Updated African biomass burning emission inventories in the framework of the AMMA-IDAF program, with an evaluation of combustion aerosols

. African biomass burning emission inventories for gaseous and particulate species have been constructed at a resolution of 1 km by 1km with daily coverage for the 2000– 2007 period. These inventories are higher than the GFED2 inventories, which are currently widely in use. Evaluation speciﬁcally focusing on combustion aerosol has been carried out with the ORISAM-TM4 global chemistry transport model which includes a detailed aerosol module. This paper compares modeled results with measurements of surface BC concentrations and scattering coefﬁcients from the AMMA Enhanced Observations period, aerosol optical depths and single scattering albedo from AERONET sunphotometers, LIDAR vertical distributions of extinction coefﬁcients as well as satellite data. Aerosol seasonal and interannual evolutions over the 2004–2007 period observed at regional scale and more speciﬁcally at the Djougou (Benin) and Banizoum-bou (Niger) AMMA/IDAF sites are well reproduced by our global model, indicating that our biomass burning emission inventory appears reasonable.


Introduction
Western African aerosols are complex mixtures of combustion aerosols emitted from biomass burning (mainly savanna fires), domestic fires (fuelwood and charcoal), fossil fuel sources (traffic, industry), together with dust particles from the Sahel and Sahara (Liousse et al., 1996;Haywood et al., 2008).The temporal distribution of these sources is rather well known: open biomass burning aerosols are only produced during the dry season, whilst domestic fires for cooking as well as traffic and industrial aerosols are present throughout the year.Dust particles are also present all year round, though with some seasonal variability, with intense dust events tending to occur in spring (Prospero et al., 2002;Tulet et al., 2008).Some contribution of secondary organic aerosol from vegetation and nitrates from soil NO x is also expected, particularly during the wet season (Delon et al., 2010;Capes et al., 2009).Subsequent to emission, anthropogenic biomass burning particles and natural mineral dust particles become internally mixed to varying degrees.The study of such aerosol mixtures is of particular interest, because their different sizes, chemistries and optical properties may result in significant radiative, climatic and health Published by Copernicus Publications on behalf of the European Geosciences Union.
Before the AMMA (Multidisciplinary Analysis of African Monsoon) program (http://www.amma-international.org/),only a few field studies have discussed Western African aerosols and only for selective periods.For example, experimental work on emissions was performed within the DE-CAFE (Lacaux et al., 1995) and EXPRESSO (Delmas et al., 1999) programs, while studies on deposition have been carried out within the IDAF (IGBP/IGAC/DEBITS in Africa) network (http://www.medias.obs-mip.fr/IDAF;Galy-Lacaux et al., 2009) and a number of studies on aerosol radiative impacts have been performed within the AERONET network (http://aeronet.gsfc.nasa.gov/;Holben et al., 1998).Initial global modeling simulations of biomass burning aerosols associated with experimental work and the development of emission inventories may be found in Liousse et al. (1996) and Liousse et al. (2004).In parallel, dust emission parameterizations (Marticorena and Bergametti, 1995) have been developed and studies have been performed investigating Saharan dust within National and European programs (e.g. the Saharan Dust Experiment, SHADE, Tanré et al., 2003).Initial experimental and modeling work discussing heterogeneous aerosol processes including HNO3 and dust interactions were conducted e.g. by Galy-Lacaux et al. (2001) and Bauer et al. (2004).
The AMMA program offers a unique opportunity to study Western African aerosols and their impacts, since participants in the project are working together on many different aerosol features, such as emissions, modeling and measurements of both dust and combustion aerosols.Within the program, different observation periods were considered: both long term (AMMA-LOP, Long Observation Period/IDAF, 2005-2010) and short term (EOP, Enhanced Observation Period, 2005-2007) observation periods were combined with intensive campaigns (SOP, Special Observation Period, 2006) (Lebel et al., 2009).Moreover, observations from the AMMA SOP experiments have been further enriched by the DODO and DABEX (Haywood et al., 2008) experiments.Many recent papers pertaining to West African aerosols show new and interesting results obtained from these campaigns (Marticoréna et al., 2010).For example, Haywood et al. (2008) determined the vertical distribution of West African aerosols during the fire season, showing that a dust layer exists at the surface while an elevated smoke layer persists above.New chemically resolved aerosol observations are described in Capes et al. (2009), Léon et al. (2009), Kim et al. (2009), Chou et al. (2008), Johnson et al. (2008), Osborne et al. (2008), Rajot et al. (2008).Heterogeneous processes have been studied by Crumeyrolle et al. (2009) and Matsuki et al. (2010).Myhre et al. (2008) and Pinker et al. (2010) simulate aerosol mixtures of mineral dust and biomass burning aerosol at regional scales and resolutions.Finally radiative and climatic impacts for such aerosols have been investigated by Mallet et al. (2008), Derimian et al. (2008), Raut and Chazette (2008) and Solmon et al. (2008).
Our present work focuses on the development of the AMMA biomass burning emission inventory AMMABB with particular focus on combustion aerosols.
A first challenge was to develop real-time biomass burning emissions for the AMMA EOP period (2005-2007) for gases and particles.Calculating emissions from savannas, forests and agricultural fires requires a knowledge of appropriate emission factors as well as spatial and temporal distributions of burnt biomass.First attempts to quantify burnt biomass on a regional/global scale were based on FAO (United Nations Food and Agriculture Organization) land use statistics complemented with general assumptions about percentages of burnt surfaces per year (Hao et al., 1990;Hao and Liu, 1994;Liousse et al., 1996).Following Hao et al. (1990), it was assumed that 50-75% of west african savannas burned each year, though with a factor of 2-3 in uncertainty.The first qualitative improvements in fire temporal evolution were achieved with the use of active fire satellite products (Cooke et al., 1996).Later, burned area products were used to improve quantitative estimates (Liousse et al., 2004) particularly over Africa where such products have been validated (Tansey et al. 2004;Jain et al., 2008;Stropppiana et al., 2010).Inventories have since been developed using either active fire, burned area products or both (Chin et al., 2002, Ito andPenner, 2004;Hoelzemann et al., 2004, Liousse et al., 2004;Michel et al., 2005;Van der Werf et al., 2006, Generoso et al., 2007), an approach which appears to reduce uncertainties (Michel et al., 2005 andStroppiana et al., 2010).Very recently, determination of satellite fire radiative energy has been seen to be directly correlated with fire emissions and allowing the development of new inventories (still in validation) without vegetation parameterization and consequently with reduced uncertainty (Wooster et al., 2002(Wooster et al., , 2005;;Ichoku and Kaufman 2005).In our work, presented in Sect.2, SPOT-VGT satellite products were used to develop emission inventories following the methods of previous studies performed over Africa, using the AVHRR burned area products for 1980-1989(Liousse et al., 2004) ) and the SPOT GBA 2000 products for the year 2000 (Mieville et al., 2010 andKonare et al., 2010).
A second challenge was to verify the use of these new biomass burning emissions in global/regional transport models, with comparisons with surface, satellite and columnintegrated measurements obtained during the AMMA experiment.In this paper, we use the global ORISAM-TM4 chemistry-transport model, which is specially adapted to our needs.ORISAM-TM4 does not separately consider different aerosol components, and AOD values are thus not calculated by adding the effects of all components as, for example in Kinne et al. (2006) or Koch et al. (2009).The new code used in the present study, ORISAM-RAD, includes the calculation of time-evolving aerosol size distributions Atmos.Chem. Phys., 10, 9631-9646, 2010 www.atmos-chem-phys.net/10/9631/2010/ in six bin sizes between 0.04 µm to 40 µm.A detailed aerosol chemistry module is coupled with an atmospheric gas scheme, through thermodynamical equilibria between gas and liquid phase for inorganic components (sulfate, nitrate, ammonium), whilst an empirical approach based on grouping VOCs with similar volatilities and chemical properties is used for secondary organic aerosol (Odum et al., 1996).Major primary aerosols are black carbon (BC), organic carbon (OC), dust and sea salt, dissociated into chlorine and sodium ions.Water is included as a reactive component in online optical properties calculations, as for all other components.Ensuing optical properties (extinction, scattering and absorption) are then calculated assuming internally core/shell mixed aerosols (Mallet et al., 2005(Mallet et al., , 2006;;Péré et al., 2009).These approaches are similar to Jacobson (2001), Ghan et al. (2001), Stier et al. (2005) and Bauer et al. (2010).A global simulation was performed, covering the EOP period, using the ORISAM-RAD code implemented in the TM4 global chemistry-transport model.Our study also took advantage of the very recent seasonal AMMA aerosol measurements.Of special interest were the AMMA LOP (with IDAF and AERONET programs) and EOP periods, which were more coherent with the global model spatial and temporal resolutions.Particular focus was made on savanna sites highly impacted by biomass burning, especially at the new Djougou (Benin) site (Serc ¸a et al., 2007, Galy et al. 2007, Pont et al., 2009).A description of the model and comparison between simulation results and measurements is displayed in Sect.3.

Methodology
Daily, open global biomass burning emissions for gases (CO 2 , CO, NO x , SO 2 and VOC) as well as black carbon (BC) and organic carbon (OC) particles were estimated for the AMMA campaign period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007), at a horizontal resolution of 1 km×1 km over Africa, using a bottom-up approach.The following relationship was considered: where EM is the total emission estimate and BA the Burnt Area described by the L3JRC product for the period 2000-2007 (Tansey et al., 2008) at a spatial resolution of 1 km×1 km, and at daily temporal resolution.These data are derived from the Spot-Vegetation satellite.GLCv represents the percentage of each class of vegetation v present within each cell of the 0.5 • ×0.5 • vegetation map.The GLC vegetation map (Global Land Cover, developed at JRC-Ispra (http://ies.jrc.ec.europa.eu/global-land-cover-2000) is used, since optimized for studies in Africa for the year 2000.Account is made of 21 different classes of vegetation from which 14 can support fire events.
BDv and BEv are, respectively, the biomass densities (biomass per unit area occupied by a vegetation type in kg/m 2 ), and burning efficiencies (fraction of biomass exposed to fire and actually consumed in fire) for each vegetation class, v.An important work performed in the frame of the AMMA and GICC (Gestion et impacts du changement climatique, www.gip-ecofor.org/gicc/)programs was to derive the values of BDv, BEv for each GLC vegetation class (see also Mieville et al., 2010) (2010).EFv is the emission factor (given in g/kg of dry matter) for the particular gaseous or particulate species considered, and depends on the vegetation class.EF values for black carbon and organic carbon are taken from Liousse et al., (2004).Other EF values for gaseous species (CO 2 , NO x , VOCs, SO 2 . . . ) were selected following Andreae and Merlet, (2001).Table 1 resumes all values used for BDv, BEv and EFv for black carbon particles.
It should be noted that a detailed comparison performed for the year 2000 between the L3JRC and Landsat satellite burned area products with a 30 m×30 m spatial resolution (http://geo.arc.nasa.gov/sge/landsat/l7.html) revealed systematic underestimation of burned areas in the L3JRC product as compared to Landsat data for the GLC3 and GLC12 vegetation classes.Consequently, corrections were applied in this study, using values obtained from this comparison.Burnt areas for the GLC3 and GLC12 vegetation classes, typical of west African ecosystems, were increased by factors of 2 and 1.67, respectively (JRC report September 2007, Grégoire, 2007).This specific version of the inventory is called L3JRcor in the following section.

Results
Maximum BC emissions (∼0.9 TgC/yr)   Werf et al., 2006Werf et al., , 2008) ) have been included for comparison.In both cases (total Africa or West and Central Africa), we note that AMMABB estimates are higher than GFED2.This feature was also noted shown in Tummon et al. (2010) and Barret et al. (2010).In order to better understand such differences, a joint international effort aimed at improved identification of uncertainties in existing biomass burning inventories has been organized.The global inter-comparison exercise INTERMEDE BBSO (see the ACCENT European Network website, http://www.accent-network.org)aimed to analyze CO emissions derived from different satellite products in order to further investigate these uncertainties (Stroppiana et al., 2010).The best agreement was found for CO emissions over the African continent, with a maximum difference of about 45% between products for the year 2003.Figure 3 presents monthly CO budget results for Africa for 2003, as derived by a number of published inventories.The same vegetation parameterization and emissions factors were used by Mieville et al., 2010, Chin et al., 2002 and in this work; however different burnt biomass distributions were obtained from ATSR active fires scaled with GBA 2000 burnt areas, MODIS active fires and L3JRC burnt areas, respectively.A dynamical vegetation model was used in combination with the MODIS active fire product to produce the GFED2 inventory, whilst the dynamical vegetation model was combined with both the MODIS active fire and burnt area products to generate the GFED3 inventory.Pétron et al. (2004) used a top-down approach from MOPITT CO concentrations.As already mentioned, the GFED2 and GFED3 inventories provide the lowest estimates throughout the year.L3JRCcor (this work) is higher than all other inventories during the JJA season due to significant burning in the savanna/grassland biomes.The Chin et al. (2002) inventory is higher in winter as a result of considerable burning in the agricultural sector.Further details of this comparison can be found in Stroppiana et al. (2010).

Evaluation of AMMABB emission inventories through comparison of model results and atmospheric observations
The AMMABB inventory for combustion aerosols has been evaluated using the global ORISAM-TM4 chemistrytransport model.For this purpose, simulation results using the emissions inventory described in the previous section were compared with the recent AMMA measurements.

Model description
The ORISAM-TM4 model (Guillaume et al., 2007) 1994;Van Velthoven and Kelder, 1996;Houweling et al., 1998).A version with 9 vertical levels is coupled with the ORISAM model, an aerosol model developed at the Laboratoire d'Aérologie (Liousse et al., 2005;Bessagnet et al., 2009).This latter model takes into account the chemistry of ozone and its precursors (Houweling et al., 1998), as well as the chemical formation of secondary organic and inorganic aerosols (sulfate, nitrate, ammonium, and water) from condensation of gaseous precursors onto primary particles (BC, primary OC, dust).Size and microphysical aging are also taken into consideration.Sea-salt (chlorine and sodium) particles are included.Thirty-eight gaseous compounds are involved in ozone chemistry and secondary aerosol formation.
In the present work, 6 classes of aerosol diameters are included (size bins from 0.04 µm to 40 µm), together with nucleation (Kulmala et al., 1998), coagulation and condensation processes (Gelbard et al., 1980).Dust and sea-salt emissions are produced following the parameterizations of Zakey et al. (2006) and Gong et al. (2003).Aqueous chemistry for sulfate particles in cloud droplets follows the work of Dentener and Crutzen, (1994).Heterogeneous reactions on sea-salt are included, following Nenes et al. (1998) whereas coarse nitrate formation on dust follows Bauer et al. (2004) and Hodzic et al. (2006).Finally, H 2 , NO 2 , N 2 O 5 , and NO 3 heterogeneous chemistry is based on Jacob (2000).
Aerosol optical properties are calculated with the RADiative (RAD) model (Mallet et al., 2005(Mallet et al., , 2006) ) applied to ORISAM-TM4 results, assuming a 3-layer internally mixed aerosol structure.The particle core is assumed made of BC and dust, with a first shell of primary organic carbon and the second one of sea salt and secondary aerosol species (SOA, sulfates, nitrates).
Biomass burning emissions for gases and particles are taken from the AMMABB inventory, as described in Sect. 2. These emissions also account for fire injection heights and different vertical distributions are introduced in the model for savanna and forest burning following Lavoué et al. (2000).The emissions from other natural processes and those resulting from anthropogenic activities are described in Guillaume et al. (2007).Natural emissions of oceanic DMS, volcanic sulfur and soil NOx are taken from the GEIA emissions portal (http://www.geiacenter.org/),while NOx produced by lightning and organic biogenic precursors (terpenes) are taken from the POET inventory (Granier et al., 2005).Anthropogenic black and organic carbon emissions from fossil and bio-fuels are adapted from Junker et Liousse (2008) for the year 2003 (Assamoi and Liousse, 2010).Anthropogenic emissions of gaseous compounds (SO2, NOx, NH3) are extracted from the EDGAR3.2inventory for the year 1995 (Olivier et al., 1999, http://www.mnp.nl/edgar/model/v32ft2000edgar/), except for CO and VOCs, from the POET inventory (Granier et al., 2005), and NOx aviation emissions taken from the ANCAT 1998 database (Gardner et al., 1997).Primary sulfate emissions are assumed to be 2.5% of SO 2 emissions.The selected anthropogenic emissions appeared to be the most suitable dataset available for this study.Recent developments, for example by Assamoi and Liousse (2010), will improve the available choice of inventories, with an expected increase of emissions in West African megapoles in this inventory.
Emissions of carbonaceous aerosols appear in two modes in the sectional model.The first mode has a mass diameter of 0.23 µm while the second mode (only including 20% of OC emissions) has a mass diameter of 2.5 µm (Cousin et al., 2005).BC and OC emissions are partitioned into hydrophobic and hydrophilic fractions.For both species, the transition from hydrophobic to hydrophilic types through absorption, condensation and coagulation is parameterized using a simple constant turnover time (Cooke et al., 1999), since no global consensus exists as for ageing and hygroscopicity of carbonaceous aerosol.Secondary organic aerosols were considered fully hygroscopic.Wet and dry deposition parameterizations are described in Guillaume et al. (2007).In particular, wet deposition, apart from including in-cloud and below cloud scavenging also accounts for SO 2 oxidation within cloud droplets and rain drops.
A global simulation was performed with the ORISAM-RAD-TM4 model for the entire AMMA campaign, including the EOP (Enhanced Observation Period), from December 2004 to March 2007, with a spin up period of one month.The model was forced with ERA 40 (ECMWF) wind and precipitation fields, updated every 6 h.

Black carbon surface concentrations
Comparisons between modeled and observed BC surface concentrations at AMMA/IDAF super-sites (see the map in Fig. 4a) are presented for Banizoumbou (Niger) from December 2004 to November 2005 (Fig. 4a), Lamto (Ivory Coast) from May 2006 to October 2006 (Fig. 4b) and Djougou (Benin) from June 2005 to March 2007 (Fig. 5a).Observed BC concentrations at Djougou and Banizoumbou were measured using a seven-wavelength aethalometer.The absorption coefficients which were measured by aethalometers at the AMMA super-sites were influenced not only by highly absorbing combustion aerosols, but also by dust particles containing various absorbing fractions (Alfaro et al., 2004).In order to take this into consideration, we applied the methodology of Fialho et al. (2005) to the measured absorption signal at Djougou and Banizoumbou to correct for dust interaction.Absorption values were thus decreased on the order of 9-13%.No correction is applied to the data obtained from Lamto, where BC values were obtained using a single-wavelength aethalometer.Many previous studies have further suggested that aethalometer measurements need to be corrected for multiple scattering by the filter fibers, scattering of aerosols embedded in filters and filter loading artifact (Liousse et al., 1993;Petzold et al., 1997;Bond et al., 1999).Collaud Coen et al. ( 2010) have summarized the different corrections (respectively the "Weingartner", the "Arnott", the "Schmid", the "Virkkula" corrections) with new correction algorithms and recommendations.Most of the corrections require parallel aethalometer and scattering measurements.In our case, without such measurements and following Collaud Coen et al. ( 2010), the corrections at a biomass burning site would be a 23% reduction of absorbing coefficients (and thus of BC concentrations).Finally, even though the relative uncertainty regarding measured BC concentrations was first given to be on the order of 10% (Hansen et al., 1982), an uncertainty of 30% on aethalometer BC measurements has been retained in this study.
Mean BC concentrations as well as seasonal variability of BC concentrations are generally well reproduced by the model.At Banizoumbou, the model presents higher BC concentrations than observed during the wet season.This is possibly due to uncertainties in the parameterization of BC wet deposition in the model or as a result of errors in simulated precipitation.At Lamto (Fig. 4b) the model gives lower con-    centrations than observations in July 2006.Again, this may possibly be explained by excessive aerosol removal due to overestimated precipitation, a known error in the ERA-40 reanalysis in the Guinean Gulf region (Uppala et al., 2005).Despite relatively good agreement between model and observations at Djougou, modeled values using the L3JRCcor inventory are lower by 40-50% than observations in December 2005.Coarse horizontal and vertical grid resolution model at a specific site may explain such differences with observations.Moreover, as shown earlier, experimental data issued from aethalometer analyses, may be overestimated by up to 23%.Discrepancies between model and measurements would be reduced if such correction was applied.In Fig. 5a, modeled results are also given at Djougou for L3JRC inventory without the L3JRC satellite data correction (see Sect. 2).It is interesting to note that this inventory has a very large impact on modeled BC concentrations (with a modeled BC concentration decreasing by a factor of 6), thus emphasizing the importance of the choice of biomass burning inventory and, in this case, of the corrected version of the inventory.In summary, the model associated with the AMMABB biomass burning inventory satisfactorily reproduces the interannual BC variability, with both observed and modeled

Djougou aerosol optical properties
This section deals with the comparison between modeled and measured aerosol optical properties at Djougou.Scattering coefficients were measured from February 2006 to February 2007 with a single-wavelength (0.52 µm) ECOTECH nephelometer, with 10% error (Formenti et al., 2002) (Fig. 5b).
Aerosol optical depth (AOD) and column-integrated single scattering albedo (SSA) are retrieved at 441nm from January 2005 to March 2007 from direct sunphotometer and almucantar measurements (Dubovik et al., 2000) (error about ±0.01 (Holben et al., 1998) and ±0.03 (Dubovik et al., 2000), for AOD (Fig. 6a) and SSA (Fig. 6b), respectively).Experimental methods used to obtain these observations are described in Mallet et al. (2008).It should be noted that variations in such parameters as AOD, SSA, scattering coefficient include the impacts of both dust and combustion aerosol.As for BC comparisons, both mean values and seasonal variability of the simulated scattering coefficient are generally well captured by the model.Two main discrepancies between the model and observations can be observed.First, in June, the modeled scattering coefficients are higher than observations.Since this difference does not appear in the simulated BC concentrations (see Fig. 5a), this is likely mainly due to uncertainties in other aerosol components, such as dust concentrations.Interestingly, the March 2006 dust event described by Tulet et al. (2008) and Mallet et al. (2009) is well captured by the model.A second problem occurs in January 2007, when modeled scattering coefficients are lower than observations.This issue was already noted regarding BC concentrations, and can possibly be explained by the fact that local biomass burning sources may not be adequately taken into account in the coarse resolution model.
AOD comparisons are shown in Fig. 6a, for both coarse (AODc) and fine particles (AODf) at 441 nm.Reasonable comparison is found for fine particles, with modeled values slightly lower than observations (on the order of 41%), except for May and June 2006 when modeled values are approximately 18% higher than experimental ones.This is consistent with the previous comparisons for both BC and scattering coefficients (see Fig. 5a and b).The comparison is quite different for coarse AOD.From January 2005 to June 2006, modeled values are higher than observations (the difference ranging from 14% to 315%), whereas they are lower from August 2006 to March 2007 (the difference ranging from 0.1% to 56%).These differences are much more significant outside the main fire season, possibly due to uncertainties in the dust scheme.Finally, Fig. 6b shows a comparison between modeled and measured single scattering albedo (SSA).Mean SSA values, as well as seasonal variations are well captured by the model.However, the differences mentioned in Figs.5a, b and 6a for December 2006 and January 2007 (underestimates of both simulated scattering coefficient and AOD) result in lower modeled SSA values (i.e. more absorption) than observed.It is also possible that the 100% core/shell mixing treatment applied in the ORISAM-RAD model tends to overestimate aerosol absorption as observed by Péré et al., (2010).Indeed, increase of water uptake in case of high hydrophilic aerosol concentrations may reinforce light refraction from the aerosol shell into the aerosol BC core, thus enhancing particle absorption (Moffet and Prather, 2009).It is thus possible that a combination of externally-and internally-mixed particles could be more realistic in the African domain.
Similar general agreement is also found at Banizoumbou (Niger), particularly for fine particles, though comparisons between observations and model are more complex than at Djougou. Figure 7a displays AOD comparisons at 441 nm, respectively for both coarse (AODc) and fine particles (AOCf).Outside the fire season, model values are higher than observed ones whereas the reverse is true during the fire season, with differences on the order of 35%.This is similar to the trend already observed for BC concentrations in figure 4a.A possible explanation for the wet season can be tentatively found in aerosol deposition and transport whereas in winter, the difference is rather explained considering the ORISAM-TM4 dust scheme.As for AODc, they behave rather similarly to Djougou, model/observation differences being much higher than for AODf with higher modeled values from March 2005 to October 2006 and lower modeled values from November 2006 to March 2007.Finally figure 7b displays modeled and observed SSA values at Banizoumbou, very similar to the ones found in Djougou (Fig. 6b).Reasonable comparisons may be found except for the november 2006 to January 2007 period when modeled values are below observations.Such differences could be due to underestimations of modeled coarse particles, again revealing a possible flaw in the dust scheme.

Vertical distribution of the extinction coefficient at Djougou
Several previous studies have investigated vertical aerosol profiles in the West African region during the AMMA dry season (the DABEX experiment).Surface lidar and aircraft observations have clearly shown that during the dry season smoke aerosols are located above a dust-rich surface layer (Osborne et al., 2008;Haywood et al., 2008;Pelon et al., 2008;Leon et al., 2009).Johnson et al ( 2008) also demonstrated that these relative profiles (a smoke-rich aerosol layer overlying a dust-rich aerosol layer) are adequately represented by the Met Office Hadley Centre HadGEM2 climate model.However they found that the biomass burning smoke was underestimated by a factor of about 2.4 for January 2006 when using GFED1 emission data, which differs little from GFED2 data in northern Africa.Thus, the use of the AMMABB emission data set where emissions are a factor of ∼1.8 larger could help reduce the discrepancy between the model and observations (Table 2).
In our work, as a result of the low vertical resolution of ORISAM-TM4, it is very difficult to capture the complex vertical profile of aerosols that is observed over West Africa.However, in order to further investigate this aspect we conducted several tests focused on the ability of the ORISAM-RAD-TM4 model to accurately reproduce aerosol vertical profiles, particularly the aged smoke layer and its vertical distribution.In this regards, we used micropulse Lidar (MPL) observations (Pelon et al., 2008) at Djougou (Northern Benin) at 0.523 µm for two specific events, one occurring in January 2006, representative of mixed aerosols, and a second one in March of the same year, when mineral dust particles are predominant.
Figure 8a displays vertical profiles of aerosol extinction coefficients (in km −1 ) as estimated by the MPL and as calculated in the model.For January 2006, the MPL observations clearly indicate the presence of two distinct aerosol layers, one located between the surface and approximately 1.5 km and a second one located above including smoke aerosols.In this case, ORISAM-RAD-TM4 presents lower extinction coefficient values than observations over the entire profile, but does appear to capture the presence of the second layer (1.5-4 km) with values around 0.07 km −1 at 1.8 km and 0.03 km −1 at 3 km.
For the March 2006 case of pure mineral dust (Fig. 8b), the model similarly gives lower extinction coefficient values than observations, with a significant underestimation within the dust layer.Indeed, maximum simulated extinction coefficients are approximately 0.25 km −1 , while lidar observations suggest values closer to 1.4 km −1 .It is, however, interesting to note that the vertical position of the first layer given by the model and the MPL observations are in agreement.In January, this layer most likely mixing dust and fresh biomass burning aerosols appears between 0 and 1 km whereas in march, this layer is located near 1-1.5 km.At this point, however, it is difficult to isolate one particular source of error, and any one or a combination of factors such as the model vertical or horizontal resolution, model dynamics or dust production scheme could be involved.

Spatial distribution of aerosol optical depth over Africa
We further compared spatial distributions of simulated AOD over Africa data from the PARASOL satellite (http://www.icare.univ-lille1.fr/parasol/browse/,Fig. 9b and d (right) for the model, and 9a and c for PARASOL (left)), for January 2006 (top) and July 2006 (bottom).Note that the aerosol optical depths in the PARASOL data are derived at 0.86 µm, leading to lower estimates of aerosol optical depths than those derived from sun-photometers (Fig. 7) owing to strong wavelength dependence of the aerosol optical depth typical of biomass burning aerosols.Model simulated and observed AOD are of the same order of magnitude.This study shows that the use of updated biomass burning emissions in a global model including a complex aerosol module with core/shell optical treatment can produce quite satisfactory results.Regional patterns simulated by the model are in agreement with satellite data, in January (biomass burning in the Northern Hemisphere) and in July (biomass burning in the Southern Hemisphere).However, in our work there are still two obvious discrepancies.First, the maximum simulated AOD occurs too far eastwards in January, and second, in July, the simulated AOD maximum occurs over a larger area, extending further southwards than observed.We evaluated the impact of the AMMABB in- to the GFED2 biomass burning inventory in order to assess the difference between the AMMABB and GFED2 inventories.No improvement in the location of AOD maximum was observed and comparisons with satellite data in fact showed that simulations using the AMMABB inventory reproduced the magnitude of AOD better (Tummon et al., 2010).Finally, this shift, also seen when comparing simulated AOD to MODIS satellite observations (http://modis.gsfc.nasa.gov/)(not shown here), could be due to errors in satellite retrievals as well.The next generation of biomass burning inventories using both burnt areas, fire radiance energy and dynamical vegetation models will hopefully contribute to understanding such complex issue (Wooster et al., 2002, 2005, Ichoku and Kaufman, 2005).

Conclusion
We developed a daily African biomass burning inventory for the 2000-2007 period at 1 km×1 km horizontal spatial resolution, for gaseous and particulate species.This inventory was then compared to other existing inventories.A factor of ∼2.4 of differences between GFED2 and AMMABB inventories may be found, which is reasonable, considering the large uncertainties inherent in the determination of biomass burning emissions (Stroppiana et al., 2010).The inventory was also tested within the ORISAM-TM4 global model for BC particle emissions.Comparisons were made with surface, column-integrated and satellite observations obtained during the AMMA campaign.Major features appear to be well reproduced by the global model when the updated biomass burning emissions are included.Seasonal variability of modeled BC concentrations is well simulated at the three AMMA super-sites, Djougou, Lamto and Banizoumbou.Satisfactory comparisons are also obtained with surface measurements of scattering coefficients, sunphotometer AOD and column-integrated SSA values at the Djougou and Banizoumbou sites.Major problems appear to be related to coarse model grid resolution, which does not capture local intense biomass burning events, and to dust modeling, which still requires further development.Similar underestimations (Malavelle et al., 2009) are also found using RegCM3 regional model implemented with the same dust scheme as ORISAM-TM4.Comparison with PARASOL aerosol optical depth data shows good agreement in terms of AOD magnitudes, though maximum AOD are shifted too far eastwards in January and too far southwards in July.These results are in favor of the existence of strong additional particulate source along the Guinean gulf coast, possibly related to developing megacities, industrial and oil activities there, this source not yet being fully accounted for in current inventories.Finally, AMMABB emission inventories appear to be systematically higher than GFED ones.As already noted by different papers, such a difference in source is required to improve the agreement between observations and models.This is the case in West Africa with HADGEM2 model (Johnson et al., 2008) and with RegCM3 model (Tummon et al., 2010) for Southern Africa.Similar improvement has been also reported in AMMA model inter-comparisons for the gas components of the AMMABB product (Barret et al., 2010).

Figure 1
Figure1shows total emissions of black carbon particles from African biomass burning for December 2005 through February 2006 and June through August 2006.BC emissions are maximum in December in the northern hemisphere while they peak in July in the southern hemisphere.Minimum BC emissions occur in March, April and May, despite persistent fires in North-East and Southern Africa (not shown here).In the Northern hemisphere, maximum BC emissions are located in Central Africa, although some hot spots can be found in West Africa, with higher local intensities in 2005 than in 2006.The contrary is indicated at the regional scale, with higher BC emissions in 2006 than in 2005, as shown in Fig.2a, which displays the interannual variability of African BC emissions averaged for December, January and February (DJF) from 2000 to 2007.Interestingly, BC emissions

Table 1 .
Correspondence between BDv, BEv and EF(BC) in each GLC vegetation class.

Table 2 .
Werf et al., 2006Werf et al., , 2008) )udgets for BC and OC in Africa and West and Central Africa as given by this inventory and by GFED 2 inventory (Van derWerf et al., 2006Werf et al., , 2008) )http://www.ess.uci.edu/∼ jranders/.Table 2 summarizes annual African and West/Central Africa BC and OC biomass burning emissions for the year 2006.Data from the GFED2 inventory (Van der Monthly CO budget for different emission inventories in Africa (year 2003).
Fig. 3. Monthly CO budget for different emission inventories in Africa (year 2003).
is based on the coarse resolution TM4 model (http://www.