Interactive comment on “ The representation of dust transport and missing urban sources as major issues for the simulation of PM episodes in a Mediterranean area ” by E . Flaounas

The main comments posted on this paper converge on the need to set our work in the context of previous studies and to highlight the innovating aspect of our results. Indeed, many works (as listed by reviewer #1) investigated the effect of Saharan dust transport over Mediterranean Europe during the last 10 years and this clearly needs to be better highlighted in our paper. This will be done. Some of these works are recent and the C4671


Introduction
Besides its importance in influencing climate (Charlson, 1992), particulate matter of anthropogenic origin is a main component of atmospheric pollution and is responsible for the lack of visibility in urban centers (Park and Kim, 2005;Vallius et al., 2005;Wåhlin et al., 2006;Lonati and Giugliano, 2006).Its impact on human health is now widely recognized; many studies indeed showed that atmospheric particles can easily enter respiratory airways, with a depth of inhalation and a deposition rate depending on their size, shape, physical and chemical properties (Monn, 2001;Moschandreas and Saksena, 2002).Epidemiological studies have demonstrated the association between the exposure to PM 10 and PM 2.5 (particulate matter with a diameter lower than 10 µm and 2.5 µm, respectively) and all-cause mortality, among which lung-cancer and cardiopulmonary mortality are listed (Scoggins, 2004).Recently, (Larrieu, 2007) confirmed that current levels of PM 10 and nitrogen dioxide in several European cities are linked to a short term increase of cardiovascular morbidity.Since 1999, the European Union has set limit values for PM 10 for the protection of human health over different averaging periods, so as to represent various exposure durations.A 24-h limit value of 50 µg/m 3 was fixed, not to be exceeded more than 35 times a year.Since 2005, the annual limit value for PM 10 is set to Published by Copernicus Publications on behalf of the European Geosciences Union.40 µg/m 3 .Van Dingenen et al. (2004) and Yttri (2005) reported that, at the beginning of the 2000s, this last standard was exceeded at all near-city, urban and kerbside places in Europe, while the annual limit value of 40 µg/m 3 was still over passed in several areas.Their studies also revealed that for hourly values, the 50 µg/m 3 threshold was extensively exceeded, especially in Italy (87 days per year), and that an upward trend in PM 10 baseline was observed since 2000 in some European countries.
In urban and suburban areas, the major sources of anthropogenic aerosols are road traffic and fossil fuel burning emissions, as well as combustion and soil-related industrial activities (Kaur et al., 2007;Yatkin and Bayram, 2007).Nevertheless, the background PM 10 levels are mainly composed of mineral crustal particles, and it is common that long-range transport of natural aerosols strongly participates in high local PM episodes.Indeed, Saharan dust outbreak events have been shown to be responsible for short-term sharp increases in PM 10 concentrations in the boundary layer over the western and eastern Mediterranean, and were associated to frequent exceedances of the EU limits for PM 10 (Viana, 2002;Rodriguez, 2004;Moreno, 2005;Ganor, 2009), with peak values that could exceed 200 µg/m 3 during 48 h for severe events (Sánchez et al., 2007).Numerous studies have described and discussed the importance of such dust import to Spain (Rodriguez et al., 2002;Lyamani et al., 2005;Salvador, 2008), and Italy (Gariazzo et al., 2007;Barkan, 2005;Mona, 2006).Several studies also focused on the import of dust to different parts of Europe (e.g. Viana, 2008), but to our knowledge no French data have been discussed in this context.Only the very recent study of Querol et al. (2009) proposes a comparison of dust impact across the Mediterranean basin, including three sites in southern France.These data support the possibility of a significant impact of African dust on this area.
Eulerian models that simulate the transport and the transformation of aerosols and gases constitute an important tool to enhance our knowledge about aerosol pollution events, and to develop efficient action plans to reduce the atmospheric levels of particulate matter (Ilacqua et al., 2007;Turpin, 2000;Thunis, 2007).However, in these models, the anthropogenic and terrestrial fractions of particulate matter need to be better described and quantified.Model intercomparisons, but also single evaluations of models against measurements, have been conducted in order to evaluate our current capabilities to perform comprehensive PM modeling (Bessagnet et al., 2004;Monteiro et al., 2007;Sartelet et al., 2007;Stern et al., 2008).The results helped to gain confidence in the simulation of the chemical constituents of particulate matter and showed that models were able to capture the general variability of PM during the simulated periods.However, they also highlighted some deficiencies in the models, which often lead to an underestimation of modeled PM concentrations relative to observations.These deficiencies have been attributed to the choice of the physical and chemical pa-rameterizations (including secondary organic aerosol (SOA) production), to meteorological input fields and the associated parameterization of vertical exchanges in the atmosphere, to the treatment of badly known sources, to long-range transport of dust particles, but also to the precision of measurements used for comparison.In addition, the studies indicate that errors in the emission data are important sources of uncertainty, that may reach 20% on a total anthropogenic annual basis for Europe, while they are known to be considerably larger for individual model grid points and short time scales (hours to days).In consequence, the exploration of particulate pollution case studies remains necessary to better identify the nature of particulate pollution events and to provide new directions to ameliorate eulerian model realism, for an improved air quality management.
The study presented here focuses on the simulation, with the CHIMERE eulerian model, of PM 10 concentrations in the south of France during two successive PM episodes that occurred in the month of June 2006.It aims at understanding the constitutive elements of this episode, at evaluating the ability of the model to restitute those elements, and at identify the elements in the model that can be enhanced for a better simulation and forecast of similar phenomena.This study was conducted in close cooperation with Atmo PACA, which is the public association in charge of air quality monitoring and modeling in the Provence-Alpes-Côte d'Azur (PACA) region.After the presentation of the site of the study and of the CHIMERE model configuration, we will analyze the ground-based observations of PM 10 during the selected pollution event and compare them with the model outputs.In a second part, sensitivity studies will help identifying the origin of the gaps between the model and the measurements, and some possibilities to enhance the representation of PM 10 in the model will be discussed.

Site and measurement data
In this study we focus on the western greater part of the PACA region.The domain is characterized by a complex topography (Fig. 1): it is covered with mountains (which don't exceed the altitude of 1000 m) forming two valleys (of Rhône and of Durance) while to the south lays the Mediterranean Sea.The largest city of this area is Marseille, which population is of about 820 000 inhabitants.Other large urban centers are Toulon, Avignon, Aix and Arles.A large industrial complex is located in Berre, 50 km to the west of Marseille.Among the numerous stations measuring pollutants over the area, PM 10 are measured using a Tapered Element Oscillating Microbalance in 5 urban areas and 1 industrial area on the simulation domain.Measurements of NO x and basic meteorological parameters such as wind strength and direction are also conducted at several of the 6 stations listed above.Two types of atmospheric circulation prevail on the region: Mistral and sea breeze conditions.The Mistral is a strong wind flowing from the north which eventually turns easterly as it propagates towards the South due to the sea-land temperature contrast.As a consequence, it mostly affects the western and central parts of the domain, with an expected beneficial effect on air quality due to strong atmospheric dilution (Salameh, 2007).Sea breeze circulation is associated with low winds blowing from the south.Figure 2 presents the wind rose plot for June and July 2006 and the associated mean PM 10 concentrations at the Marseille groundbased station.It shows evidence that the sea breeze (southwest, south-east direction) is associated with the greatest frequency and with high PM 10 concentrations, probably due to PM advection from the urbanized coast, but also due to longrange transport of particles.On the reverse, Mistral events (310 • -340 • ) show stronger wind speeds and are accompanied with a drop in particulate pollution intensity.The western area of Arles-Avignon is the most exposed to Mistral events, while the whole region is exposed to the slow advection of air masses by the sea breeze in low-dispersion conditions, as air masses are further blocked by the large mountainous volumes laying to the north-east (Drobinski, 2007).Its intense emissions, but also its southern location, make of this region the most polluted area in France in terms of ozone but also in terms of particulate matter, with 50% of the days having a non good quality of air.Thus, the PM 10 24 h-limit of 50 µg/m 3 is exceeded 43 to 116 times a year at several urban sites (especially in Marseille and Toulon), while only 35 excess are tolerated per year.

Model configuration
For this study we used the Chemistry Transport Model CHIMERE (http://www.lmd.polytechnique.fr/chimere/).The model was first run on a domain covering the entire PACA region with an horizontal resolution of 9 km and forced at its boundaries by the national and continental PREV'AIR simulations (http://www.prevair.org).
In a second step, the model was run over a nested domain with a finer horizontal resolution of 3 km along 41×51 grid points, covering all the urban agglomerations shown in Fig. 1.On the vertical, the domain is divided into 8 levels from the ground to 750 hPa.
The chemical scheme is the reduced MELCHIOR2 mechanism (Lattuati, 1997), which is composed of 44 species Fig. 3. Anthropogenic emissions of NOx from the regional Atmo PACA inventory.The numbers correspond to the measurement sites listed in Table 1.
including 19 organic species and 120 reactions.Photolysis rates are calculated under clear sky conditions as a function of height using the TUV model of Madronich (1998).Dynamical calculations were provided by the MM5 meso-scale model (http://www.mmm.ucar.edu/mm5/)running at the same horizontal resolution as CHIMERE.The CHIMERE aerosol module accounts for 8 species (primary particulate matter -including not only black carbon and organic carbon but also industrial mineral dusts -nitrate, sulfate, ammonium, biogenic and anthropogenic secondary organic aerosol -SOA -sea salts and water) and uses 6 size bins from 10 nm to 40 µm.Physical processes taken into account are the following: -Coagulation is calculated from the theory used by Gelbard (1980) -Absorption is taken into account for both inorganic and organic species.For inorganic species, equilibrium concentrations are computed with a tabulated version of the thermodynamic module ISORROPIA (http:// nenes.eas.gatech.edu/ISORROPIA).For secondary organic species, equilibrium concentrations are calculated through a temperature dependent partitioning coefficient (Pankow, 1994).
-Concerning nucleation, the parameterization of Kulmala (1998) for sulfuric acid nucleation is used.This process is affected to the smallest bin in the sectional distribution.Although the nucleation of condensable organic species has been clearly identified in many experimental studies (Kavouras, 1998), no parameterization was available for this version of the CHIMERE model.However, since the sulfuric acid nucleation process competes with absorption processes, it is expected to occur in weakly particle polluted conditions.
Up to now, a very simplified scheme for SOA formation has been implemented in the chemical module MEL-CHIOR, based on biogenic and anthropogenic aerosol yields.In the model, precursor volatile organic compounds able to form secondary aerosol species are high chain alkanes, aromatics and monoterpenes.Mass transfer between gas and aerosol phases is not only driven by the gas phase diffusion but also by the thermodynamic equilibrium through a temperature dependent partition coefficient (see http://www.lmd.polytechnique.fr/chimere/) for more details).Aqueous sulphate chemistry is considered, as well as a few heterogeneous reactions relative to nitric acid chemistry.Dry deposition for model gas species is parameterized as a downward flux from Wesely (1989).Dry deposition for aerosols also makes use of a resistance scheme.
Anthropogenic emissions are provided by the kilometrical regional inventory developed by Atmo PACA with a bottom-up approach for air quality forecast, last updated in 2006 and which considers NO x , speciated volatile organic compounds, SO 2 , CO, PM 10 and PM 2.5 from fixed industrial sources, mobile sources, and biogenic sources (AIR-MARAIX, 2005).Concerning PM emissions, the inventory (Fig. 3) reveals an important participation of the industrial sector (approximately 60% of the regional emissions) associated to punctual sources in the industrial area of Berre, at the west of Marseille, followed by road transports (approximately 30%).However, the weight of the road transport sector increases to 40% if we consider the finest particles (especially PM 2.5 ).It is necessary to note here that the inventory of PM sources is not exhaustive.In particular, the diffuse emissions of industrial fabrics, the agricultural sector, the particles stemming from the wear of tires, brakes and roads and their resuspension due to road traffic are not taken into consideration here, as in many current inventories, due to a lack of data.Biogenic emissions of isoprene and terpenes are parameterized as fluxes depending on temperature and insulation.
The evaluation of CHIMERE is processed everyday at Atmo PACA through a statistic evaluation of the model air quality forecasts against the ground-based network of chemical and meteorological measurements.The results are very satisfactory for gaseous pollutants, but show an underestimation of the observed PM 10 levels.Concerning meteorology, the ability of MM5 to restitute the local air mass circulations on this area was shown in the frame of a previous modeling and measurement program (Pirovano, 2007).The simulations were conducted for the whole June and July periods with hourly outputs.

Analysis of the episode simulation
The model was run over the whole period in the configuration described above.The simulated PM 10 concentrations were compared to observations and analyzed, so as to evaluate the ability of the model to perform comprehensive PM simulations.

PM episode restitution
In Fig. 4 (upper graph) are represented the time series of the daily mean PM 10 concentrations measured at all the PACA stations during the period of June-July 2006.The observations reveal a mean PM background of 20 to 30 µg/m 3 .This value is consistent with longer term observations, as described in the annual air quality reports of this region (At-moPACA, 2008).However, during this period, the measured concentrations twice overstep the limit of 50 µg/m 3 : from 18 to 21 June and from 25 to 30 June.The PM maxima then reach 60 to 75 µg/m 3 .These episodes of particulate pollution can be associated to synoptic influences and not to local phenomena, as the peaks are observed at all sites and prove to be correlated site by site.
We can notice from the lower graph (modeled values) that the model is able to capture the main features of the dayto-day variability of PM 10 during the whole period.The correlation between observations and model outputs varies This point does not constitute a bias of the model.However, when comparing the simulated daily mean time series with the observations, it appears that the model systematically underestimates the particulate pollution levels (i) by 10 µg/m 3 (factor of 1.5) during the non-event days and (ii) by 25 µg/m 3 (factor of 2) during the two episodes of June.

Analysis of the episode
We focused on the first episode, which peaks on 21 June.Due to its regional amplitude, we investigated the possibility of large-scale transport of aerosols to the simulated area, through a back trajectory analysis for the whole period.The NCEP Final analysis (FNL) showed that during the episode high pressures were prevailing over western Mediterranean.Rodriguez et al. (2001) showed that similar meteorological conditions (anticyclonic circulations) were responsible for PM pollution episodes over Spain, by driving air masses from north Africa to southern France.
Figure 5 presents 4-day back trajectories for the air masses arriving at the southern border of our simulation domain on 21 June, as generated by the HYSPLIT model from FNL (http://www.arl.noaa.gov/HYSPLIT.php).In parallel, we analyzed Angstrom coefficient values (Fig. 5 an approach was previously used by Lyamani et al. (2005) to demonstrate the occurrence of a similar event in southeastern Spain.
It is interesting to notice that CHIMERE represents the occurrence of the two dust events in proper time (see the temporal evolution of modeled dust at Marseille in Fig. 6), each time with a slight increase in the PM background level over the whole area.We could associate this increased level with the import of particulate matter through the southern boundary of the simulation domain.However, as large-scale CHIMERE simulations are forced at their boundaries by climatological dust concentrations, it comes out that such data are nor able to reproduce the intensity of the PM background during this period, neither to describe the variability in the composition of incoming air masses at the limits of the domain.It thus cannot provide a good representation of summer PM evolution in southern Europe.
We investigated the possibility to improve the model predictions in such situations by implementing, as boundary conditions in our run, the concentration fields of dust particles simulated by the DUST version of the CHIMERE model (http://www.lmd.polytechnique.fr/dust/)for the same period.This version of CHIMERE is dedicated to the transport of mineral particles and reproduces the emission, dry and wet deposition and transport of these particles over west Africa.Dust concentrations are represented through 20 size bins from 0.01 to 34 µm.The model is driven by MM5 meteorological fields and has been run for the whole year 2006 on a 1 • ×1 • grid covering a rectangular domain between 90 • W-90 • E and 10 • S-60 • N with 15 vertical levels from the surface to 200 hPa.More details about this version of the model can be found in Menut (2007).
Outputs are available every 12 h.We interpolated the concentration fields so as to fit our spatial and temporal resolutions, and the 20 dust concentration classes were redistributed into our 6 larger size bins to replace our initial and boundary conditions for the dust component of the particulate matter.This simulation, called DUSTBOUND, was run for from 9 June to 24 June.The results are presented in Sect. 4.

Local features
A second interesting aspect of the model results concerns the modeled hourly mean concentrations.When comparing observed and simulated time series for a period without any supposed dust event, we can see (as illustrated in Fig. 7 for the Toulon grid point, with a temporal zoom on the second week of June) that CHIMERE cannot capture the intensity of the urban PM variability that results from human activity.This bias mainly finds its expression in lower peak values along the day, making difficult the restitution of urban air quality.This graph is representative of the model results at all the urban sites.An underestimation of the local production of secondary organic aerosol in the model may be partly responsible for the low variability in the model outputs.The underestimation of SOA is a common feature to many models (Zhang, 2007).But although this point clearly needs refinement, it may not be the main reason for the smooth aspect of the model outputs as we consider source areas and we mainly miss morning peaks.
In urban areas, the hourly variability of PM 10 is mainly dependent on road traffic emissions.As NO 2 and PM urban emissions are mainly related to road traffic, we plotted (figures not shown) the observed PM 10 concentrations versus the observed NO x concentrations at all urban stations during the periods with negligible wind, in order to obtain data representative of local traffic emissions.We observed a linear dependency between the two species, with a PM to NO x ratio of about 0.4 while in the emission inventory, the dependency between PM 10 and NO x in urban cells varies between 0.1 and 0.2 only.May this result from an insufficient description of aerosol resuspension related to traffic in anthropogenic emission inventories, as discussed previously?According to recent studies, the contribution of resuspension phenomena is estimated to a level of 20% of road traffic emissions, while 20% further can be attributed to coarse fraction abrasion source emissions, the rest of emissions including com-bustion and fine fraction abrasion (Thorpe et al., 2007).In our case, the inventory PM to NO x ratio points out an underestimation of urban PM sources by at least a factor of 2, which fits with the omission of resuspension and road wear sources.
We thus investigated the improvement of the model that can follow from a re-adjustment of PM urban emissions, if supplementary traffic-related sources are taken into account.We launched an empirical sensitivity test by replacing the anthropogenic particle emissions by that of NO x adjusted by a factor of 0.4 according to the observations.The sensitivity test, called EMISS, was conducted from 9 to 24 June in order to explore both periods with and without dust import.The results are presented in the next section.

Sensitivity tests results
The CHIMERE DUST simulation results reveal that the model well predicts the transport of dust particles from northwestern Africa to the Mediterranean during the third week of June 2006.The arrival of a dust plume is represented in Fig. 8 which shows dust concentration maps calculated at 800 m asl for 14 June and 17 June at 12:00 UT.
When used as boundary conditions, these dust concentration fields strongly impact the regional simulation of PM 10 .The time series in Fig. 9 present the regional model results for the reference simulation and for the simulation DUST-BOUND, together with measurements, for the grid cells of the Aix and Marseille sites, respectively.We can clearly see that the DUSTBOUND simulation better reproduces the phenomenology of a PM event than the reference run, although its intensity remains strongly overestimated (20 to 30 µg/m3) from 17 to 19 June.This may either follow from inaccuracies in the dust plume location or from an overestimation of its intensity during the early days of the event.However, the sudden and pronounced increase in the PM 10 baseline (up to 80 µg/m3) on 16 June and its decrease on 22 June, that much better fit the observations, constitute a visible improvement of the model outputs.In this coupled configuration, the model thus proposes a more realistic PM background evolution, and provides a better understanding of particulate matter composition in the area.It shows that dust climatological values, commonly used as boundary conditions for CTMs, constitute a limitation to explain common particulate pollution events in southern Europe.
We can see that this new simulation however does not reproduce the high PM peaks observed in all urban stations during the dust event.From the observation of the large dust plumes in Fig. 8, we assumed that long-range transport of terrestrial particles may not generate such variability in the PM 10 levels, and we supposed a local origin for these peaks.The results of the simulation EMISS should thus indicate us if they may originate in local urban traffic emissions.
The time series in Fig. 10 show the model outputs for the reference simulation and for the simulation EMISS, together with observations for the urban sites of Toulon and Marseille.This figure reveals a much more realistic evolution of PM 10 concentrations in the simulation EMISS.Although  the exact intensity of the PM peaks measured at urban sites is not expected to be reproduced in the model due to the wellknown dilution of CTM emissions in the grid cells, the model now better captures the hourly features particulate pollution.This simulation produces PM 10 peak values of the right order of magnitude, and fully restitutes in time some of the main peaks observed in urban areas.
We have to keep in mind that, from the point of view of emission parameterization, this simulation is a simplistic one, and may not be consistent over the whole domain.However, these results highlight the fact that current anthropogenic PM emissions are underestimated in urban areas and can be easily improved using a refined empirical approach.Even if these emissions are not as influential as dust advection on the regional PM levels, their improvement remains necessary to explore the role of local PM emissions on urban air quality.

Conclusions
During the third week of June, an important episode of particulate pollution with PM 10 concentrations ranging from 60-100 µg/m 3 was observed in the south of France.We showed that it originated in the long-range transport of African dust, confirming that outbreak phenomena can influence particle levels in southern France even under the form of moderate episodes.This result is in agreement with the observations reported for other Mediterranean countries (Rodriguez et al., 2001(Rodriguez et al., , 2002;;Querol et al., 2009;Mitsakou, 2008).Although the CHIMERE model was able to reproduce the nature and the timing of this episode, we show that the use of climatological values to determine the amount of dust transferred to the domain from the larger scale are insufficient to reproduce the amplitude of the event.During this episode, the simulation outputs indeed showed a negative bias reaching 25 µg/m 3 over the whole region.Additionally, during all the simulation period, we highlighted a poor representation of the hourly variability of PM 10 in urban cells compared to observations which was attributed to an incomplete representation of urban driving sources.Our conclusions meet those of Astitha et al. (2008) which show in their overview on regional modeling that emissions inventories and dust transport are major elements in air quality forecasts.
In the second part of this paper, the model sensitivity to urban emissions, as well as the impact of enhanced dust boundary conditions on the model outputs was investigated.Our objective was to evaluate CTM improvement strategies to better represent particulate pollution in southern places of Europe.We highlighted the need for an improvement of anthropogenic PM emission inventories, especially concerning road traffic.We showed that the current gap between inventories and observations could reach a factor of 2 for PM 10 in urban centers.Quite recently, the COPERT methodology aiming at the calculation of air pollutant emissions from road transport has been updated to include resuspension by vehicles (http://lat.eng.auth.gr/copert/).The evaluation of inventories using this methodology needs to be performed to evaluate the degree of improvement reached in the model outputs.Finally, by implementing the CHIMERE-DUST outputs at the boundaries of our domain, we showed that the use of large-scale dust emissions and transport models provide most reliable PM 10 forecasts for public information, but also for PM 10 event analyses and further PM pollution regulation.Indeed, the "Thematic Strategy on Air Pollution" of the European Environmental Agency that follows from the Clean Air For Europe initiative (http://ec.europa.eu/environment/archives/air/cafe/activities/pdf/cbabaselineresults20002020.pdf)relies on modeling as its primary source of information, while using monitoring data to calibrate its chemical transport model EMEP (http://www.emep.int/).The modeled data obtained are further used in support of negotiations on emission reduction measures and for assessing of the feasibility of such measures.The precise determination of the reducible (anthropogenic) part of PM 10 by models thus constitutes a challenge for Mediterranean countries within the frame of the elaboration of particle-related European directives.In this study we underline the necessity for the nesting of CTMs by a large-scale model.
In Fig. 11 we present the results of a hybrid DUSTBOUND-EMISS simulation.The time series represent the hourly values in Aix and Marseille.It clearly appears that this simulation provides more consistent results and allows quantifying more precisely the regional particulate pollution, whether we focus on background pollution or on its urban variability and intensity.The hybrid simulation is still exposed to several biases and its empirical form cannot constitute a frame for daily air quality forecasts.However, it shows that a more realistic view of the particulate levels and evolution over the area can be reached by simple improvements.

Fig. 1 .
Fig. 1.Presentation of the simulation domain and location of the urban (black star) and industrial (black triangle) PM 10 measurement sites.

Fig. 2 .
Fig. 2. Wind (top) and PM concentration (bottom) sector plots at Arles for June and July 2006.

Fig. 4 .
Fig. 4. Observed (upper graph) and modeled (lower graph) daily mean concentrations of PM 10 at the 6 measurement sites

Fig. 6 .
Fig. 6.PM 10 hourly evolution in Marseille as calculated by CHIMERE (white line) compared to observations (black line).The simulated dust and SOA parts composing PM 10 are represented as cumulative areas.

Fig. 7 .
Fig. 7. Observed (blue line) and modeled (red line) PM 10 concentrations in Toulon, for a representative week of the studied period.

Fig. 9 .
Fig. 9. PM 10 time series from the reference (red line) and DUST-BOUND (black line) simulations in Aix (upper graph) and Marseille (lower graph) for the 3 selected weeks of June 2006.Observations are reported in blue.

Fig. 10 .
Fig. 10.Reference (red line) and EMISS (black line) simulations of PM 10 concentrations in Aix (upper graph) and Toulon (lower graph) for a representative week of June 2006.Observations are reported in blue.

Fig. 11 .
Fig. 11.PM 10 in Aix (upper graph) and Marseille (lower graph) as simulated by CHIMERE in the reference configuration (red line) and for the hybrid DUSTBOUND-EMISS simulation (black line) for a 3-week period in June 2006.Observations are shown in blue.