meteorological input computed by MM5 at 0.5° horizontal resolution and input emission fluxes are diagnosed by the EMEP inventory also at a 0.5° horizontal resolution.

  • View in gallery

    Differences (%) between modeled and measured daily ozone maxima, at different types of stations (7 urban, 8 periurban, and 7 rural sites) and for the three run configurations: FINE (circle), AVER (triangle), and COARSE (square).

  • View in gallery

    Time series of surface O3 concentrations (ppb) from 30 Jul to 3 Aug 2003 at three different sites: (top) downtown Paris, (middle) Les Ulis, and (bottom) Tremblay. The black solid, red, and black dashed lines correspond to the simulations FINE, COARSE, and AVER, respectively. The temporal profiles of measured ozone at the corresponding sites are represented with circles. The error bars represent the standard deviation in AVER configuration at the hour of the daily ozone concentration maximum.

  • View in gallery

    CHIMERE model performance in forecasting the daily peak of surface ozone concentrations. Modeled daily maximum ozone concentrations at the different configurations (FINE: circle, AVER: triangle, COARSE: square) for the cases where the alert threshold value (90 ppb) is exceeded according to measurements (star). The error bars represent the standard deviation in AVER configuration.

  • View in gallery

    Ozone surface concentrations (ppb) modeled with the CHIMERE model at the fine-resolution grid (6 km2) over the IdF domain on (top) 1, (middle) 2, and (bottom) 3 Aug 2003. Input data for emissions and meteorology are used at different resolutions: EMIc and EMIf for emissions and MM5c and MM5f for meteorology as described in text. (a) Both emission fluxes and meteorological data are used at coarse resolution (EMIc + MM5c). (b) Emissions input is used at coarse resolution and meteorological data are used at fine resolution (EMIc + MM5f). (c) Emissions are used at fine resolution and meteorological data at coarse resolution (EMIf + MM5c). (d) Both emission fluxes and meteorological data are used at fine resolution (EMIf + MM5f).

  • View in gallery

    Differences (%) between modeled and measured daily ozone maxima at different types of stations (7 urban, 8 periurban, and 7 rural sites) for four different run configurations: COARSE (square), EMIc + MM5f (emissions input are used at coarse resolution and meteorological data are used at fine resolution: triangle down), EMIf + MM5c (emissions input are used at fine resolution and meteorological data are used at coarse resolution: triangle up), and FINE (circle).

  • View in gallery

    Difference (%) between modeled and measured surface O3 daily maxima as a function of the horizontal resolution of emission input fluxes. Model error is averaged over all stations where observed O3 concentration crossed the alarm threshold of 90 ppb.

  • View in gallery

    Time series of modeled O3 dry deposition velocity over different subgrid surfaces (colored lines) and ponderated with the land use fractions, mean dry deposition velocity (dashed black line) for the grid cells corresponding to the sites (top left) Paris 1, (top right) Les Ulis, and (bottom left) Tremblay. (bottom right) The map shows the maximum spatial, subgrid variability of O3 concentration modeled with CHIMERE model over the IdF domain for 3 Aug 2003 with FINE resolution.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 271 143 3
PDF Downloads 177 97 4

Does an Increase in Air Quality Models’ Resolution Bring Surface Ozone Concentrations Closer to Reality?

View More View Less
  • 1 Laboratoire de Météorologie Dynamique, Ecole Polytechnique, Palaiseau, France
Full access

Abstract

A persistent challenge for small-scale air quality modeling is the assessment of health impact and population exposure studies. Despite progress in computation and in the quality of model input (i.e., high-resolution information on land use and emission patterns), the uncertainty associated with input parameters cannot be eliminated. The aim of this paper is to study different sources of uncertainty that affect model results as the resolution increases. Mesoscale chemistry transport simulations at different resolutions are used and modeled 03 concentrations are compared with surface measurements. The case study consists of CHIMERE model simulations over the city of Paris. It is shown that the principal source of noise in model results is the resolution of the input emission fluxes. The O3 concentrations modeled with simulations forced by several horizontal resolutions of input emission data (from Δx = 48 km to Δx = 6 km) indicate that model results do not improve monotonously with resolution, but that after a certain point discrepancies become larger. Based on this result and as an alternative to the deterministic downscaling that resolves explicitly the finer scale (beyond the 1-km range), the authors propose a subgrid-scale approach that uses a statistical description of spatial scales finer than model resolution. As an example, the subgrid variability of modeled O3 concentration has been quantified, when modeled dry deposition processes occur over subgrid surfaces (land use fractions). The implementation of this modified calculation gives access to subgrid fluxes and subgrid surface concentrations instead of the mean values provided by the commonly used model calculation.

Corresponding author address: Myrto Valari, Laboratoire de Météorologie Dynamique, Ecole Polytechnique, 91128 Palaiseau, France. Email: myrto.valari@lmd.polytechnique.fr

Abstract

A persistent challenge for small-scale air quality modeling is the assessment of health impact and population exposure studies. Despite progress in computation and in the quality of model input (i.e., high-resolution information on land use and emission patterns), the uncertainty associated with input parameters cannot be eliminated. The aim of this paper is to study different sources of uncertainty that affect model results as the resolution increases. Mesoscale chemistry transport simulations at different resolutions are used and modeled 03 concentrations are compared with surface measurements. The case study consists of CHIMERE model simulations over the city of Paris. It is shown that the principal source of noise in model results is the resolution of the input emission fluxes. The O3 concentrations modeled with simulations forced by several horizontal resolutions of input emission data (from Δx = 48 km to Δx = 6 km) indicate that model results do not improve monotonously with resolution, but that after a certain point discrepancies become larger. Based on this result and as an alternative to the deterministic downscaling that resolves explicitly the finer scale (beyond the 1-km range), the authors propose a subgrid-scale approach that uses a statistical description of spatial scales finer than model resolution. As an example, the subgrid variability of modeled O3 concentration has been quantified, when modeled dry deposition processes occur over subgrid surfaces (land use fractions). The implementation of this modified calculation gives access to subgrid fluxes and subgrid surface concentrations instead of the mean values provided by the commonly used model calculation.

Corresponding author address: Myrto Valari, Laboratoire de Météorologie Dynamique, Ecole Polytechnique, 91128 Palaiseau, France. Email: myrto.valari@lmd.polytechnique.fr

1. Introduction

A current trend in small-scale air quality modeling is to quantify the health impact of exposure to atmospheric pollution (WHO 2000). The associated health risk can be evaluated if the output of air quality models is linked to information on citizen activities (Ott 1985; Kousa et al. 2002; Kruize et al. 2003; Baklanov et al. 2007). Such a numerical approach requires high temporal and spatial resolution in air quality models to better represent the heterogeneity of the urban environment. The variability of pollutants inside cities is mainly due to (i) the temporal evolution and spatial distribution of emission sources and (ii) the dispersion of pollutants inside street canyons.

Several modeling tools exist to describe the evolution of meteorological properties at a finescale. Large eddy simulation (LES) models resolve finescale turbulent structures, such as small eddies, by adapting higher-moments closure to the formulation of atmospheric turbulence (Moeng and Wyngaard 1989). Street canyon models (Vardoulakis et al. 2003) calculate explicitly the flow within a designated city structure [building height to street width ratio must be provided (Rotach 1993a, b)]. The horizontal resolution of such models ranges from a few meters up to a hundred meters.

In some cases, meteorological models have integrated more complex parameterizations (Masson 2000; Mestayer et al. 2005) that assess the impact of street canyons on the mean flow by statistical representations. In this approach an a priori knowledge of city features, linked to the modeled mean meteorological flow, can provide an estimate of subgrid-scale effects. On the other hand, mesoscale chemistry transport models (CTMs) commonly use a K-diffusion theory to account for the turbulent mixing inside model cells. Such parameterizations lead to enhanced mixing at high model resolution (especially close to high emissions sources) and model results can turn out to be significantly biased as discussed in Brücher et al. (2000).

When chemical processes are taken into account, the challenge of small-scale modeling becomes rather different. Surface emissions are characterized by heterogeneous spatial patterns and highly fluctuating temporal profiles associated with transport. The inherent variability of surface emissions has a nonlinear impact on the chemical transformation of pollutants (Galmarini et al. 2007) and consequently, any increase in the resolution of meteorology and/or emissions might lead to more uncertainty of modeled concentrations. Mass et al. (2002) discuss similar effects on mesoscale meteorological modeling, but at least to our knowledge, this uncertainty has never been quantified for chemistry transport models. Nevertheless, the uncertainty can become significant for secondary pollutants such as ozone, because all errors related to meteorology, emissions, chemistry, and transport are accumulated into the final modeled concentration (Hanna et al. 1998, 2001). In addition, the validation of high-resolution CTM is not a trivial task since validation data, mainly surface measurements, are not always representative of model grid cells (Vardoulakis et al. 2005).

Another important statement concerns the sensitivity of model results to the chemical boundary conditions. The simulation domain needs to be large enough around local sources so that species background concentrations would be represented correctly (Menut 2003). This condition is not satisfied in LES and high-resolution mesoscale modeling (i.e., horizontal resolution finer than 1 km).

Thus, in the present study we begin by addressing the impact of the horizontal resolution of a CTM on modeled ozone concentrations. The case study consists of simulations of the Paris urban plume during a photo-oxidant episode modeled with the CHIMERE model (section 2). The chosen event represents a typical summer smog event allowing for a certain generalization of the concluding remarks. Different horizontal resolutions are used, while the vertical resolution is identical in all cases. We compare model results at two different resolutions to surface measurements (section 3). Surprisingly, ozone concentrations modeled with the coarse resolution showed, globally, the greatest agreement with observations. Previous studies (van Loon et al. 2007; Vautard et al. 2007) have shown that the CHIMERE model discrepancies are mainly due to uncertainties related to input emissions and meteorological data, and not to systematic bias. Based on this observation, we argue that even if the finer resolution gives large errors when it is directly compared with measurements, its results can provide a reliable statistical representation of the subgrid variability inside the coarse-resolution cells. Model results at a high resolution were averaged over the coarse mesh grid cells and the standard deviation was used to represent O3 spatial variability. We show that this statistical representation of model results at the finescale is more realistic (smaller bias) than the deterministic mean output concentrations.

It is possible to identify the different sources of noise affecting model calculation as the resolution increases and, eventually, to quantify their impact. We assess the question by using different resolutions of meteorological input and emissions data. We study the model’s response as the resolution increases (section 4), and show that surface ozone was most affected by changes in the emission inventory resolution and that this impact is highly nonlinear. We gradually increased the resolution of input emission fluxes by averaging the values over larger areas around the “standard” fine-resolution cells. Modeled ozone concentrations are compared to measurements with the intention of studying how the resolution of primary species emission affects model results with regard to a secondary species (O3). In practice, simulations forced by 10 different horizontal resolutions of input emission fluxes, starting at 48 km × 54 km and down to 6 km × 6 km, showed that model results do not improve linearly with the resolution. On the contrary, model discrepancies become larger when the resolution becomes too high, suggesting that there exists an optimal point where the equilibrium between model resolution and input errors is inversed.

Based on this result, we suggest that CTM simulations at high spatial resolution can lead to unrealistic pollutants concentrations. An alternative way to represent the unresolved scale is to describe statistically the heterogeneity at subgrid space and model the corresponding variability of species concentrations. We present a first study toward this approach that focuses on the ozone dry deposition process (section 5). Ozone dry deposition variability over model grid cells is used to generate concentration fluctuations during each model time step. Mean dry deposition velocity of gaseous species is usually calculated in CTMs as the weighted average of the different velocities parameterized over different subgrid fractions of the grid cell (fractional land use). For the same case study, we change the model to address the subgrid variability of the dry deposition process to the final modeled ozone concentrations. Thus, instead of averaging the different subgrid dry deposition velocities to a single mean flux, we allowed each different component to affect the mean model concentration in an independent way. At the end of each model time step, the resulting concentrations are weighted with land use data and the subgrid variability of ozone concentration per time step is represented by the standard deviation around the mean value.

2. Case study

a. Synoptic situation

A persistent anticyclone over western Europe during the first half of August 2003 resulted in exceptionally high temperatures over the continent. The stagnation of a polluted air mass over urban areas favored photochemical activity and ozone production inside urban plumes downwind of intense sources of O3 precursors. A complete analysis of this event is presented in Vautard et al. (2005).

b. Model setup

The episode is modeled with the CTM CHIMERE (see online at euler.lmd.polytechnique.fr/chimere). The domain covers Ile de France (IdF) (48°–49.5°N, 1.25°–3.6°E), an area of about 180 km × 180 km centered around Paris (Fig. 1). The highly urbanized center of the domain represents an intense source of polluted air mass principally related to traffic. Urbanization gradually decreases with the increasing distance from the city center. The isocontours of Fig. 1 correspond to the emitted mass of total NOx (NO + NO2) at 0800 UTC 1 August 2003. The model is used at two different resolutions: (i) the fine resolution with Δx = Δy = 6 km (FINE), and (ii) the coarse resolution with Δλ = Δϕ = 0.5° (COARSE). The simulation results are compared over the IdF domain, which consists of 900 FINE and 20 COARSE grid cells (the actual domain of COARSE covers western Europe from 10°W to 22°E and from 35° to 57°N). Vertically, the two model configurations are identical: eight vertical layers with the top of the first layer at 995 hPa (≈50 m above surface) and up to 500 hPa, which, as discussed in Vautard et al. (2005), is high enough to include the boundary layer during the period of the simulation. The modeled boundary layer height, during the 5 simulated days, reached a maximum height of 1850 m. Hourly meteorological data are interpolated every 10 min at the FINE configuration whereas the integration over the chemical mechanism uses a 1.5-min time step. COARSE runs use a single integration time step of 10 min for both the interpolation of meteorological inputs and chemical reactions. Meteorological parameters are calculated with the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) (Dudhia 1993) at a vertical resolution of 32 vertical layers (from the surface to 200 hPa) and a diagnostic preprocessing is applied to provide turbulence-related parameters for the CHIMERE model [friction velocity, boundary layer height, deep convective fluxes, and turbulent diffusivity (Kz)]. This processing of meteorological parameters is described and validated in previous studies (Schmidt et al. 2001; Vautard et al. 2001). All other parameters are described in Bessagnet et al. (2005). In the standard configuration of the model the surface emission fluxes used in FINE simulation are taken from the inventory developed by the AIRPARIF local association [discussed and validated in Vautard et al. (2003a)]. Emissions for COARSE (a configuration designed to apply to continental-scale simulations) are provided by the European Monitoring and Evaluation Programme (EMEP) inventory (www.emep.int). In section 4, where we no longer discuss model scores versus measurements but focus on pure modeling features and model intercomparisons, we scaled the AIRPARIF input emission fluxes to obtain consistency between the total emitted mass diagnosed by the different inventories. Thus, we conserved spatially the emitted mass for each species and each time step over the surface of COARSE cells.

Chemical boundary conditions for the coarse-resolution simulation (continental domain) are driven by GOCART model monthly climatologies (Ginoux et al. 2001) for aerosol species, while boundary conditions for gas-phase species are driven by the LMDz-INCA (Interaction with Chemistry and Aerosol model) global chemical weather forecast system. The LMDz-INCA model refers to the coupling between the Laboratoire de Meteorologie Dynamique general circulation model (Van Leer 1979; Tiedtke 1989; Hourdin and Armengaud 1999) and INCA (Hauglustaine et al. 2004). The fine-resolution simulation was nested inside the domain of the coarse-resolution simulation and thus it used COARSE output concentrations as boundary conditions.

Simulation results are discussed for 1, 2, and 3 August 2003, when the model ozone plume is clearly distinguishable from background values. Model concentrations at different resolutions are compared with each other and against measurements of the AIRPARIF air quality network [as described in Vautard et al. (2003b)]. Surface stations are divided in three groups characterized as urban, periurban, and rural according to their location in the domain and the emission activity that best describes the vicinity of each station (and thus their spatial representativeness).

3. Direct comparison of simulations with measurements

In this section, we compare the different standard model configurations FINE and COARSE directly to each other, as well as against measurements. In addition, we estimate FINE mean ozone concentration spatially averaged over COARSE. This configuration (hereafter called AVER) does not represent a simulation by itself, but it allows us to intercompare model results and estimate the differences due to model resolution.

Surface ozone concentration maps illustrate the behavior of ozone plumes depending on model configuration. We compare model performance with regard to [O3] at each configuration and evaluate the gain associated with the use of a finer resolution (i.e., additional features of ozone chemistry and transport being captured). A zoom to selected sites is then done and we study the temporal evolution of model results at different resolutions. We also discuss how each model configuration triggers air pollution alarms.

a. Surface maps of ozone concentration

As the grid size decreases, the model’s spatial and temporal resolution approaches the scales of the physical and chemical processes driving O3 formation and thus more detailed spatial patterns become evident (Fig. 2). For the 3 days, FINE predicts more O3 at the center of the plume while the total area covered by high [O3] values is larger in COARSE. Differences in modeled [O3] can go up to 40 ppb, for both the excess in the center of the FINE plume and its deficit at background levels.

These differences are due not only to spatial averaging over concentrations as shown in the maps by AVER, but they also reflect the impact of emissions resolution (e.g., NOx) on the modeled O3. This relates specifically to the sources due to vehicular traffic (localized but intense). By using COARSE surface emission fluxes, the respective roles of “ozone titration” (close to the sources) and “ozone precursor” (far from the sources) are smoothed and may disappear.

With FINE, O3 depletion is pronounced inside Paris where NO is emitted at high rates by traffic, leading to an underestimation of O3 concentration. For 1 August, for example, surface ozone concentrations ranged from 80 to 115 ppb at 1700 UTC in the cells representing Paris. On the other hand, COARSE results at the same area and time give a homogenous surface ozone concentration of 90 ppb.

The O3 is continually produced downwind with FINE where the most intense plumes are formed far from the sources. FINE is able to predict surface ozone peaks up to 30 ppb more than COARSE inside such plumes. This shows the inability of COARSE to correctly represent O3 production away from the city (more than 40 km for such stagnant conditions).

The integration time step had to be decreased with the increase in model spatial resolution, so that the Courant–Friedrichs–Lewy (CFL) condition (u × Δtx < c) would be satisfied. The use of a small time step may also be an explanation for the enhanced O3 titration modeled with FINE because it lets the model better represent the fast reaction between O3 and NO, which is not as well captured by the COARSE time step. On 2 and 3 August 2003, the impact of resolution is also reflected on the location of high O3 concentrations. FINE gives relatively low, background [O3] levels in areas where COARSE crosses the pollution alert threshold.

Finally, when the modeled O3 concentrations of FINE are averaged over the coarse grid (AVER configuration), a great difference between AVER and COARSE is found (middle columns of Fig. 2). This gives an idea of the nonlinear effects related to O3 chemistry and transport.

b. Ozone daily maxima

Measured daily ozone maxima are compared to corresponding model simulations for all stations available in the Paris area (22 stations: 7 urban, 8 suburban, and 7 rural, shown in Fig. 1). Surface stations are split into three categories: urban, periurban, and rural (following the European classification); for each site type, we report the difference between modeled and measured ozone concentrations for the three model configurations, FINE, COARSE, and AVER. The results corresponding to each day of the simulation are superimposed on the same column (Fig. 3).

These scores give the spread of each model configuration around the observed values. Logically, the largest spread corresponds to COARSE: because of its low resolution, the model fails to capture finescale features driving air quality in the vicinity of surface stations (especially rural). Differences with measurements reach the same extreme values for FINE and COARSE configurations (i.e., higher than 30%). At the urban stations, the discrepancies of FINE can be greater than those of COARSE. The large spread of FINE highlights the model sensitivity to accumulated errors: a higher resolution ensures a greater variability of model results, but at the same time acts as a source of uncertainty often leading to larger errors when we compare to point sites. This argument is reinforced by the fact that AVER gives the smallest spread.

A concluding remark would be that even if the spread of COARSE around measurements is larger, the mean value of the discrepancies is well centered around zero. On the contrary, FINE has a net tendency to underestimate measured values. This aspect reflects model deficiency in correctly predicting O3 titration close to high NOx emissions. The part of the FINE error that is “corrected” with the use of AVER configuration has been attributed to input data errors (especially emissions) being smoothed by the averaging. Nevertheless, the underestimation of ozone concentration remains at the AVER configuration, suggesting different sources of model error. Ozone underestimation might also be due to enhanced mixing of emissions leading to a more efficient ozone depletion by NO. At high model resolution and in particular at areas with sharp horizontal emission gradients, the instantaneous mixing of emitted species at model cells leads to large discrepancies (Stockwell 1995; Vinuesa and de Arellano 2005; Ebel et al. 2007). Even though FINE results represent realistic features of the variability of ozone concentration, an averaging approach can be a more accurate tool, from a forecasting point of view.

c. Temporal evolution of ozone concentration

The impact of the horizontal resolution on model results can be better understood when we focus on individual stations and compare the temporal evolution of the model’s behavior at different configurations with surface observations (Fig. 4). To illustrate this impact, we selected three stations: Paris 01, Les Ulis, and Tremblay (see also Fig. 1 for locations). Paris 01 is at the center of Paris; Les Ulis and Tremblay are small cities 20 km southwest and northeast of Paris, respectively. During 30 and 31 July, ozone concentrations remained lower than 60–70 ppb for all stations; this is reasonably estimated by the model for FINE, COARSE, and AVER configurations. During the first 3 days of August 2003, the “official alert threshold” for ozone concentration (90 ppb) was exceeded for a large number of stations in the Paris area. Recorded ozone concentrations crossed the 90-ppb threshold at different stations, on one, two, or all three days of the simulation (e.g., Paris station), depending on the wind direction.

In downtown Paris, FINE always estimated values lower than the measured ones. Using AVER instead of FINE and adding the modeled variability (standard deviation), we see, however, that for 2 days out of 3 the measured crossing of the alert threshold value is included in the model’s result. COARSE estimated higher concentrations than FINE and is closest to the measurements. It should also be noted that AVER is the only configuration able to simulate realistic nighttime ozone values (i.e., nonzero values).

Measured values southwest of Paris, in Les Ulis, on 1 August showed moderate ozone values (below the alert threshold). All model configurations showed an equal underestimation and gave concentration peaks of 60 ppb when 85 ppb was observed: this is associated with a very low subgrid variability (shown with AVER), suggesting that the model underestimates the real ozone values for a large spatial area around the station and explains the fact that FINE and COARSE results are very close to one another.

For the same day, the Tremblay station is found in the ozone plume. The ozone peak is well retrieved, especially with FINE.

For the 3 days when measured ozone exceeded the alert threshold (1, 2, and 3 August 2003) and for the three sites of Fig. 4, the general remarks are mainly that (i) COARSE or AVER ± ΔAVER are able to model properly ozone’s diurnal cycle in downtown Paris (ΔAVER = σ, with 2σ standing for the standard deviation); (ii) at periurban stations, depending on whether the station is located inside the plume or not, FINE can give very realistic results (e.g., Tremblay) while the low resolution of COARSE is unable to distinguish the urban center from the remote plume (this led to false alerts such as in Tremblay for 3 August); and (iii) the use of FINE averaged over COARSE (AVER configuration) allows one to diagnose correctly most of the ozone peaks for the 3 days. This latter configuration benefits from a high resolution of emissions inventory and meteorological fields, and by averaging the results to a lower resolution, the impact of the potential errors is smoothed.

d. Alarm threshold triggers

Using the same kind of information as in the previous sections, we focus on daily peaks that cross the alert threshold (i.e., cases where recorded values exceed 90 ppb). Up to now it was shown that the performance of COARSE is better when model results are directly compared to measurements. It is, however, certain that a finer-scale modeling can better reproduce concentration variability that is, if not precise, at least realistic. It is thus important to evaluate the supplementary information delivered by FINE compared to COARSE. The standard deviation around AVER is considered as the ozone variability that FINE can model at COARSE subgrid scale. Figure 5 presents the cases where the measured daily peak exceeded 90 ppb and the corresponding modeled daily peak is reported for FINE, COARSE, and AVER.

  • COARSE: 28 peaks modeled out of 34. This result, however, should be tempered by the fact that there is more than one station inside the same COARSE grid cell and thus if modeled ozone concentration reaches the threshold a single time, the model will be considered to have successfully triggered the alarm for all the included stations.

  • FINE: 12 peaks modeled out of 34. Depending on the day (and the meteorology), the results show no clear tendency to underestimate or overestimate the observed ozone concentrations. It should be noted that observed high values (e.g., above 120 ppb) were reproduced by FINE, although at the wrong place, but were totally missed by COARSE (i.e., COARSE did not exceed 110 ppb in any case).

  • AVER ± ΔAVER: 20 peaks modeled out of 34. If AVER ± ΔAVER is considered, instead of directly comparing the results of FINE with measurements, it becomes clear that the standard deviation of FINE is a realistic representation of the variability of modeled ozone.

In general, COARSE is able to model peaks when measurements remain close to the threshold. The low resolution “sacrifices” the high-resolution variability of model output ozone concentrations but on the other hand, by using input already averaged over large areas, the accumulation of model errors is limited. When FINE results are spatially averaged (AVER) the aforementioned errors are also averaged but the variability of the concentration can be represented statistically using the standard deviation. When we use AVER ± ΔAVER, much better forecast scores are obtained. The question is whether we are able to find the critical point where the uncertainty inherent to input resolution overcomes the benefits of model resolution increase.

In the following we assess this question by using different resolutions of input datasets as forcings for the standard FINE resolution simulation and we study the impact on modeled O3 concentrations.

4. The impact of resolution of emissions inventory and meteorological input on modeled surface ozone

Up to this point, we have illustrated that the resolution increase introduces high uncertainty to the model and thus the information gained on the variability of ozone concentration by refinement of the scale becomes of little help because different sources of noise interfere. Furthermore, it was shown that if modeled variability is used in a statistical way (ΔAVER) rather than a deterministic one (direct comparison with measurements), the resulting image becomes much more realistic. The problem is that the variability that one can retrieve from model outputs cannot be easily linked to the variability of the input parameters and all other sources of noise, since their interaction is highly nonlinear. Ching et al. (2006) carried out statistical analyses of the subgrid distribution of model results in fine cells inside coarser cells. Apart from a certain tendency of inhibited O3 production in cases of high NO concentrations, this study concluded that parameters are too variable in space and time across the modeling domain to derive any conclusion of a prognostic character.

a. Qualitative analysis

We set up a modeling experiment that allowed us to make a qualitative evaluation of the parameters that most affect model results as the resolution increases. The model is used with a constant spatial resolution of Δx = 6 km and we change the resolutions of the entire dataset of meteorological inputs and emissions from COARSE to finescale.

Different model configurations are defined:

  • Surface emissions: COARSE used the EMEP dataset projected to the CHIMERE coarse grid (0.5° horizontal resolution). This configuration is denoted EMIc. FINE used high-resolution emissions taken from the AIRPARIF emissions inventory (1-km horizontal resolution). Emission fluxes are summed over the CHIMERE fine grid (6-km horizontal resolution). This simulation is denoted EMIf. So as not to take into account the difference in the total emitted mass of the EMEP and the AIRPARIF inventories, the AIRPARIF emissions were adjusted to match exactly those of COARSE. For each hour of the simulation, we forced the sum of all FINE-resolution emission fluxes included in the same COARSE cell to equal the corresponding COARSE emission flux. With this approach, we ensure that the high-resolution information on the spatial distribution of the emitting sources is not lost, but the EMLc total mass and hourly profiles are conserved. Therefore, a more objective comparison of model results is carried out and the impact of the emissions resolution can be clearly highlighted.

  • Meteorology: for COARSE it is calculated with the mesoscale model MM5 at 0.5° horizontal resolution. This configuration is denoted MM5c. For FINE it is calculated with MM5 model at a 5-km horizontal resolution. This configuration is denoted with MM5f. A lot of parameters are thus perturbed at the same time (e.g., wind speed, temperature, friction velocity, etc.).

The impact of the resolution of meteorological data on modeled [O3] can be seen by comparing Figs. 6a,b. Figure 6a shows model results when all input is given at COARSE resolution (EMIc + MM5c) while Fig. 6b gives the results when the meteorological parameters are given at finescale (EMIc + MM5f). The impact on surface ozone concentrations is relatively low. This stems from the fact that the two setups of the model are not so different: the ozone plumes are very dispersed in both cases [compared to the references configuration (Fig. 6d) EMIc + MM5f]. Apart from a certain low-ozone variability found at the center of the plumes, the two model configurations share the same horizontal patterns. The aspect of enhanced O3 depletion pointed out when FINE results are compared with measurements is still present in this model configuration.

On the contrary, when the model is forced by high-resolution emissions (EMIc + MM5c), a great difference is observed on model results (Figs. 6a,c). The ozone plume is much less dispersed horizontally with EMIf. In EMIc configuration the strong horizontal gradients of NOx emissions are not represented in the model because of resolution insufficiency. This leads to (i) surface O3 being less titrated over urban areas, especially on 2 and 3 August (longer residence over the city) and (ii) more O3 being formed much farther from the center compared to EMIf. Both aspects can give an explanation for the better results of COARSE when the different model resolutions are compared with measurements.

It should be noted that the difference between Fig. 6d and the first column of Fig. 2 stems from the fact that AIRPARIF emission fluxes were modified to match the EMEP totals.

Figure 7, where the model response on variable input data resolution is directly compared to measurements, gives a more quantitative picture of simulation results. When only meteorological inputs are used in a fine resolution (EMIc + MM5f), model error remains close to the COARSE simulation. On the contrary, when only emissions are used in their fine configuration (EMIf + MM5c), model results become much more similar to the FINE simulation. At the same time it is clear that the model is more sensitive to changes in the resolution of emissions than in meteorological input.

For the present case study, the resolution of meteorological data affected model results to a much lesser extent compared to emissions. Even if wind direction was variable during the days of the simulation, the spatial variability of wind fields was very low. In general, over the Paris area, emissions exhibit very strong horizontal gradients: the landscape changes from a highly urbanized city center to rural areas over a very short distance (20–30 km). On the contrary, meteorological fields may be relatively homogeneous, as was indeed the case during the specific episode.

b. Quantification of the impact of emissions resolution

Having shown the great impact of the resolution of emissions on modeled surface ozone concentrations, and at the same time that the reliability of model results does not necessarily improve with resolution, one should logically wonder whether a critical point can be found beyond which further resolving the emission patterns becomes pointless. The nonlinear response of model error to horizontal resolution stems from the fact that at a certain scale, input errors overcome model resolution. In this section we attempt to define this point for the particular model and the presented case study. The high-resolution emission set denoted above as EMIf (modified AIRPARIF input fluxes) was averaged spatially over gradually increasing surfaces, thus decreasing its resolution. Ten different resolutions of the emission dataset are used, starting from the initial AIRPARIF set used for FINE simulations (6 km × 6 km) and up to 48 km × 48 km. Simulations are run over the FINE CHIMERE grid so that model spatial and temporal resolutions are identical for all cases. Model differences with surface measurements are evaluated for each simulation when observed ozone daily peak crosses the 90-ppb threshold. The result is given in Fig. 8 for O3 concentration, where it is shown that globally, model results improve almost linearly with resolution from 48 km × 48 km to 12 km × 12 km and then discrepancies begin to increase.

Observed O3 concentration crosses the alarm whether at urban or at periurban sites (the same cases shown in Fig. 5) where spatial O3 variability is large. Error increase with resolution at such stations is due to the fact that measurements represent the actual O3 concentration only at a small spatial scale (around 1 km). Errors in the emission inventory and wind direction are also accumulated, leading to larger model discrepancies at higher resolution.

An additional explanation is related to the design of the emission inventory and the way mesoscale models parameterize the mixing of the emitted species. Mesoscale models suppose an instantaneous dilution of emissions to the model cell. Since inventory validation is based on these models, an inherent correction accounting for the false dilution effect characterizes emission input data. This assumption makes emission inventories resolution-dependent and is a possible explanation for the worse results found in the FINE resolution simulation.

5. A statistical representation of model subgrid-scale variability: An example based on ozone dry deposition

It was shown that there exists an upper limit in the increase of model resolution in mesoscale simulations of photo-oxidant pollution events. The exact value of this critical point depends also on model design (physical parameterizations, chemical mechanism, quality of input data, etc.) and on the studied event. The order of magnitude of the critical resolution resulting from the present case study can be generalized since (i) a typical case of summer smog was simulated, (ii) a previously validated model was used, and (iii) one cannot imagine unbiased input emission data at high resolution. In this context, we propose an alternative way to gain information on the unresolved scale by using a statistical representation of the subgrid space.

In the present study we showed the fundamental role of emission resolution on modeled [O3]; thus it would be logical to go ahead with this argument with an example based on a statistical representation of subgrid-scale emissions into the model. It is, however, a complicated task and a large quantity of data is needed to proceed to a correct statistical representation of emissions. For this reason we present here an application of the suggested methodology on the dry deposition process. The choice is justified by the fact that dry deposition is a major sink for O3 (Michou et al. 2005) and also because common model dry deposition parameterizations are already based on statistical representations of the subgrid scale.

Ozone deposition flux depends strongly (if not primarily) on the type of underlying surface (Tetzlaff et al. 2002). Models often use statistical information providing the percentage of the grid-cell surface occupied by different surface types. These land-use fractions account for the surface heterogeneity at subgrid scale and provide a more realistic mean dry deposition flux. In the present study, ozone dry deposition velocity is calculated over each subgrid surface according to the parameterization of Wesely (1989). Nine types of surfaces are used in the model. At each model time step, different subgrid velocities are calculated (colored lines in Fig. 9). Ozone deposes faster over “grassland” or “forest” surface types (0.8–1 cm s−1) and slower over crops (0.3 cm s−1). In addition, ozone dry deposition velocity may show a marked diurnal cycle over a certain type of surface (e.g., grassland type) or retain a constant behavior throughout the day (e.g., urban surface or over water). Similarly, the impact of meteorological variations (from one day to another or between different grid cells) on modeled ozone dry deposition is a function of the surface type, as is also shown in Fig. 9. However, the different dry deposition velocities are instantly ponderated by the land use fractions to a single mean value. For example, the grid cell where the Paris site is situated has zero spatial variability since 100% of the corresponding cell surface is characterized as urban and the mean ozone dry deposition velocity coincides with the one modeled for the urban type of surface. The largest variability for the three sites depicted in Fig. 9 is logically modeled over Tremblay since land use fractions represent the most marked surface heterogeneity. The mean value of the normalized standard deviation of ozone dry deposition velocity at the corresponding cell is 39%.

At the moment of averaging of the subgrid velocities, the statistical information (land use fractions) is lost and ozone concentration at each time step is finally affected only by the mean dry deposition flux. Nevertheless, the dependence of dry deposition on land use is too high for the mean deposition velocity to sufficiently represent the process, as already discussed in Miao et al. (2006). The present study goes a step further than Miao et al. (2006), because it transfers the variability of O3 dry deposition velocity to the finally modeled ozone concentration (bottom right panel in Fig. 9).

In practice, at each model time step and for all model cells, instead of averaging dry deposition velocities in the single mean deposition flux, we kept the different components until the end of model calculation (during each time step). The mean model concentration of all species was affected by a different dry deposition flux, each one corresponding to a subgrid land use fraction. In this way the dry deposition variability due to the heterogeneity of the grid is used to generate concentration fluctuations. The result of this calculation at the end of the time step is a number of concentrations (equal to the number of land uses encountered in the cell) corresponding to deposition over hypothetical cells entirely covered with each different land use. Weighting this set of concentrations with the land use fractions we obtain a mean concentration to be used at the next time step. By calculating the standard deviation of these concentrations we gain access to the corresponding part of ozone subgrid variability.

The described method was applied over the Ile de France domain for FINE (Δx = 6 km) model resolution. Land use data were taken from the EMEP database (see section 2). The maximum variability modeled on 3 August 2003 is 0.28 ppb per time step, around a mean concentration of 90 ppb. This corresponds to a variability of 0.3% (left panel of Fig. 9). The low values of O3 variability can be explained to a large extent by the fact that the provided land use data were not sufficiently detailed to allow higher variability to emerge. A sensitivity analysis (not shown) was carried out over grid cells occupied by a great number of random combinations of land uses. This study sorted out particular combinations of surface types that lead to high O3 concentration variability when coexisting in the same grid cell. The maximum variability was found for combinations of water (inland or oceanic) with grassland, and for urban surface with grassland. Even though such combinations are realistic for the domain of the study (e.g., parks, rivers, streets, and buildings), they were not represented in the land use data used by the model.

6. Conclusions

We used a validated mesoscale chemistry transport model at two different resolutions, and showed that even if the refinement of the scale evidently gives a more realistic representation of ozone evolution inside an urban plume, the accumulation of errors related to emission and meteorological input makes the final result more biased when it is directly compared with measurements. We used an alternative approach, where model results at fine resolution account for a statistical ozone variability at the unresolved scale of a coarser resolution. This representation brought model results closer to reality.

Modeled surface ozone was found to be especially sensitive to the resolution at which input emissions are given. The increase of emission resolution was found to improve model results only up to a certain point, beyond which the induced noise became large. In the present study, we attributed this model sensitivity to the double role of NOx emissions that can alternatively produce or titrate O3. Previous studies over the same area also showed that ozone formation inside the Paris urban plume is often very close to the transition between two different photochemical regimes (Sillman et al. 2003). Modeled O3 production may be more sensitive to either NOx or volatile organic compounds (VOC) emission depending on the chemical regime. A net trend of underestimation of [O3] in the finer-resolution simulation is probably due to the enhanced mixing of NOx emission, leading to “artificially” efficient O3 titration.

Parameterizations used in mesoscale models impose a limit to the downscaling, beyond which model results become too sensitive to input errors and uncertainties of input data overcome model resolution. It was shown that an optimum model resolution can be defined even though the actual value quantified in the present study (6 km < Δx < 12 km) does not provide a universal value. Since air quality modeling in an urban environment needs to assess health impact issues and thus reach spatial and temporal scales inaccessible to explicit mesoscale calculations, we explored the alternative of statistically representing the subgrid scale.

We applied this approach to the ozone dry deposition process, and we modeled ozone concentration subgrid variability by taking into account a division of the grid-cell space to subsurfaces. The quantified variability was very low, probably because of a poor representation of the subsurfaces by the land use data, but the interest of the study lies in the fact that it is possible to use statistical information to represent concentration variability at a subgrid space without explicitly resolving the finer scale.

REFERENCES

  • Baklanov, A., and Coauthors, 2007: Integrated systems for forecasting urban meteorology, air pollution and population exposure. Atmos. Chem. Phys., 7 , 855874.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bessagnet, B., Hodzic A. , Blanchard O. , Lattuati M. , Le Bihan O. , and Marfaing H. , 2005: Origin of particulate matter pollution episodes in wintertime over the Paris Basin. Atmos. Environ., 39 , 61596174.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Brücher, W., Kessler C. , Kerschgens M. , and Ebel A. , 2000: Simulation of traffic-induced air pollution on regional to local scales. Atmos. Environ., 34 , 46754681.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ching, J., Herwehe J. , and Swall J. , 2006: On joint deterministic grid modeling and sub-grid variability conceptual framework for model evaluation. Atmos. Environ., 40 , 49354945.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1993: A nonhydrostatic version of the Penn State/NCAR mesoscale model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev., 121 , 14931513.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebel, A., Memmesheimer M. , and Jacobs J. , 2007: Chemical perturbations in the planetary boundary layer and their relevance for chemistry transport modelling. Bound.-Layer Meteor., 125 , 265278.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Galmarini, S., Vinuesa J-F. , and Martilli A. , 2007: Modeling the impact of sub-grid scale emission variability on upper-air concentration. Atmos. Chem. Phys., 8 , 141158.

    • Search Google Scholar
    • Export Citation
  • Ginoux, P., Chin M. , Tegen I. , Prospero J. M. , Holben B. , Dubovik O. , and Lin S-J. , 2001: Sources and distributions of dust aerosols simulated with the GOCART model. J. Geophys. Res., 106 , 2025520273.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hanna, S., Chang J. , and Fernau M. , 1998: Monte Carlo estimates of uncertainties in predictions by a photochemical grid model (UAM-IV) due to uncertainties in input variables. Atmos. Environ., 32 , 36193628.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hanna, S., Lu Z. , Frey H. , Wheeler N. , Vukovitch J. , Arunachalam S. , Fernau M. , and Hansen D. , 2001: Uncertainties in predicted ozone concentrations due to input uncertainties for the UAM-V photochemical grid model applied to the July 1995 OTAG domain. Atmos. Environ., 35 , 891903.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hauglustaine, D. A., Hourdin F. , Jourdain L. , Filiberti M-A. , Walters S. , Lamarque J-F. , and Holland E. A. , 2004: Interactive chemistry in the Laboratoire de Météorologie Dynamique general circulation model: Description and background tropospheric chemistry evaluation. J. Geophys. Res., 109 .D04314, doi:10.1029/2003JD003957.

    • Search Google Scholar
    • Export Citation
  • Hourdin, F., and Armengaud A. , 1999: The use of finite-volume methods for atmospheric advection of trace species. Part I: Test of various formulations in a general circulation model. Mon. Wea. Rev., 127 , 822837.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kousa, A., Kukkonen J. , Karppinen A. , Aarnio P. , and Koskentalo T. , 2002: A model for evaluating the population exposure to ambient air pollution in an urban area. Atmos. Environ., 36 , 21092119.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kruize, H., Hänninen O. , Breugelmans O. , Lebret E. , and Jantunen M. , 2003: Description and demonstration of the EXPOLIS simulation model: Two examples of modeling population exposure to particulate matter. J. Exposure Anal. Environ. Epidemiol., 13 , 8799.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mass, C., Ovens D. , Westrick K. , and Colle B. , 2002: Does increasing horizontal resolution produce more skillful forecasts? Bull. Amer. Meteor. Soc., 83 , 407430.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Masson, V., 2000: A physically-based scheme for the urban energy budget in atmospheric models. Bound.-Layer Meteor., 94 , 357397.

  • Menut, L., 2003: Adjoint modelling for atmospheric pollution processes sensitivity at regional scale during the ESQUIF IOP2. J. Geophys. Res., 108 .8562, doi:10.1029/2002JD002549.

    • Search Google Scholar
    • Export Citation
  • Mestayer, P. G., and Coauthors, 2005: The urban boundary-layer field campaign in marseille (ubl/cluescompte): Set-up and first results. Bound.-Layer Meteor., 114 , 315365.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Miao, J-F., Chen D. , and Wyser K. , 2006: Modelling subgrid scale dry deposition velocity of O3 over the Swedish west coast with MM5-PX model. Atmos. Environ., 40 , 415429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Michou, M., Laville P. , Serca D. , Fotiadi A. , Bouchou P. , and Peuch V-H. , 2005: Measured and modeled dry deposition velocities over the ESCOMPTE area. Atmos. Res., 74 , 89116.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Moeng, C., and Wyngaard J. , 1989: Evaluation of turbulent transport and dissipation closures in second-order modeling. J. Atmos. Sci., 46 , 23112330.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ott, W., 1985: Total human exposure: An emerging science focuses on humans as receptors of environmental pollution. Environ. Sci. Technol., 19 , 880886.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Rotach, M., 1993a: Turbulence close to a rough urban surface. Part I: Reynolds stress. Bound.-Layer Meteor., 65 , 128.

  • Rotach, M., 1993b: Turbulence close to a rough urban surface. Part II: Variances and gradients. Bound.-Layer Meteor., 66 , 7592.

  • Schmidt, H., Derognat C. , Vautard R. , and Beekmann M. , 2001: A comparison of simulated and observed ozone mixing ratios for the summer of 1998 in Western Europe. Atmos. Environ., 35 , 62776297.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Sillman, S., Vautard R. , Menut L. , and Kley D. , 2003: O3-NOX-VOC sensitivity and NOX-VOC indicators in Paris: Results from models and Atompheric Pollution over the Paris Area (ESQUIF) measurements. J. Geophys. Res., 108 .8563, doi:10.1029/2002JD001561.

    • Search Google Scholar
    • Export Citation
  • Stockwell, W., 1995: Effects of turbulence on gas-phase atmospheric chemistry: Calculation of the relationship between time scales for diffusion and chemical reaction. Meteor. Atmos. Phys., 57 , 159171.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tetzlaff, G., Dlugi R. , Fridrich K. , Gross G. , Hinneburg D. , Pahl U. , Zegler M. , and Molders N. , 2002: On modeling dry deposition of long-lived and chemically reactive species over heterogeneous terrain. J. Atmos. Chem., 42 , 123155.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117 , 17791800.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Van Leer, B., 1979: Towards the ultimate conservative difference scheme. V. A second order sequel to Godunov’s method. J. Comput. Phys., 32 , 101136.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • van Loon, R., and Coauthors, 2007: Evaluation of long-term ozone simulations from seven regional air quality models and their ensemble. Atmos. Environ., 41 , 20832097.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vardoulakis, S., Fisher B. E. A. , Pericleous K. , and Gonzalez-Flesca N. , 2003: Modelling air quality in street canyons: A review. Atmos. Environ., 37 , 155182.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vardoulakis, S., Gonzalez-Flesca N. , Fisher B. , and Pericleous K. , 2005: Spatial variability of air pollution in the vicinity of a permanent monitoring station in central Paris. Atmos. Environ., 39 , 27252736.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vautard, R., Beekmann M. , Roux J. , and Gombert D. , 2001: Validation of a hybrid forecasting system for the ozone concentrations over the Paris area. Atmos. Environ., 35 , 24492461.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vautard, R., and Coauthors, 2003a: Paris emission inventory diagnostics from ESQUIF airborne measurements and a chemistry transport model. J. Geophys. Res., 108 .8564, doi:10.1029/2002JD002797.

    • Search Google Scholar
    • Export Citation
  • Vautard, R., and Coauthors, 2003b: A synthesis of the air pollution over the Paris region (ESQUIF) field campaign. J. Geophys. Res., 108 .8558, doi:10.1029/2003JD003380.

    • Search Google Scholar
    • Export Citation
  • Vautard, R., Honore C. , Beekman M. , and Rouil L. , 2005: Simulation of ozone during the August 2003 heat wave and emission control scenarios. Atmos. Environ., 39 , 29572967.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vautard, R., and Coauthors, 2007: Evaluation and intercomparison of ozone and PM10 simulations by several chemistry transport models over four European cities within the CityDelta project. Atmos. Environ., 41 , 173188.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vinuesa, J-F., and de Arellano J. V-G. , 2005: Introducing effective reaction rates to account for inefficient mixing in the convective boundary layer. Atmos. Environ., 39 , 445461.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Wesely, M., 1989: Parameterization of surface resistances to gaseous dry deposition in regional-scale numerical models. Atmos. Environ., 23 , 12931304.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • WHO, 2000: Air Quality Guidelines for Europe. 2nd ed. WHO European Series, Vol. 91, WHO Regional Publications, 273 pp.

Fig. 1.
Fig. 1.

IdF domain. The contours represent the surface emission flux of total NOx (NO + NO2) at 0800 1 Aug 2003 (10−10 gm−2 s−1). AIRPARIF monitoring stations with red stand for urban, orange for periurban, and black for rural stations. The exact location of each station is marked with a blue triangle.

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 2.
Fig. 2.

Surface O3 concentrations (ppb) modeled with the CHIMERE model for (top) 1, (middle) 2, and (bottom) 3 Aug 2003 at 1700 UTC for region IdF. The model configurations are (left) FINE, (middle) AVER, and (right) COARSE. FINE runs use meteorological input computed by MM5 at 5-km2 horizontal resolution and emission fluxes are provided by the AIRPARIF inventory (1-km2 resolution). COARSE runs use meteorological input computed by MM5 at 0.5° horizontal resolution and input emission fluxes are diagnosed by the EMEP inventory also at a 0.5° horizontal resolution.

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 3.
Fig. 3.

Differences (%) between modeled and measured daily ozone maxima, at different types of stations (7 urban, 8 periurban, and 7 rural sites) and for the three run configurations: FINE (circle), AVER (triangle), and COARSE (square).

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 4.
Fig. 4.

Time series of surface O3 concentrations (ppb) from 30 Jul to 3 Aug 2003 at three different sites: (top) downtown Paris, (middle) Les Ulis, and (bottom) Tremblay. The black solid, red, and black dashed lines correspond to the simulations FINE, COARSE, and AVER, respectively. The temporal profiles of measured ozone at the corresponding sites are represented with circles. The error bars represent the standard deviation in AVER configuration at the hour of the daily ozone concentration maximum.

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 5.
Fig. 5.

CHIMERE model performance in forecasting the daily peak of surface ozone concentrations. Modeled daily maximum ozone concentrations at the different configurations (FINE: circle, AVER: triangle, COARSE: square) for the cases where the alert threshold value (90 ppb) is exceeded according to measurements (star). The error bars represent the standard deviation in AVER configuration.

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 6.
Fig. 6.

Ozone surface concentrations (ppb) modeled with the CHIMERE model at the fine-resolution grid (6 km2) over the IdF domain on (top) 1, (middle) 2, and (bottom) 3 Aug 2003. Input data for emissions and meteorology are used at different resolutions: EMIc and EMIf for emissions and MM5c and MM5f for meteorology as described in text. (a) Both emission fluxes and meteorological data are used at coarse resolution (EMIc + MM5c). (b) Emissions input is used at coarse resolution and meteorological data are used at fine resolution (EMIc + MM5f). (c) Emissions are used at fine resolution and meteorological data at coarse resolution (EMIf + MM5c). (d) Both emission fluxes and meteorological data are used at fine resolution (EMIf + MM5f).

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 7.
Fig. 7.

Differences (%) between modeled and measured daily ozone maxima at different types of stations (7 urban, 8 periurban, and 7 rural sites) for four different run configurations: COARSE (square), EMIc + MM5f (emissions input are used at coarse resolution and meteorological data are used at fine resolution: triangle down), EMIf + MM5c (emissions input are used at fine resolution and meteorological data are used at coarse resolution: triangle up), and FINE (circle).

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 8.
Fig. 8.

Difference (%) between modeled and measured surface O3 daily maxima as a function of the horizontal resolution of emission input fluxes. Model error is averaged over all stations where observed O3 concentration crossed the alarm threshold of 90 ppb.

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Fig. 9.
Fig. 9.

Time series of modeled O3 dry deposition velocity over different subgrid surfaces (colored lines) and ponderated with the land use fractions, mean dry deposition velocity (dashed black line) for the grid cells corresponding to the sites (top left) Paris 1, (top right) Les Ulis, and (bottom left) Tremblay. (bottom right) The map shows the maximum spatial, subgrid variability of O3 concentration modeled with CHIMERE model over the IdF domain for 3 Aug 2003 with FINE resolution.

Citation: Journal of Atmospheric and Oceanic Technology 25, 11; 10.1175/2008JTECHA1123.1

Save