Atmospheric nitrogen budget in Sahelian dry savannas

The atmospheric nitrogen budget depends on emission and deposition fluxes both as reduced and oxidized nitrogen compounds. In this study, a first attempt at estimating the Sahel nitrogen budget for the year 2006 is made, through measurements and simulations at three stations from the IDAF network situated in dry savanna ecosystems. Dry deposition fluxes are estimated from measurements of NO 2, HNO3 and NH3 gaseous concentrations and from simulated dry deposition velocities, and wet deposition fluxes are calculated from NH+4 and NO − 3 concentrations in samples of rain. Emission fluxes are estimated including biogenic emission of NO from soils (an Artificial Neural Network module has been inserted into the ISBA-SURFEX surface model), emission of NOx and NH3 from domestic fires and biomass burning, and volatilization of NH3 from animal excreta. Uncertainties are calculated for each contribution of the budget. This study uses original and unique data from remote and hardly-ever-explored regions.The monthly evolution of oxidized N compounds shows that emission and deposition increase at the beginning of the rainy season because of large emissions of biogenic NO (pulse events). Emission of oxidized compounds is dominated by biogenic emission from soils (domestic fires and biomass burning of oxidized compounds account for 0 to 13% at the most at the annual scale, depending on the station), whereas emission of NH 3 is dominated by the process of volatilization from soils. At the annual scale, the average gaseous dry deposition accounts Correspondence to: C. Delon (claire.delon@aero.obs-mip.fr) for 47% of the total estimated deposition flux, for both oxidized and reduced compounds. The average estimated wet plus dry deposition flux in dry savanna ecosystems is 7.5±1.8 kgNha−1yr−1, with approximately 30% attributed to oxidized compounds, and the rest attributed to NH x. The average estimated emission flux ranges from 8.4( ±3.8) to 12.4(±5.9) kgNha−1yr−1, dominated by NH3 volatilization (72–82%) and biogenic emission from soils (11–17%), depending on the applied volatilization rate of NH 3. While larger, emission fluxes are on the same order of magnitude as deposition fluxes. The main uncertainties are linked to the NH3 emission from volatilization. When scaled up from the 3 measurement sites to the Sahelian region (12 ◦ N:18 N, 15 W:10 E), the estimated total emission ranges from 2( ±0.9) to 3(±1.4) TgNyr−1, depending on the applied volatilization rate of NH 3 and estimated total deposition is 1.8( ±0.4) TgNyr−1. The dry savanna ecosystems of the Sahel contribute around 2% to the global (biogenic + anthropogenic) nitrogen budget.


Introduction
Nitrogen is a key compound both as a nutrient for plants and animals and as an atmospheric pollutant.In the atmosphere, several nitrogen trace compounds are present, such as NO, NO 2 , HNO 3 , N 2 O and NH 3 , as well as particulate and aqueous forms such as NO − 3 and NH + 4 .At the global scale, the reactive N cycle has been widely impacted by human activities, notably for food production.Indeed, the creation of reactive nitrogen has increased by 120% since 1970 and reached 187 TgNyr −1 in 2005, which is a consequence of the increase of the world population (Galloway et al., 2008).Anthropogenic emissions of NO x are better quantified than natural emissions, and NH 3 emissions from all sources and at all scales remain largely uncertain (Sutton et al., 2007).Nitrogen emissions from the soil contribute to the formation of the above mentioned gaseous products, whereas enrichment of nitrogen in the soil will take place through biological nitrogen fixation, nitrogen wet and dry deposition or nitrogen fertilization (Vlek, 1981).Reactive Nitrogen emissions from soils are influenced by several environmental and physical parameters, such as soil temperature and moisture, soil pH, texture, wind speed, plant cover, floristic composition (e.g.legumes) and N input (fertilization) (Williams et al., 1992;Yienger and Levy, 1995 and references therein;Potter et al., 1996;Bouwman et al., 2002b;Meixner and Yang, 2004;Delon et al., 2007).In semi arid and arid regions, limited water resources will have significant consequences on nitrogen cycling in the soil and the atmosphere.The maximum emission will occur for an optimal value of the Water Filled Pore Space (WFPS).Above (and below) this optimal value (depending on the type of soil and climate), the emission will decrease.The seasonal rainfall distribution leads to an accumulation of N in soils during the dry season, and to large pulses of N emission at the beginning of the rainy season (Austin et al. (2004); Jaeglé et al. (2004)).Resulting emissions from these pulse events release high quantities of NO x in the atmosphere, contributing to increased ozone formation in the troposphere and long range transport of ozone.In Sahelian regions, (the part of West Africa defined as (12 • N:18 • N, 15 • W:10 • E) nitrogen emissions may come from different sources throughout the year.In this study, we will focus on simulated biogenic soil emissions of NO x , calculated biomass burning and domestic fuel emissions of NO x and NH 3 , and calculated volatilization of NH 3 from animal manure.Neither industrial N sources nor N emissions from synthetic fertilization are taken into account because of the remote location of the Sahel from big cities or industrial centres, and because synthetic fertilizers are either used sparingly or not at all in this region of the world.Similar to emission, wet and dry deposition play an essential role in determining the concentration of nitrogen compounds in the atmosphere, and the nitrogen input to the soil/plant system.
The IDAF (IGAC/DEBITS/AFrica) programme started in 1995 with the establishment of 10 measurement sites representative of major African ecosystems.The objectives of the programme are to study dry and wet deposition of important trace species and more generally the biogeochemical cycles of key nutrients.In this way, the IDAF activity is based on high quality measurements of atmospheric chemical data (gaseous, precipitation and aerosol chemical composition) on the basis of a multi-year monitoring.Since 2005, the ORE (Environmental Research Observatory) IDAF has been part of the AMMA (African Monsoon Mutidisciplinary Analyses) EOP (Extensive Observation period) and Banizoumbou Agoufou Katibougou kgN.ha -1 .yr - kgN.ha -1 .yr - 7 6.5 6 5.5 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 7 6.5 6 5.5 5 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 The IDAF network and the resulting scientific research have been presented in several papers, such as Galy-Lacaux and Modi (1998); Galy-Lacaux et al. (2001, 2009); Yoboué et al. (2005); Adon et al. (2010).The objective of this study is the calculation of the balance between N compounds emission and deposition, in order to quantify the atmospheric nitrogen budget in dry savannas.A focus will be made in 3 study sites which are representative of rural semi-arid savannas: Banizoumbou (Niger,13.33• N, 2.41 • E), Katibougou (Mali, 12.5 • N, 7.3 • W) and Agoufou (Mali, 15.3 • N, 1.5 • W), for the year 2006 (see Fig. 1).
In order to calculate N compound emission and deposition, and to quantify the distribution of different emission sources, methods of calculation of emission and deposition fluxes will be presented.The assumptions made to estimate each contribution will be explained, as well as all sources of uncertainties.
In the following, the budget will be calculated first for oxygenated N compounds (NO x for emissions, NO 2 , HNO 3 and NO − 3 for deposition) at the monthly and annual scale for the year 2006, and second for NH x products (NH 3 , NH + 4 ) at the annual time scales.Not all the nitrogen gaseaous compounds are taken into account in this budget, but the main species are represented and a nitrogen compounds budget will be calculated at the 3 IDAF stations, and scaled up to the Sahelian regional scale (12 • N:18 • N, 15 • W:10 • E).

Deposition
The IDAF network is the French contribution to the international IGAC (International Global Atmospheric Chemistry)/DEBITS (Deposition of Biogeochemically Important Trace Species) program.The DEBITS committee has defined a set of experimental and analytical protocols to have comparable measurements from all the DEBITS stations.The IDAF network has adopted these protocols as well.Further information on the IDAF network may be found at http://medias.obs-mip.fr/idaf/.
Continuous measurements of N compounds concentration in air and rain have been performed in Banizoumbou (13.3To estimate atmospheric nitrogen deposition, including both wet and dry processes, we compiled the IDAF nitrogen data from gas and rain chemistry measurements.This study will give the relative contribution of dry and wet deposition processes to the total nitrogen deposition in 2006.

Dry deposition estimate
Gaseous measurements (NH 3 , HNO 3 , NO 2 ) are monthly integrated samples using passive sampling techniques following the work of Ferm et al. (1994).This technique has been tested in different tropical and subtropical regions (Ferm and Rodhe, 1997;Carmichael et al., 2003;Martins et al., 2007).All samples of the IDAF west-central African stations are brought to the Laboratory of Aerology (LA) in Toulouse, France, for ionic chromatography analysis.All details and performances of IDAF passive samplers are given in Martins et al. (2007).
At each IDAF station, passive samplers have been exposed monthly in duplicate, at 1.5 m height.To give an indication of the precision of this sampling technique, the covariance of all duplicate samples were obtained for a ten year period (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007) and found to be 20%, 9.8%, 14.3% for HNO 3 , NO 2 and NH 3 , respectively.The covariances compare well with those reported by Martins et al. (2007), at the Southern African sites, for the same pollutants (20%, 8.3%, 15.3%, for HNO 3 , NO 2 and NH 3 , respectively).The average of the duplicate samplers was used in all cases except when contamination of one of the samplers was suspected (this happened in less than 5% of all data).The detection limit of the passive samplers was evaluated with the variations in the blank amounts of the impregnated filters and found to be 0.05(±0.03)ppb for HNO 3 , 0.2(±0.1)ppb for NO 2 and 1(±0.5)ppb for NH 3 , respectively.For the HNO 3 , NO 2 and NH 3 measurements, 12%, 4% and 0% were below the detection limit, respectively.Adon et al. (2010) presents all the IDAF database from 1998 to 2007 and gives all the details on the samplers, the quality checking and the associated uncertainties.In this study, we used the uncertainty on gas concentrations calculated from the covariance of all duplicate samples.The uncertainties on HNO 3 ,NO 2 and NH 3 concentrations are equal to 20%, 9.8% and 14.3%, respectively.However, it has to be noted that using the covariance for the estimated uncertainty of the monthly passive sample measurements does not account for the inaccuracy of these measurements.As such, the uncertainties quoted here are lower than actual.
The monthly estimated dry deposition flux (generally expressed in units of molecules cm −2 s −1 , and converted to kgNha −1 yr −1 here) is computed using the inferential method, which is defined as taking the product of the measured gas concentration in the air (molecules cm −3 ) and the corresponding deposition velocity in cm/s.The deposition velocity is estimated for each site and species.This method has been widely used in other studies for different types of ecosystems, as in Shen et al. (2009), Pineda and Venegas (2009), Jin et al. (2006), Zhang et al. (2005) and Zhang et al. (2009).The estimate of the gaseous dry nitrogen deposition has been calculated as the sum of the dry deposition flux estimates of ammonia (NH 3 ), nitric acid (HNO 3 ) and nitrogen oxide (NO 2 ).
Dry deposition velocities have been estimated using the Soil Vegetation ATmosphere (SVAT) big-leaf model ISBA (Interactions between Soil Biosphere and Atmosphere, Noilhan and Mahfouf (1996), presented below).In the model the general resistance parameterization for dry deposition velocities of Wesely and Hicks (1977) has been introduced.The surface resistance incorporates both the physical and biological surface characteristics together with the solubility of deposited species (Baer and Nester, 1992).For vegetated surfaces (Wesely, 1989), one further considers the relative contributions of stomata, mesophyllic tissues, and cuticle whereas for liquid surfaces, the parameterization of Erisman and Baldocchi (1994) is used.For the cuticle and soil resistances, the parameterization developed Zhang et al. (2003) has been used.These parameterizations have been included in ISBA and coupled with the 255 surface classification types.ISBA calculates such evolving parameters as aerodynamical, quasi-laminar, stomatal resistances, and drag coefficients for different vegetation types.Chemical dry deposition velocities evolve at each time step together with surface wind, turbulent conditions and chemical parameters.Meteorological conditions are provided by the forcing, derived from satellite data, and developed in ALMIP (AMMA Land surface Model Intercomparison Project), described below.Validation of surface temperature and moisture have been made within the ALMIP project (Boone et al. (2009), DeRosnay et al. (2009)).The forcing in ALMIP is available at 10 m, whereas the concentrations of gases are measured at 1.5 m.A logarithmic decrease of the wind forcing from 10 to 1.5 m, depending on the rugosity, has been applied, to calculate deposition velocities at 1.5 m.
We have therefore calculated monthly means (from 3hourly values) of deposition velocities for NO 2 , HNO 3 and NH 3 , in order to reproduce the seasonal cycle of the deposition process.The monthly means deposition velocities range from 0.1 to 0.33 cm/s for NO 2 , from 0.34 to 0.61 cm/s for HNO 3 and from 0.11 to 0.39 cm/s for NH 3 , where the higher values are found during the wet season.These values are in the lower range of usual deposition velocities found in the literature.The underestimation is larger for HNO 3 and NH 3 than for NO 2 , for the same meteorological forcing.This is due to the fact that wind speed is low in the forcing (2.5, 2.4 and 1.9 m/s as a yearly average in Agoufou, Banizoumbou and Katibougou respectively).Low wind speed involves high aerodynamic resistance (Ra).Large Ra will have a strong influence on HNO 3 deposition velocity, because canopy resistance for this specie is negligible, a lower influence for NH 3 and a small influence for NO 2 .These processes have been comprehensively described in Hanson and Lindberg (1991), giving different deposition velocities depending on the wind speed at a same site, Duyzer and Fowler (1994), and more recently, Zhang et al. (2009).The main uncertainty of the deposition velocities is linked to the wind forcing, and to the type of soil represented in the model.The type of soil is not considered directly in this calculation, but indirectly via the roughness length.The difference between the wind in the forcing and the wind measured in situ is between 5 and 35% depending on the site.A mean rate of 20% is applied for the wind speed uncertainty, leading to an uncertainty rate of 15% for NH 3 , 13% for NO 2 and 19% for HNO 3 deposition velocities.
The authors are however aware that the inferential method is not the best one to evaluate NO 2 and NH 3 fluxes.Exchange of NH 3 appears to behave as if a compensation point exists: NH 3 is both emitted and deposited.The direction of the net flux will depend on the strength of the volatilization process and environmental conditions.The NH 3 compensa-tion point has been widely studied for temperate climate vegetation (Sutton et al. (2007) and references therein), Trebs et al. (2006) have also described a detailed N budget in a tropical pasture site in Brazil, but the lack of micrometeorological data in the Sahel prevents from parameterizing and estimating, for example, stomatal emissions, cuticular desorption emissions, cuticular recapture of stomatal emissions (Sutton et al., 1998), key points for the compensation point concept.As a consequence, the emission and deposition fluxes of NH 3 are estimated individually in this study.Concerning NO 2 fluxes, chemical reactions leading to the rapid production or consumption of NO 2 (ozone photochemistry) may induce errors in the calculation of dry deposition fluxes of NO 2 .A lot of studies have stated the limits of the inferential method for NO 2 fluxes (as examples, Kramm et al., 1995, De Arellano and Duynkerke, 1992, Kristensen et al., 2010), and use detailed micrometeorological measurements and parameterizations to calculate the fluxes.According to De Arellano and Duynkerke (1992), if turbulence is parameterized in terms of an exchange coefficient, the chemical reactions may induce a 35% difference between inert and reactive exchange coefficients.As a consequence, an additional uncertainty rate of 35% is added to the NO 2 flux.
In this study, concentrations are measured with passive samplers (indirect method), and are integrated monthly.Although preferable, no micrometeorological measurements are available, because not affordable or not practical in such remote regions.While highly uncertain, the NO 2 flux is estimated with the inferential method (which gives an idea of its contribution in the N budget), keeping in mind that it has to be validated with micrometeorological measurements in future field campaigns.As we use monthly means for concentrations and deposition velocities, the covariance between the two may induce an additional uncertainty (the missing covariance term), especially for species having strong diurnal variations, like NO 2 (Matt and Meyers, 1993).This missing covariance term is reduced for species like NH 3 and HNO 3 .A missing covariance uncertainty rate of 20% has been applied for NO 2 , and 10% for NH 3 and HNO 3 .
Particulate N dry deposition (pNH + 4 and pNO − 3 ) is not taken into account in this budget.
The comparison of these concentrations with NH 3 concentrations in both sites (2.9 to 10.4 ppb in Banizoumbou, 1.8 to 6.9 ppb in Katibougou, and 3.5 to 10.5 ppb in Agoufou) leads to the conclusion that particulate deposition is negligible.As far as the authors know, no other particulate dry deposition measurements have been made in the remote regions of Sahel.
The total uncertainty applied to the fluxes is linked to the concentration measurements, the deposition velocity, the missing covariance (plus a term of flux divergence for NO 2 ).The total rate of uncertainty applied for deposition fluxes is 70% for NO 2 , 31% for NH 3 and 38% for HNO 3 .

Wet deposition
An automatic precipitation collector specially designed for the IDAF network has been installed at all stations.A local operator collects water from each rainfall event in a Greiner tube (50 ml).Preserving the rainwater samples from contamination is an important issue since microbial input could modify its chemical composition.Samples are refrigerated at 4 • C and preserved with 15 mg of thymol biocide or stored in a deep freeze environment.Ion Chromatography (IC) procedures are given in Galy-Lacaux and Modi (1998).The uncertainty of annual and monthly rainfall depth measured with standard rain gauges is about 10%.Precipitation coverage length (PCL%) is about 100% for the studied sites in 2006.The percent Total Precipitation (TP%) that represents the completeness of precipitation depth associated with valid chemical analysis is around 80-100% for all the sites.Chemical composition of rain is determined using Ionic Chromatography.The laboratory of Aerologie participate since 1996 to the international intercomparison program organized annually by WMO (World Meteorological Organization, 1997).According to prior results and through intercomparison tests organized by WMO, analytical precision is estimated to be 5% or better for all ions, within the uncertainties on all measured values presented here.The concentrations are given in µeqL −1 and we assume that the first digit after the comma is significant.Combining all the uncertainties of measurements and calculations, the uncertainty of the wet deposition fluxes is estimated to be about 10%.
To calculate wet nitrogen deposition in African dry savannas for the year 2006, we have compiled the annual Volume Weighed Mean (VWM) concentrations of nitrate and ammonium from the precipitation collected at the 3 IDAF stations.The computation of nitrate and ammonium wet deposition has been done according to the 2006 annual rainfall for the studied sites.The rainfall depths registered at Banizoumbou (Niger), Katibougou and Agoufou (Mali) are 505.2mm, 588.2 mm and 374 mm respectively.Nitrate and ammonium concentrations at the dry savannas sites are in the upper range of all the African ecosystems, with values from 37 (in Banizoumbou) to 31 (in Katibougou) µeqL −1 of NH + 4 and a range of 7 (Banizoumbou) to 16 (Katibougou) µeqL −1 of NO − 3 (Sigha et al., 2003;Yoboué et al., 2005;Dentener et al., 2006).
In the Sahel region, the major source of precipitation nitrate content comes from natural NO x emissions from soils, whereas NO x production from lightning plays a minor role for wet deposition (Tost et al., 2007), and it is not taken into account in this budget.Nitrate concentration in precipitation represents the final result of homogeneous and heterogeneous processes of nitrogenous gases and particles in the atmosphere and cloud water.Nitric oxide (NO) is the major nitrogen compound released from savanna soils in the nonburning season (Serc ¸a et al., 1998), and a large fraction of NO produced is oxidized in the atmosphere through photochemical reactions, into HNO 3 or organic nitrates.HNO 3 , which is extremely soluble in water, is thus easily scavenged by clouds.Galy-Lacaux et al. (2001) have demonstrated that heterogenous processes between nitric acid and mineral particles are present all over the African continent.
Ammonium content in precipitation results from inclusion of gaseous ammonia and particles containing NH + 4 in precipitating clouds.The highest values of ammonium compounds in precipitation registered in the semi-arid regions have been attributed to strong sources of ammonia from domestic and pastoral animals during the wet season (Galy-Lacaux and Modi, 1998).The concentration of dissolved organic nitrogen (DON) in rain is not measured within the IDAF network, and it remains rather difficult to quantify the part of DON in the total wet deposition budget.A study performed in North Carolina from Whitall et al. (2002) states that DON could represent a third of the total wet deposition budget, whereas Hill et al. (2005) found that DON represents 18% of the wet deposition in an open field in Michigan.The budget presented in this study is therefore underestimated, in a proportion that is difficult to asses, but could represent 20% of the wet deposition.

NO biogenic emission from soils: model characteristics
Biogenic emissions from soils are derived from an Artificial Neural Network (ANN) approach.The resulting algorithm provides on line biogenic NO emissions and is developed in Delon et al. (2007).It is fully coupled to the SVAT model ISBA, and has been previously tested in the 3-D coupled chemistry-dynamics model MesoNH-C (which uses surface scheme ISBA) to reproduce NO pulses after a rain event in Niger (Delon et al., 2008).NO emissions from soils in ISBA are obtained for the year 2006 at a spatial resolution of 0.5 • and a time resolution of 3 h.In the following sections, NO fluxes are averaged on a 3 • /3 • , window around each specific station.The simulated domain extends from 5 • S to 20 • N in latitude, and from 20 • W to 30 • E in longitude, but only the Sahel region (15 • W:10 • E, 12 • N:18 • N) will be explored in this study.The meteorological forcing, developed within ALMIP, is obtained from a data set based on the merging of ECMWF (European Centre of Medium Range Weather Forecast) atmospheric state variables, and TRMM-3B42 3-hourly data for the precipitation.LAND-SAF data is used for downwelling longwave and shortwave radiative fluxes (Geiger et al., 2008;Huffman et al., 2007).A more comprehensive description of ALMIP may be found in Boone et al. (2009).
The NO flux from soil is a non linear function of seven soil surface parameters: surface WFPS, surface and deep soil temperatures, pH, sand percentage, fertilisation rate and wind speed.Surface WFPS is deduced from the simulated soil surface moisture.Surface and deep soil temperatures are also calculated by the model.Wind speed is provided by the atmospheric forcing.The pH map is obtained from IGBP-DIS (1998, http://www.sage.wisc.edu/atlas)soil data base, at a spatial resolution of 0.5 • /0.5 • .Sand percentage is obtained from the ECOCLIMAP data base (Masson et al., 2003), at a resolution of 1 km/1 km.We assume that the most important driving parameters for our study are surface soil temperature and WFPS (calculated from soil moisture).Therefore, the uncertainty of the biogenic NO flux is calculated from the derived of the non linear function regarding soil surface temperature and moisture in ISBA (we approximate that the flux fluctuation in our study is mainly due to the fluctuation of these two parameters).The uncertainty of the NO flux from soil is calculated month by month and site by site.It ranges from 3 to 45% (18% in annual mean) in Agoufou, from 5 to 68% (25% in annual mean) in Banizoumbou, and from 8 to 56% (20% in annual mean) in Katibougou, depending on the month.
The fertilization rate provided to the model is based on the calculation of N released by organic fertilization (i.e.cattle dung), for each country.Indeed, data bases of land use do not provide enough information in the Sahel concerning synthetic fertilization.Galloway et al. (2004) give an estimate of 2.5 TgNyr −1 for the N input by fertilization production for the whole African continent, and Bouwman et al. (1997) show no emission from synthetic fertilizer use in the Sahelian band (12-18 • N).Schlecht and Hiernaux (2004) state that manure from livestock is an important source of organic matter and nutrients due to limited access to mineral fertilizers.These low assessments corroborate the assumption that the use of synthetic fertilizers in the Sahel is not a common feature, and can be neglected in our inventory.
Therefore, significant work was done to develop our own data base for organic fertilization for 23 countries of West and Central Africa which are contained in our simulation domain.These countries are, in alphabetical order: Benin, Burkina Faso, Cameroon, Chad, Congo, Democratic Republic of Congo, Equatorial Guinea, Gabon, Gambia, Ghana, Guinea Bissau, Guinea, Ivory Coast, Liberia, Mali, Mauritania, Niger, Nigeria, Republic of Centre Africa, Senegal, Sierra Leone, Sudan, and Togo.
The N quantity released by livestock is calculated from Schlecht et al. (1998), in gNhead −1 day −1 , for cows, sheep and goats.This estimate is multiplied by the number of animals per km 2 in each country.The animal population is obtained in each region of each country from the FAO (Food and Agriculture Organization) Global Livestock Production and Health Atlas GLiPHA (2009, http://kids.fao.org/glipha/).The resulting inventory of N is introduced in the ANN algorithm as N input in kgNha −1 h −1 .The data needed to build this inventory are, however, a source of uncertainty: the N input calculated from Schlecht et al. (1998), which gives 45 kgNanimal −1 yr −1 in a small region of Niger, remains low when generalized to the West African region.Indeed, Mosier et al. (1998) gives an estimate of 170 kgNanimal −1 yr −1 for the whole African continent, and Bouwman and Van Der Hoek (1997) gives an estimate of 130 kgNanimal −1 yr −1 in developing countries.These differences may be explained first because poultry, camels and donkey excreta are not taken into account in Schlecht et al. (1998) study, and second because this study is limited to a small region in Niger where the acidic sandy soils, which are predominant in the region, are inherently poor in N (Schlecht et al., 2004).Compared to the West African region, they do not provide a representative uptake of N in food for grazing.Therefore, we have taken an intermediate value of 70 kgNanimal −1 yr −1 for cows and 20 kgNanimal −1 yr −1 for sheep and goats for our study domain, which yields a total N input of 23 kgNha −1 yr −1 in Agoufou (Mali), 25 kgNha −1 yr −1 in Banizoumbou (Niger) and 11 kgNha −1 yr −1 in Katibougou (Mali).
30 to 50% this N input will be used for the calculation of NH 3 volatilization, the rest is used as input for the calculation of NO biogenic emissions by the ANN algorithm in the ISBA surface model (the justification for this distribution is explained below, Sect.2.2.2).The uncertainty concerning this part of the budget is one of the most difficult to calculate.It is linked to the animal population from GliPHa, certainly the more recent estimate available.Bouwman et al. (1997) have stated in their global inventory that the overall uncertainty caused by the animal population is probably inferior to 10%.The same rate is applied in our inventory.The uncertainty depends also on the N release by livestock.The overall uncertainty for this source is estimated to be 50% from the same author.As a consequence, the uncertainty applied on NH 3 volatilization is 51% in our study.
Figure 1 shows the NO biogenic flux from soils in the simulation domain in terms of the annual mean in Fig. 1a and JJAS mean (June July August September) in Fig. 1b.Indeed, JJAS corresponds to the wet season in West Africa, when the monsoon flux (cool and wet air coming from the ocean) reaches its northernmost position bringing humid air and intense precipitation in the Sahel.As mentioned in Yienger and Levy (1995), Jaeglé et al. (2004), Butterbach-Bahl et al. (2004) and references cited therein, NO emission is principally driven by soil moisture in tropical regions, and the most intense emissions occur in the Sahel when the first rains fall on the very dry soils.As shown in Fig. 1b, soil emissions in the Sahel reach their strongest values during the rainy season from June to September.In this study, we will only focus on the Sahelian region (12 • N:18 • N, 15 • W:10 • E) and the magnitude of fluxes in wet savanna or forests areas or other tropical ecosystems will not be discussed here.

NH 3 emission by volatilization
Ammonia is formed in soils from biological degradation of organic compounds and ammonium, as represented by the equation 1: Where (aq) stands for aqueous, (g) for gaseous and (atm) for atmospheric loss.NH + 4 (aq) depends on soil cation exchange reactions, soil moisture content and net mineralization.The concentration of NH 3 (aq) depends on ammoniacal N, temperature and pH.The rate of gas dispersion in the natural environment depends on the temperature gradient, and wind speed (Vlek, 1981).After NH 3 is emitted to the atmosphere, each nitrogen atom can participate in a sequence of effects, known as the nitrogen cascade in which a molecule of NH 3 can, in sequence, impact atmospheric visibility, soil acidity, forest productivity, terrestrial ecosystem biodiversity, stream acidity, coastal productivity (Galloway and Cowling, 2002), soil acidification and eutrophication (Bouwman and Van Der Hoek, 1997).
The current annual NH 3 emission in developing countries of 15 TgN accounts for 2/3 of the global emissions from animal excreta.The fraction of the N excretion that is lost as NH 3 ranges from 10 to 36%, depending on animal-waste management and animal category (Bouwman and Van Der Hoek, 1997;Bouwman et al., 1997).More recently, Bouwman et al. (2002a) stated that 25, 28 and 33% of N use in Western Africa is released as NH 3 in intensive grasslands, upland crops and wetland rice, respectively.The difficulty of obtaining reliable data concerning the use of animal manure, and management practices in tropical countries leads to uncertainties in estimating the NH 3 loss, particularly in semiarid regions like the Sahel.However, one can suppose that better conditions are encountered in the Sahel region which favours NH 3 volatilization, such as high temperatures, low soil moisture and bare soil surfaces.As a consequence, two different loss rates (30 and 50%) have been applied to the input of N by animal manure previously prescribed for the calculation of NO emissions.This leads to a N-NH 3 volatilization estimated between 7.2±3.7 and 12.4±6.3kgNha −1 yr −1 in Agoufou, 8.1±4.2 and 12.8±6.6kgNha −1 yr −1 in Banizoumbou, and between 3.9±2 and 6.2±3.2 kgNha −1 yr −1 in Katibougou, reported in Table 2. Bouwman et al. (1997) estimate the emission from domesticated animals in the Sahelian band between 0.5 and 5 kgNha −1 yr −1 for the year 1990.

NO x and NH 3 emission from biomass burning
Global biomass burning inventories for gases and particles are available from January 2005 to December 2006 on the Laboratoire d'Aérologie website (http://www.aero.obs-mip.fr:8001/).The available spatial resolution is 1 km-1 km, and the time scale is daily.These global inventories use the L3JRC burnt area product (Liousse et al., 2010) based on the SPOT-VGT vegetation satellite and Global Land Cover (GLC) vegetation map, together with data on biomass densities and burning efficiencies.Such data, which were not available for the GLC2000 vegetation types but for the UMD (Hansen et al., 2000) global land cover product (Michel et al., 2005) have been established from correspondences between the 13 UMD vegetation classes and GLC2000 vegetation classes (Liousse et al. (2010), Mieville et al. (2010)).Emission factors for gaseous species were chosen following Andreae and Merlet (2001).In the present study, we used the inventories for NO, NO 2 and NH 3 species in the region of interest (−5 to 20 • N, −20 to 30 • E).Monthly means and averages in a 5 • /5 • window around each specific station (to ensure a sufficient sampling of fire events) are used to evaluate the potential impact of biomass fires at the local/regional scale.
Long range transport of nitrogen emissions from remote fires over southern Africa are not considered here due to the northward position of the stations in dry savanna areas.Indeed, Mari et al. (2008) and Sauvage et al. (2007) have shown that the impact of fire emissions occurring in the southern hemisphere from June to September was limited to coastal regions in the gulf of Guinea, and did not influence sites to the north.

NO x and NH 3 emission from domestic fires
Combustion of biofuel is mainly used for cooking in the Sahel.Biofuel use provides a constant emission all year long and is a potential source of trace gases.Ludwig et al. (2003) (and references therein) show that 0.5 TgNO/yr is emitted over the African continent due to biofuel consumption.In the present study, NO x and NH 3 emission from domestic fires uses the methodology from Junker and Liousse (2008), for the most recent existing year (2003).Due to a lack of knowledge of the emission factor as a function of countries, fuels, activity, technology/norm combinations, this methodology is based on a lumping procedure to account main factors of vulnerability.Nine emission factors are defined for each fuel category, and 3 main activities are considered (traffic, domestic, industrial), with 3 levels of technology linked to each countryś development level (developed, semi-developed, developing).Reference NO x emission factors values were given by Andreae and Merlet (2001), and on the 2 following websites: http://www.naei.org.uk/reports.php,and http: //www.transport.govt.nz/.
NH 3 emissions factors are also given in Andreae and Merlet (2001).Consumption data are given by the United Nations database.Annual emissions are calculated country by country, and then gridded at 25 km/25 km resolution (Assamoi and Liousse, 2010).The monthly input of nitrogen compounds is therefore constant all year long, and is averaged over a 5 • /5 • window around each specific station to consider the local/regional impact of these emissions.
Uncertainties are mainly linked to wood and charcoal consumption estimates (50%, citetAssamoi09) and emission factors (31% Andreae and Merlet (2001).The total uncertainty applied to domestic fires is 60% for both compounds.

Results and discussion
As mentioned in Sect.2.1, deposition fluxes (dry + wet) are estimated from concentration measurements at the IDAF stations.All deposition fluxes are converted in kgNha −1 yr −1 and compared to emission fluxes in the same unit.Oxidized and reduced N compounds will be first treated separately.Then, a total budget will be estimated at the annual scale.

Monthly evolution of Nitrogen oxidized compounds
Nitric oxide (NO) and nitrogen dioxide (NO 2 ) react rapidly in the atmosphere and are referred to jointly as NO x .In turn, NO x can be incorporated into organic compounds such as peroxyacetyl nitrate (PAN) or alkyl nitrate, or further oxidized to HNO 3 .Gas-phase HNO 3 can be converted to aerosol nitrate (NO − 3 ) (e.g., by reaction with ammonia).PAN can convert back into NO 2 , and in hot temperature conditions its lifetime is short (few hours), so concentrations may remain low despite abundant photochemical radicals that promote PAN formation (Munger et al., 1998).We can therefore reasonably consider that the emission of NO (both biogenic and anthropogenic) at the surface is the beginning of the formation of all other reactive oxidized nitrogen compounds in the atmosphere.Non reactive N compounds (N 2 , N 2 O, N 2 O 5 ) are not included in the budget, because emission fluxes of these compounds in arid regions are negligible.According to Meixner and Yang (2004), N 2 O fluxes from semiarid soils are expected to be low, and those from arid soils are very low, due to the fact that N 2 O emission is favoured by denitrification processes at high soil moisture content.Galbally et al. ( 2008) (and references therein) suggest, however, that 5% of the global soil emissions of N 2 O from agricultural disturbances (i.e.land conversion to agriculture) may occur in semi arid and arid zones.This fraction is considered to be low, hence N 2 O emissions are not considered.Organic compounds such as PAN are also omitted from this budget, assuming that they are converted into NO 2 and that their concentrations are negligible in non polluted areas (around 100 ppt; Seinfeld and Pandis, 1998), compared to statistical values of NO 2 (2.5±0.6 ppb) and HNO 3 (0.5±0.2 ppb) concentrations, based on a 10 year period (Adon et al., 2010).Hence, it has to be mentioned that the three N gases measured and deposited (and thus considered in the budget) do not represent all the N gaseous compounds in the atmosphere.
As a consequence in this study, the total NO x emission flux is defined as the sum of biogenic NO soil emission flux + biomass burning NO x flux + domestic fire NO x flux.It is compared to deposition flux of oxidized components, defined as the sum of estimated dry deposition of NO 2 + HNO 3 in the gas phase, + NO − 3 wet deposition flux.Figure 2 presents the monthly evolution of emission and deposition oxidized N fluxes in Agoufou, Banizoumbou and Katibougou.Mean fluxes for each station at each season Significant deposition fluxes commence at the beginning of the wet season (usually around May in dry savanna areas) due to a pulse emission flux which is produced over very dry soils when the first rains fall (Johansson et al. (1988), Yienger and Levy (1995), Otter et al. (1999), Yan et al. (2005), Delon et al. (2008).Adon et al. (2010) showed that the surface NO 2 concentration has its first maximum at the beginning of the rainy season.Figure 2 shows that emission and deposition fluxes increase simultaneously, but note that the deposition maximum is early compared to the emissions.In the model, the biogenic NO emission is highly correlated with rainfall (i.e. the soil has become wet enough to produce sufficient emissions).Later in the season, the model keeps emitting, which is in contrast to site measurements which show a decrease in concentrations.Two factors may explain this: the emission module emits longer than it should, and sometimes rain events do not compare well between measurements and model (whereas the total annual measured rainfall is comparable with that deduced from the forcing).The latter results because it is difficult to compare point rainfall measurements to a satellite-based product at a 0.5 degree spatial scale, espe-cially because of the mostly convective nature of the rainfall in this region (e.g.Ali et al., 2003).
A second emission maximum is observed in november in Banizoumbou and Katibougou, and corresponds to fire occurence during the dry season, which imply an increase in NO x fluxes.The highest uncertainties represented on the emission flux are found in the dry season (November), when biomass burning emissions are important in Banizoumbou and Katibougou, and also in August, when soil moisture is high, as well as biogenic fluxes.Indeed, the biomass burning flux uncertainty is large, and the NO flux uncertainty is larger when soil moisture is larger.A second maximum in deposition is observed in September in Katibougou, and in October in Banizoumbou, shifted compared to the emission maximum, and less intense.The shift could be due to an overestimate of fire emissions compared to local deposition measurements.Furthermore, HNO 3 deposition fluxes, superior to NO 2 fluxes (because of higher deposition velocities), do not increase when fire emissions increase.Even if emission fluxes are larger than deposition fluxes, they keep similar orders of magnitude both in wet and dry seasons.Considering the high level of uncertainty of the NO 2 flux, the maximum uncertainties in deposition fluxes are found when NO 2 concentrations are large, i.e. during the wet season, and in september and october when fire emissions occur.
The comparison of these fluxes in Fig. 2 at the monthly scale shows a good agreement between emission and deposition magnitude, while underscoring the difficulties of analysing the nitrogen budget in such remote areas where too few measurements are available.Emission modelling presents an alternative to this lack of measurements, but many questions remain of concern, for example, the quantity of nitrogen contained in the soils, or the amount of emissions  after the first rainfall in semi-arid regions.However, the joint exploitation of IDAF measurements and modelling results gives a unique opportunity to provide a nitrogen budget in Sahelian ecosystems.In the following, we will try to evaluate the contribution of biogenic and fire emissions to atmospheric deposition at the annual scale, both for oxidised and reduced nitrogen compounds.It is obvious from these figures that the emission flux is dominated by the biogenic part, both in the wet season and for the annual mean.Mean values and uncertainties for each season are reported in Table 1.Emission fluxes at the annual scale range from 1.0±0.1 to 2.4±0.8kgNha −1 yr −1 .These estimates are in the range of emission fluxes reported by Meixner and Yang (2004), which range from 0.02 to 34 kgNha −1 yr −1 , in semi arid and arid ecosystems.In Bani- In Figs.3b and 4b, the estimated dry deposition appears to be half of the total deposition, except in Agoufou, with the remaining half being wet deposition.This has already been observed in dry savanna stations and other IDAF stations in West African wet savanna (Galy-Lacaux et al., 2003).Of course, the high level of uncertainty of the dry deposition flux of NO 2 could drastically change this repartition, in one direction or the other.

Annual and seasonal budget of N oxidised compounds
Large deposition fluxes are observed in Katibougou and Banizoumbou because of high wet deposition of NO − 3 .Katibougou, which is the southernmost station, receives more rainfall than Banizoumbou and Agoufou, which could explain the high wet deposition flux.But in Banizoumbou, the strong deposition flux in the wet season can be explained by significant biogenic emissions at the beginning of this season, well reproduced by the model.Figures 3a and 4a show that no biomass burning contribution appears in Agoufou, due to the small quantity of biomass available during the dry season.NO 2 concentration measurements do not show any increase in Agoufou in that season (Adon et al., 2010).The quantity of N available in the soil is not part of the systematic measurements in IDAF stations, but measurements in the Gourma region (Diallo and Gjessing, 1999) do not reflect any strong reservoir of N. The release of nitrogen compounds in Agoufou and Banizoumbou specifically (because of the reduced rainfall amount) seems to be very powerful.Preliminary explanation could lie in the N input from N-fixing crusts of cyanobacteria and lichens (Barger et al., 2005), and N-fixing plants and trees (e.g.Acacia).Meixner and Yang (2004) state that despite the fact that deserts have low plant productivity, inorganic nitrogen can accumulate in desert and semi-desert soils during long dry periods.
Generally speaking, the deposition fluxes of oxidised N compounds in dry savanna stations is the consequence of strong biogenic emission fluxes in the wet season.Indeed, the contribution of emissions from biomass burning at the annual scale is only 12% and 5% of the total oxidised N emission flux, respectively, in Banizoumbou and Katibougou, and the contribution of domestic fires is negligible.The contribution of traffic from large cities to dry deposition during the wet season is also neglected, as shown by very low particulate concentrations whatever the season.

Annual and seasonal budget of reduced compounds
The dry and wet deposition of NH 3 and NH + 4 will be referred to as NH x in the following.
The calculation of NH 3 volatilisation has already been developed in Sect.2.2.2.The subsequent NH 3 volatilization flux (30 to 50% of the N input) is reported in Table 2.The assumption that the release is constant during the year is obviously not exact because of seasonal variations in the absorption and release depending on soil humidity, wind speed, and food quality for cattle.Owing to the lack of measurements in remote Sahelian areas, the lack of statistical data on cattle repartition and nitrogen content leads to many assumptions, implying of course a high level of uncertainty in determining the exact quantity of NH x released.But the relative level of volatilization from excreta compared to other sources of NH 3 is maintained, when compared to emissions at the global scale (Bouwman et al., 1997;Galloway and Cowling, 2002).Figure 5 shows the annual repartition of emission and deposition NH x fluxes and the associated uncertainties as error bars.Annual mean values for each station are given in Table 2.The average NH x emission flux ranges from 6.4±3.3 to 10.5±5.4 kgNha −1 yr −1 and is dominated by the volatilization flux, despite the high uncertainty rate applied on it.The mean deposition flux (dominated by dry deposition processes) is 5.2±1.0 kgNha −1 yr −1 .These values are of the same order as deposition fluxes calculated in a tropical pasture site in Brazil, giving a range of NH 3 fluxes between 6 and 12 kgNha −1 yr −1 (Trebs et al., 2006).

Annual budget of oxidized and reduced nitrogen
compounds at each station, and scaled up to the regional scale Figures 6 and 7 show the total deposition and emission fluxes of oxidized and reduced nitrogen compounds at the annual scale for each station.Uncertainties are shown in parenthesis.These figures allow the synthesis between Sects.3.2 and 3.3, and show that the maximum (wet and dry) deposition and emission fluxes are dominated by the NH 3 contribution, with respectively 77%, 68% and 52% at Agoufou, Banizoumbou and Katibougou stations for deposition flux, and respectively 84%, 79% and 63% at Agoufou, Banizoumbou and Katibougou for the emission (for a 30% NH 3 volatilization rate).Whelpdale et al. (1997) have estimated that the contribution of oxidized nitrogen to the total deposition was about 23% in arid savannas, which corroborates our results.
The second most important emission flux is given by the biogenic NO from soils.The contribution of NO 2 dry deposition flux, which presents the highest uncertainty, is only 9% in Agoufou and Banizoumbou, and 12% in Katibougou.An important decrease or increase of this flux (in the range of its uncertainty) will not impact drastically the total budget.The budget could be however highly impacted by the large NH 3 emission uncertainty.
Table 3 summarizes the mean N emission and deposition fluxes at the annual scale for each station.From these 3 sites, the average deposition flux, attributed to dry savanna ecosystems in Sahel, is 7.5 (±1.8) kgNha −1 yr −1 , and the average emission flux ranges from 8.5±3.8 to 12.5±5.9kgNha −1 yr −1 .The ± values given in parenthesis represent the mean uncertainties carried through the entire calculation.Averaging the flux values of the 3 sites is a rather simple method to scale up the N budget, but the lack of data and information around the sites prevented us from using a more elaborated method.The deposition fluxes calculated here are of the same order of magnitude as deposition fluxes in temperate rural site reported by Zhang et al. (2009), ranging from 4.3 to 11 kgNha −1 yr −1 .Deposition fluxes values remain in the same order of magnitude from one station to another.Common characteristics (same climate regime, same type of emission sources and amplitude, same type of vegetation and soil characteristics) deduced from these 3 stations, can be attributed to dry savanna ecosystems, and may also be scaled up to the Sahelian region (12 • N:18 • N, 15 • W:10 • E, i.e. 2.4 × 10 6 km 2 ).Emission characteristics also give common features from one site to another, despite the stronger contribution of NH 3 volatilization in Agoufou, due to the very weak contribution of biomass burning and biofuel fires in this remote area.
According to Mayaux et al. (2003), Aubréville (1949), and White (1986), we can consider that dry savanna (i.e; open grassland and open grassland with sparse shrubs) is representative of the main Sahelian ecosystem.It is therefore possible from these fluxes estimates to give an N annual budget of dry savanna Sahelian ecosystems.However, during the wet season, the portion of Sahel between 12 • N and 13 • N is highly cultivated.This zone will not emit and deposit N-NH 3 the same way as non cultivated and pastoral zones.Therefore, the budget calculated here, without considering a cultivated zone during the wet season, has to be considered as a maximum value.
Figure 8 gives the mean repartition that should be observed in the Sahel concerning nitrogen compound emission and deposition fluxes.Total deposition has been estimated to range between 1.4 and 2.3 TgN/yr (with a mean value of 1.84±0.44,1.3±0.2TgN/yr attributed to reduced compounds, 0.5±0.2TgN/yr attributed to oxidized compounds).Depending on the NH 3 volatilization rate, the emission has been estimated to be between 1.1 and 4.5 TgN/yr, 1.1 being the lower flux estimated with the 30% volatilization rate, and 4.5 being the highest estimate for 50% volatilization rate (for a 30% volatilization rate, the estimated emission is 2.0±0.9,1.5±0.8TgN/yr for reduced compounds and 0.5±0.1 for oxidized compounds), for the Sahel region.
These values have to be compared to global emission and deposition estimates: global NO x emission has been estimated at 44 (30-73) TgN/yr (Galbally et al., 2008), and NH 3 global emission in 1990 has been estimated at 54 (40-70) TgN/yr (Bouwman et al., 1997).Galloway et al. (2008) give a 100 TgN/yr for NH 3 +NO x emission in 1995.Our mean estimate of N compound emissions for the Sahel reaches 2.0 TgN/yr, which is around 2% of the estimate of global NH 3 +NO x emissions cited above, from only 0.48% of the global surface area.Furthermore, Bouwman et al. (1997) have estimated the contribution from animal excreta to NH 3 emissions at 21.7 TgN/yr at the global scale, plus 5.9 TgN/yr due to biomass and biofuel combustion.Our estimate of NH 3 emission from volatilization and combustion would give 84% of 2.0 TgN/yr = 1.7 TgN/yr in the Sahelian region only.These mean estimates give a rough idea of the importance of the N budget in the Sahelian dry savannas.As stated by Davidson and Kingerlee (1997) and Meixner and Yang (2004), these ecosystems constitute a significant part in the emission of Nitrogen compounds, whereas our knowledge of these remote areas is limited by a lack of measurements and studies.

Conclusions
This study is a first and original attempt to estimate both deposition and emission fluxes of a set of nitrogen species in dry savanna ecosystems, using simulated and calculated inventories, and in situ measurements.Deposition fluxes of HNO 3 , NH 3 and NO 2 are calculated according to the inferential method with measured concentrations and simulated deposition velocities, at the monthly scale.Emission fluxes of biogenic NO from soils are simulated with a NN derived  In this study, we have first tried to reproduce the nitrogen oxidized compound emission and deposition evolution month by month at three different IDAF stations in dry savanna areas during 2006.The magnitude of deposition and emission fluxes is similar, but the maximum emission is later  and larger than the maximum deposition during the rainy season.An annual budget of reduced and oxidized N emission and deposition fluxes has been calculated.It gives a mean estimate of 7.5 (±1.8) kgNha −1 yr −1 for total deposition, slightly dominated by wet deposition (53% of the total), and 8.4 (±3.8) kgNha −1 yr −1 for total emission during the year 2006, dominated by NH 3 volatilization (72%) and biogenic emission from soils (17%), whereas emissions from biomass burning and domestic fires accounts for 11% only.If we make the reasonable assumption that dry savanna is representative of the main Sahelian vegetation type, we can scale up those fluxes to the Sahelian region: the estimated emission in the Sahel could be 2.0 (±0.9)TgN/yr, whereas estimated deposition would be 1.8 (±0.4) TgN/yr.The uncertainties are numerous, but they are linked to necessary assumptions considering the small amount of data available in this region.Uncertainties have been calculated for each contribution of the budget, and are linked to the concentration measurements (in gas and rain), to the dry deposition velocities estimated from modeling, to the lack of validation data in the Sahel, to the estimated livestock population in the Sahel, to the ANN algorithm estimate of biogenic NO, to the assumption that deposition fluxes are calculated with the inferential method, and to the fact that only three N compounds are used in the deposition calculation (Organic Nitrogen not taken into account).The highest uncertainties are found for NO 2 dry deposition flux, for the NH 3 volatilization flux and for the fire emissions.However, only the NH 3 volatilization flux uncertainty has a significant impact on the overall budget, and could change significantly the amount of released Nitrogen in the atmosphere.The overall uncertainty in emission is 45%, and 25% in deposition.We have assumed that long range transport of nitrogen compounds did not influence local deposition (except wet deposition) at the three studied stations.If applied to other regions, this assumption has to be carefully re-considered.This work does not claim to be totally exhaustive, but wants to bring a supplementary knowledge for the unknown Sahel region.These values show that dry savanna ecosystems in the Sahel have to be taken into account for the N budget calculation.A lot of work remains to improve this budget however.The improvement of the ANN module for suitability in other West African ecosystems (wet savannas, forests) is foreseen in future research studies, as well as planning field campaigns in the Sahel to validate our deposition flux calculations.
Furthermore, in order to validate the assumptions given in this study, in collaboration with specialized teams, further measurements are needed to improve the biogeochemical description of the soils in arid and semi-arid regions.This is needed to better understand the strong release of N compounds at the beginning of the wet season, despite their low N content, and to improve the quantification of N release to the atmosphere.
This study, involving original and unique data from remote and seldom explored regions, will be extended to other West African ecosystems in the future.

Fig. 2 .
Fig. 2. Monthly evolution of fluxes and associated uncertainties: in purple, dry deposition of NO 2 and HNO 3 + wet deposition of NO 3 and in blue, biogenic NO emission from soils + biomass burning flux + domestic fires flux in kgNha −1 yr −1 for IDAF dry savanna stations.

Figure 3
Figure 3 shows the distribution of the N oxidized component emission fluxes (a) and deposition fluxes (b), for the wet season.The same repartition is given in Fig. 4a and b but for annual means in the three dry savanna stations.Uncertainty bars are given for each contribution in an approximated color.It is obvious from these figures that the emission flux is dominated by the biogenic part, both in the wet season and for the annual mean.Mean values and uncertainties for each season are reported in Table1.Emission fluxes at the annual scale range from 1.0±0.1 to 2.4±0.8kgNha −1 yr −1 .These estimates are in the range of emission fluxes reported byMeixner and Yang (2004), which range from 0.02 to 34 kgNha −1 yr −1 , in semi arid and arid ecosystems.In Bani-

Fig. 8 .
Fig.8.Mean repartition of emission (using 30% for NH 3 volatilization rate) and deposition (oxidized and reduced) N fluxes, scaled up at the Sahelian region (2.4 millions km 2 ).Given as absolute value (relative uncertainty); contribution in % • N, 2.41 • E, Niger) from 1994 until now, in Agoufou (15.3 • N, 1.5 • W, Mali) from 2004 to 2007, and in Katibougou (12.5 • N, 7.3 • W, Mali) from 1997 up to now.Comprehensive descriptions of the stations can be found in Mougin et al. (2009) for Agoufou, in Galy-Lacaux and Modi (1998) for Banizoumbou, and in Adon et al. (2010) for Katibougou.Agoufou and Banizoumbou are part of the AMMA CATCH (Couplage de l'Atmosphère Tropicale et du Cycle Hydrologique) observatory.Banizoumbou is located in a rural and agro pastoral area of the Sahelian region of Niger, approximately at 60 km from Niamey.Katibougou is located at 60 km from Bamako.No major source of anthropogenic pollution influences the station.Agoufou is situated towards the northern limit of the area reached by the West African monsoon.

Table 1 .
Seasonal and annual mean emission and deposition fluxes and uncertainties (in kgNha −1 yr −1 ) of oxidized nitrogen compounds in dry savanna stations.

Table 2 .
Annual mean emission and deposition fluxes and uncertainties (in kgNha −1 yr −1 ) of reduced nitrogen compounds for dry savanna stations.

Table 1 .
The wet season lasts from June to September in Agoufou, from May to September in Banizoumbou, and from May to October in Katibougou.The first maximum during the wet season corresponds to biogenic emissions.During the wet season, emission fluxes are larger than deposition fluxes in Banizoumbou and Agoufou, due to the large amount of biogenic NO emissions in July and August, associated to the maximum of rainfall.It seems that all the N emitted is not represented in the deposition measurements, which include only 3 N gaseous compounds and do not include Organic Nitrogen, as mentioned above.During the dry season, emission fluxes are close to deposition fluxes.

Table 3 .
Total emission and deposition fluxes and uncertainties (in kgNha −1 yr −1 ) of reduced and oxidized nitrogen compounds in dry savanna.