Atmospheric emission inventory of Maurienne valley for an atmospheric numerical model

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.


Introduction
Air pollution modelling is a powerful tool capable of assisting policy makers in developing abatement strategies to reduce pollutant emissions and improve air quality. One of the most crucial data sets needed to initiate chemical and photochemical reaction mechanisms included in the model is the emission inventory. In fact, the precision with which the emission scenario is generated can be considered as one of the major limiting factors in performing valid simulations of the emission fields for primary and secondary pollutants.
Since the beginning of the 1980s, a number of anthropogenic and biogenic emission inventories have been carried out, mainly in the most industrialized countries. These inventories generally cover large areas like Europe (OECD, 1984;Veldt, 1991;Lenhart and Friedrich, 1995) or the United States (Voldner et al., 1980;Clark, 1980;Lamb et al., 1987Lamb et al., , 1993 and are used to perform large-scale transport modelling. More recently, other modelling programs have focused on pollutant dispersion and ozone emission at regional or more local scales (Ponche et al., , 2000Khatami et al., 1998;Cros et al., 2004), most of them using the CORINAIR methodology (e.g. Aleksandropoulou and Lazaridis, 2004). There are also emission inventories for air quality modelling in mountainous areas but in valleys significantly broader than the Maurienne (Sturm et al., 1999;Couach et al., 2003).
There is now a growing concern over the impact of atmospheric pollution on public health. Following the accident in the Mont-Blanc tunnel on 24 March 1999, (France). The heavy-duty traffic (about 2130 trucks per day) was diverted to the Maurienne valley, with up to 4250 trucks per day. Within the framework of the Pollution in Alpine Valleys project (Jaffrezo et al., 2005) measurement campaigns were undertaken for several one week periods. Modelling in POVA should help to explain the process leading to episodes of atmospheric pollution, both in summer and in winter. The general topics of the program are comparative studies of air quality and the modelling of atmospheric emissions and transport in these two French alpine valleys before and after the reopening of the tunnel to heavy-duty traffic in order to identify the sources and characterize the dispersion of pollutants. The program combines 3D modelling with field campaigns in order to study the impact of traffic and local development scenarios.
Bottom-up and top-down methodologies are combined (depending on available data) to map emissions with a yearly resolution. The emissions are then distributed for modelling at the hourly level by taking account of the season and the day of the week. The corresponding software was designed in order to update the emission inventory as easily as possible. Changes in pollution may therefore be documented year after year. Abatement scenarios may be built by appropriately changing total vehicle characteristics, traffic and industrial activities. It is also easy to update emission factors whose values are expected to be improved in the future thanks to better knowledge of emissions.
After a description of the Maurienne valley, we discuss the preparation of two inventories with emission factors from CORINAIR and OFEFP: species, spatialisation methodology and emission sources. The two inventories are then used and after disintegration (with time and according to chemical species), the results from the photochemical model are analysed by comparing them with the measurements from the summer 2003 intensive period of POVA observations. The Maurienne valley differs noticeably from sites considered in most emission inventory studies, where urban contributions are significant. Because of the narrowness of the valley, sources have to be described at a fairly short length-scale. Inhabitants and industrial activities are concentrated in limited areas along the bottom of the valley. A large part of the domain of interest is covered with forests and pastures.
Finally, results from an air quality model of the Maurienne valley are considered since the emission inventory is designed to be used as an input to such a model and the emission fluxes cannot be compared with measurements, in contrast to concentrations computed from modelling to be examined side by side with POVA data.

Area of interest
The Maurienne valley (130 km) is a typical arcshaped ( Fig. 1)   Vegetation is relatively dense (Fig. 2). The mountain slopes are devoted to residual agriculture and above all full nature tourism activities. The Vanoise national park (peripheral and central zones) extends from a line across Méribel, Val-Thorens and Modane to the east of the domain; it was the first French park to be created, in 1963. This duality makes the Maurienne valley a sensitive site, particularly representative of the problems of sustainable development in alpine areas, with air quality as an important challenge.

Reference year and spatial resolution of the emission inventory
The year 2003 was chosen as reference year and all the data gathered in order to build the emission inventory are related to this period. However, when data were missing for this time span, those from the year 2001 were used. Considering the domain for modelling air quality (74*62 km), the spatial resolution was chosen to be 1 km in order to be the same as that of the chemistry model in d4.1 Simulation modelT. One of the primary tasks was to locate all the emission sources and to determine the spatial emission distributions in order to draw up emission maps for each species or group of compounds, for different time resolutions (yearly or hourly emissions). This was achieved by using Geographic Information Systems (GIS). These GIS were also used to determine the land use and geographic coverage of the domain.

The different species of the emission inventory
All the major compounds emitted both by anthropogenic and natural sources (mainly forests and grassland) are considered. They are necessary and sufficient to model atmosphere chemistry. The emission inventory of anthropogenic and biogenic emissions of NOx and NMVOC makes it possible to model ozone chemistry at mesoscale. Emissions from anthropogenic sources mainly result from the use of different types of fossil fuel and from solvent evaporation. The levels and composition of these emissions depend on several main parameters such as combustion processes, temperature, filtration devices etc. (combustion) and on the conditions of use and recovering procedures, when they exist (solvents). Finally, the compounds in our emission inventory are: sulphur dioxide (SO 2 ), nitrogen oxides (NOx), carbon monoxide (CO), methane (CH 4 ), non-methane volatile organic compounds (NMVOC) and particles. Contrary to most emission inventories, methane is explicitly considered since it is a precursory gas of ozone when NOx concentration and solar exposure are high enough. The NMVOC come both from natural (vegetation) and anthropogenic sources. Due to the wide range of volatile organic compounds (VOC) emitted, individual species cannot be taken into account. Disintegration of VOC for the purpose of modelling will be explained in section d4.2 NMVOC disintegration for RACMT. Particles are not yet taken into account in chemistry modelling.

Spatial distribution of emissions
The main problem arising in such a study is due to the need to collect a very large number of data of many different origins, which are organized in quite different ways. Data come from different levels: national, regional and local administrations and offices, and private and public companies for some particular sources. Primary data generally correspond to activity measurement associated with road traffic, private heating and industrial production. In order to be used as an input for the atmosphere chemistry model, they have to be processed to produce species mass fluxes. The methodology for spatialisation proposed and developed in several emission inventories (Ponche, 1992 Khatami et al., 1998) is summarized in Fig. 3.

Classification of emission sources
The classification includes both natural and anthropogenic sources. Although anthropogenic sources represent the major part of atmospheric emissions, natural sources cannot be ignored, especially when their contributions to NMVOC emissions are considered. The sources were split into 4 main categories, according to their origins: ! Biogenic emission sources, which include emissions from forests and grassland (NMVOC).
! Emissions from the service sector, which include emissions from public and private administrations and services (hospitals, public offices, etc.) where there is energy consumption. The contribution of private housing (heating, warm water production) is taken into account.
! Emissions due to transport, which include road traffic (light and heavy vehicles, motorbikes) out of and in the towns.
! Emissions due to industrial activities.
Whereas the above classification was used mostly for data collection, the emission inventory was built by considering a different classification based on the geometrical characteristics of the different types of sources. In this way, we distinguished three types of source: ! Point sources: large amounts of pollutants are emitted from very small and well-located areas (small factories, boilers, gas stations).
! Line sources: these take into account emissions from road traffic. ! Surface sources: these include all sources which are not included in the two previous categories (forest, grassland, large factories, use of solvents by the population).

5
All the classes of the emission inventory are given in Table 1. There is considerable rail traffic in the valley but it does not contribute to emissions since the lines are electrified.
2.6. Source evaluation method 2.6.1. Heating emissions The heating of housing and industries as well as the production of hot water contribute to the emission of pollutants. Inhabitants and industries in the valley mainly use fuel oil. The emissions from heating were spatially distributed within the population (INSEE data from 1999 census).

Domestic solvent emissions
Emissions of domestic NMVOC (mostly solvents) are very difficult to estimate. Each person is a potential emission source; the CORINAIR (CORINAIR, 2003) gives an emission level of 2.6 kg/year/inhabitant and the OFEFP 1.1 kg/year/ inhabitant. The emissions were then distributed spatially within the population (INSEE data from 1999 census).

Industrial emissions
All the major stationary combustion and production installations are considered in this category. The point sources are defined according to the levels of SO 2 , NOx and VOC emissions or according to energy consumption. In France, point sources have been listed by the authorities to recover taxes on SO 2 and NOx emissions. Therefore these large emission sources are generally well identified and located.

Traffic emissions
Road traffic is one of the most significant anthropogenic sources of NOx and NMVOC. To perform this part of the emission inventory, we had to make several assumptions according to the characteristics of the traffic. Two types of traffic are distinguished: urban and non-urban. Outside the towns, all the roads were considered as line emission sources, while in the towns, due to the dense network of streets, these were classified as surface sources. Even vehicles were divided into several categories according to their weight and engine type. Table 2 shows that average speeds are adapted to the type of road (COPERT III, 2001).
To determine traffic intensity, official vehicle figures from the dDirection Départementale de l'Equipement de la SavoieT and dSociété Française du Tunnel Routier du FréjusT (highway traffic figures) were used. Records are available at each exit from the highway, and along the main road (5 recording places). Data from the INSEE were used to describe urban traffic. This national institute reckons that every inhabitant over 6 years old makes an 8.7 km journey three times per day. The emissions were then spatially distributed within the population.
The emission inventory adjusts emissions to the road gradient with COPERT III software (COPERT   of transit traffic (Table 3). Heavy vehicles passing through the tunnel are classified according to their emission levels into four classes, namely euro0, 1, 2 and 3. They are charged fees according to their class.

Biogenic emissions
Emissions from vegetation mainly result from biophysical processes such as evapotranspiration. They depend on brightness and temperature. A simple calculation is made over the domain (Eq. (1)), without road gradient effects (Guenther et al., 1996).
where e is the average emission potential (emitted mass/vegetation mass/time, yr À1 ) for each particular species, D is the foliar biomass density (g dry weight foliage m À2 ), and C represents the integrated value of a dimensionless environmental correction factor representing the effects of short-term temperature and solar radiation changes in emissions. The real vegetation area (2867 km 2 ) (Fig. 2) is taken into account by making a difference between emissions from deciduous forests, coniferous forests, grassland, tundra and soil. The specific species which are identified in the biogenic emission inventory are: isoprene, monoterpenes, other NMVOC and NO (from soil). The entire methodology is described in the Atmospheric Inventory Guidebook (CORINAIR, 2003).

Temporal disintegration
Hourly temporal disintegration is carried out in order to provide hourly emissions for the photo- (2)) were applied to all types of sources.
With E h,i the hourly emitted volume; E y,i the yearly volume of pollutant I; M k the monthly disintegration coefficient; W k the weekly disintegration coefficient; H k the hourly disintegration coefficient.
The different disintegration coefficients are detailed in Figs. 4, 5 and 6.

Results of the two emission inventories
Two different inventories are considered below: ! The emission inventory referred to as inventory A was drawn up using industrial, heating and solvent emission factors from CORINAIR. ! The emission inventory referred to as inventory B was drawn up using industrial, heating and solvent emission factors from OFEFP.
Inventories A and B are based on the same emission inventory for biogenic sources (Table 7 with CORINAIR sources) and road traffic (Table 8) with the use of COPERT III (COPERT III, 2001).
Past comparisons of emission inventories already showed disparities (Seika et al., 1996). Significant differences can be observed in the emission factors from CORINAIR and OFEFP. Variations in the yearly total emissions show differences ranging from 14% to 95%. The OFEFP emission inventory values are always lower than the CORINAIR values (Tables 9, 10 and 11).
The spatial distribution of the total emissions of SO 2 , NOx, NMVOC and CO is presented in Fig. 7.
Similar maps could easily be produced for each type of source. In this Fig. 7, the main point sources, roads and urban areas can easily be located. Emissions are concentrated in the bottom of the valley.

Simulation model
We used a modelling system: numerical simulations combine the mesoscale atmosphere model ARPS 4.5.2 and the troposphere chemistry model TAPOM 1.5.2. This modelling system was improved for the Chamonix valley, where a horizontal resolution of 300 m was adopted. The meteorological fields from ARPS may be viewed as realistic enough to induce the transport and mixing of the chemical species. The numerical simulation and satisfactory results are detail in Brulfert, 2004 andChemel et al., 2004) for both the Chamonix and the Maurienne valleys. The methodology for the Chamonix emission inventory is basically identical but suffered from a lack of data.

Atmospheric dynamics model
Large eddy simulation was used to compute mesoscale flow fields. The numerical simulations presented here were run with the Advanced Regional Prediction System (ARPS), version 4.5.2 (Xue et al., 2000(Xue et al., , 2001 with a horizontal resolution of 1 km. Lateral boundary conditions were externally-forced    (Grell et al., 1995).

Atmosphere chemistry model
ARPS is coupled off-line with the TAPOM 1.5.2 atmospheric chemistry code (Transport and Air Pollution Model) developed at the LPAS of the EPFLausanne (Clappier, 1998;Gong and Cho, 1993). With 1-km grid cells to calculate dynamics and reactive chemistry, it is possible to represent dynamics in the valley accurately (valley and slope winds) (Anquetin et al., 1999) and to process chemistry at a fine scale.
TAPOM is a three-dimensional eulerian model with terrain following mesh using finite volume discretisation. It includes modules for transport, gaseous and aerosol chemistry, dry deposition and solar radiation. TAPOM uses the Regional Atmospheric Chemistry Modelling (RACM) scheme (Stockwell et al., 1997). The mechanism includes 17 stable inorganic species, 4 inorganic intermediates, 32 stable organic species (four of these are primarily of biogenic origin) and 24 organic intermediates, in 237 reactions. In RACM, the NMVOCs are aggregated into 16 anthropogenic and three biogenic species. Boundary conditions are calculated from CHIMERE, a regional ozone prediction model, from the Institut Pierre Simon Laplace, which gives concentrations of chemical species at five altitude levels (Schmidt et al., 1998) using its recent multi-scale nested version.

NMVOC disintegration for RACM
The TAPOM model uses the RACM scheme. It includes 32 organic species. The total mass of NMVOC is disintegrated for this mechanism with coefficients presented in Table 12 (SEDE S. A., 1996). These coefficients depend on the emission sources.
In the case of biogenic emissions, NMVOC consist of isoprene, monoterpenes and other VOC's (described in Table 7). The mass percentage of  Aromatics with OH rate constant less than 13.6Â10 À12 cm 3 s À1 isoprene and monoterpenes in the total NMVOC is defined as shown in Table 13 (Hewitt, 1998). In the case of NOx, the distribution is 95% of NO and 5% of NO 2 for all classes except biogenic, with only NO emission from soil.

Model results
Two simulations were performed according to the results of the two inventories (A and B) during the summer 2003 intensive period of observations (24-30 June), the first (run A) with the CORINAIR inventory (inventory A), and the second (run B) with the OFEFP inventory (inventory B). Runs A and B differ in terms of the evaluation of domestic and industrial emissions. A and B are based on the same inventory of biogenic sources and road traffic as presented in Table 14.
During this intensive period of observations, several gas measurements were made at different sites along the valley. The measurements of CO, NOx, SO 2 and O 3 may be compared with the model values in order to evaluate the impact of emission factors. Particles are not taken into account at the moment by the model.
Six measurement sites were selected to compare the model and the measurements (Table 15). Comparisons of CO, NOx, SO 2 and O 3 measurements with results from simulations A and B are presented in Figs. 8, 9 and 10 from June 24th to 30th. Results from runs A and B do not differ significantly from one another. This is due to the predominance of traffic emissions with no significant emission from heating in summer.
In the case of SO 2 , NOx and O 3 , there is a good agreement between the model and the measurements. Measurement sites are representative of the mean pollution in the valley. The discrepancies which are observed in CO peak values at night (Fig. 10) are induced by inaccuracies in calculating inversion height and should not be attributed to errors in the emission inventory. Anyway these concentrations are weak, since they are 50 times lower than the air quality standard (8591 ppb, on 8 h).
In the case of ozone (Fig. 8), both runs give a good agreement with measurements from urban and rural stations. Both the spatial and temporal variability of the simulated ozone concentrations correspond reasonably well with the measured values. The measured (and model) concentrations of CO, NOx, O 3 , CO and SO 2 are weak compared with air quality standards.

Conclusion
An atmospheric emission inventory was developed over an alpine area (the Maurienne valley) with a spatial resolution of 1 km 2 for the reference year 2003, with the methodology combining bottom-up and topdown reasoning. The area, which covers 4588 km 2 , is sensitive to emission sources: traffic, industries, heating, domestic solvents, biogenic. During the POVA summer period of intensive observations in 2003, the major sources were road traffic and   industrial pollution according to the results of the inventory. As expected, the maps of the different major pollutants show significant emissions due mainly to the highway and industries. The emission factors from BUWAL-OFEFP differ from those from CORINAIR. Differences in factor values are up to 538% and variations in the total emissions, for year 2003, range from 14% to 95%. Several interesting points can be noted: the main pollutant by far (in terms of quantity) is CO, 60% of which results from road traffic, the remaining part being nearly entirely due to domestic heating (31%). The road traffic itself is the major source of NOx, VOCs and CO. It represents 75% of the emissions of NOx, 65% of the anthropogenic emissions of VOC, 28% of the total VOC emissions (biogenic and anthropogenic). In the case of SO 2 , the main sources are industrial emissions, which represent 78% of the total. Biogenic emissions of NMVOCs, which represent 47% of the total VOC emissions, are far from being negligible and must be taken into account in modelling the atmosphere.
Two simulations were performed with the photochemical model. Both take account of the same traffic and biogenic emission factors. They differ in terms of emissions from industries, heating and solvents. Concentrations computed from the numerical simulation do not show any significant difference and are in agreement with the measurement values. This is due to the preponderance of traffic in summer. Concentrations of CO, NOx, O 3 , CO and SO 2 are weak compared with air quality standards but nevertheless the values computed from the model remain significant.