Tropospheric chemistry in the integrated forecasting system of ECMWF

Abstract. A representation of atmospheric chemistry has been included in the Integrated Forecasting System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF). The new chemistry modules complement the aerosol modules of the IFS for atmospheric composition, which is named C-IFS. C-IFS for chemistry supersedes a coupled system in which chemical transport model (CTM) Model for OZone and Related chemical Tracers 3 was two-way coupled to the IFS (IFS-MOZART). This paper contains a description of the new on-line implementation, an evaluation with observations and a comparison of the performance of C-IFS with MOZART and with a re-analysis of atmospheric composition produced by IFS-MOZART within the Monitoring Atmospheric Composition and Climate (MACC) project. The chemical mechanism of C-IFS is an extended version of the Carbon Bond 2005 (CB05) chemical mechanism as implemented in CTM Transport Model 5 (TM5). CB05 describes tropospheric chemistry with 54 species and 126 reactions. Wet deposition and lightning nitrogen monoxide (NO) emissions are modelled in C-IFS using the detailed input of the IFS physics package. A 1 year simulation by C-IFS, MOZART and the MACC re-analysis is evaluated against ozonesondes, carbon monoxide (CO) aircraft profiles, European surface observations of ozone (O3), CO, sulfur dioxide (SO2) and nitrogen dioxide (NO2) as well as satellite retrievals of CO, tropospheric NO2 and formaldehyde. Anthropogenic emissions from the MACC/CityZen (MACCity) inventory and biomass burning emissions from the Global Fire Assimilation System (GFAS) data set were used in the simulations by both C-IFS and MOZART. C-IFS (CB05) showed an improved performance with respect to MOZART for CO, upper tropospheric O3, and wintertime SO2, and was of a similar accuracy for other evaluated species. C-IFS (CB05) is about 10 times more computationally efficient than IFS-MOZART.


Introduction
Monitoring and forecasting of global atmospheric composition are key objectives of the atmosphere service of the European Copernicus Programme. The Copernicus Atmosphere Monitoring Service (CAMS) is based on combining satellite observations of atmospheric composition with state-of-the-art atmospheric modelling (Flemming et al., 5 2013;Hollingsworth et al., 2008). For that purpose, the integrated forecasting system (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF) was extended for forecast and assimilation of atmospheric composition. Modules for aerosols (Morcrette et al., 2009;Benedetti et al., 2009) and greenhouse gases (Engelen et al., 2009) were integrated on-line in the IFS. Because of the complexity of the chemical mechanisms for reactive gases, modules for atmospheric chemistry were not initially included in the IFS. Instead a coupled system (Flemming et al., 2009a) was developed, which couples the IFS to the Chemical Transport Model (CTM) Model for OZone and Related chemical Tracers 3 (MOZART, Kinnison et al., 2007) or Transport Model 5 (TM5, Huijnen et al., 2010) by means of the Ocean Atmosphere Sea Ice Soil cou- 15 pling software (OASIS4) coupler software (Redler et al., 2010). Van Noije et al. (2014) coupled TM5 to IFS for climate applications in a similar approach. The coupled system made it possible to assimilate satellite retrievals of reactive gases with the assimilation algorithm of the IFS, which is also used for the assimilation of meteorological observations as well as for aerosol and greenhouse gases. 20 The coupled system IFS-MOZART has been successfully used for a re-analysis of atmospheric composition (Inness et al., 2013), pre-operational atmospheric composition forecasts (Stein et al., 2012), forecast and assimilation of the stratospheric ozone (O 3 ) (Flemming et al., 2011a;Lefever et al., 2014) and tropospheric carbon monoxide (CO) (Eligundi et al., 2010) and O 3 (Ordonez et al., 2010). The coupled system IFS-including modules for atmospheric composition is named Composition-IFS (C-IFS). C-IFS makes it possible (i) to use the detailed meteorological simulation of the IFS for the simulation of the fate of constituents (ii) to use the IFS data assimilation system to assimilate observations of atmospheric composition and (iii) to simulate feedback processes between atmospheric composition and weather. A further advantage of C-IFS 10 is the possibility of model runs at a high horizontal and vertical resolution because of the high computational efficiency of C-IFS. C-IFS is the global model system run in pre-operational mode as part of the Monitoring Atmospheric Composition and Climate -Interim Implementation project (MACC II and MACC III) in peraration of CAMS.
Including chemistry modules in general circulation models (GCM) to simulate interac- 15 tion of stratospheric O 3 (e.g. Steil et al., 1998) and aerosols (e.g. Haywood et al., 1997) in the climate system started in the mid-1990s. Later, more comprehensive schemes for tropospheric chemistry were included in climate GCM such as ECHAM5-HAMMOZ (Pozzoli et al., 2008;Rast et al., 2014) and CAM-chem (Lamarque et al., 2012) to study short-lived greenhouse gases and the influence of climate change on air pollution (e.g. Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | discretization uses 60 levels up to the model top at 0.1 hPa (65 km) in a hybrid sigmapressure coordinate. The vertical extent of the lowest level is about 17 m; it is 100 m at about 300 m above ground, 400-600 m in the middle troposphere and about 800 m at about 10 km height. The modus operandi of C-IFS is one of a forecast model in a NWP framework. The 5 simulations of C-IFS are a sequence of daily forecasts over a period of several days. Each forecast is initialised by the ECMWF's operational analysis for the meteorological fields and by the 3-D chemistry fields from the previous forecast ("forecast mode"). Continuous simulations over longer periods are carried out in "relaxation mode". In relaxation mode the meteorological fields are relaxed to the fields of a meteorological 10 re-analysis, such as ERA-Interim, during the run (Jung et al., 2008) to ensure realistic and consistent meteorological fields.

Transport
The transport by advection, convection and turbulent diffusion of the chemical tracers uses the same algorithms as developed for the transport of water vapour in the 15 NWP applications of IFS. The advection is simulated with a three-dimensional semi-Lagrangian advection scheme, which applies a quasi-montonic cubic interpolation of the departure values. Since the semi-Lagrangian advection does not formally conserve mass a global mass fixer is applied. The effect of different global mass fixers is discussed in Diamantakis and Flemming (2014) and Flemming and Huijnen (2011b). 20 A proportional mass was used for the runs presented in this paper because of the overall best balance between the results and computational cost. The vertical turbulent transport in the boundary layer is represented by a first order K-diffusion closure. The surface emissions are injected as lower boundary flux in the diffusion scheme. The lower boundary flux condition also accounts for the dry deposi- 25 tion flux based on the projected surface mass mixing ratio in an implicit way. The vertical transport by convection is simulated as part of the cumulus convection. It applies a bulk mass flux scheme which was originally described in Tiedtke (1989 and aerosol precursors are assumed to be scavenged in the convective rain droplets 5 and are therefore excluded from the convective mass transfer. The operator splitting between the transport and the sink and source terms follows the implementation for water vapour (Beljaars et al., 2004). Advection, diffusion and convection are simulated sequentially. The sink and source processes are simulated in parallel using an intermediate update of the mass mixing ratios with all transport 10 tendencies. At the end of the time step tendencies from transport and sink and source terms are added together for the final update the concentration fields. Resulting negative mass mixing ratios are corrected at this point by setting the updated mass mixing ratio to a "chemical zero" of 1.0 × 10 −25 kg kg −1 . 15 The anthropogenic surface emissions were given by the MACCity inventory (Granier et al., 2011) and aircraft NO emissions of a total of ∼ 0.8 Tg N yr −1 were applied (Lamarque et al., 2010). Natural emissions from soils and oceans were taken from the Precursors of Ozone and their Effects in the Troposphere (POET) database for 2000 Olivier et al., 2003). The biogenic emissions were simulated by Introduction

Dry deposition
Dry deposition is an important removal mechanism at the surface in the absence of precipitation. It depends on the diffusion close to the earth surface, the properties of the constituent and on the characteristics of the surface, in particular the type and state 5 of the vegetation and the presence of intercepted rain water. Dry deposition plays an important role in the biogeochemical cycles of nitrogen and sulphur, and it is a major loss process of tropospheric O 3 . Modelling the dry deposition fluxes in C-IFS is based on a resistance model (Wesely et al., 1989), which differentiates the aerodynamic, the quasi-laminar and the canopy or surface resistance. The inverse of the total resistance 10 is equivalent to a dry deposition velocity V D .
The dry deposition flux F D at the model surface is calculated based on the dry deposition velocity V D , the mass mixing ratio X s and air density ρ s at the lowest model level s, in the following way: The calculation of the loss by dry deposition has to account for the implicit character of the dry deposition flux since it depends on the mass mixing ratio X s . itself The dry deposition velocities were calculated as monthly mean values from a oneyear simulation using the approach described in Michou et al. (2004). It used meteorological and surface input data such as wind speed, temperature, surface roughness 20 and soil wetness from the ERA-interim data set. At the surface the scheme makes a distinction between uptake resistances for vegetation, bare soil, water, snow and ice. The surface and vegetation resistances for the different species are calculated using the stomatal resistance of water vapour. The stomatal resistance for water vapour is calculated depending on the leaf area index, radiation and the soil wetness at the up-25 permost surface layer. Together with the cuticlular and mesophyllic resistances this is combined into the leaf resistance according to Wesely et al. (1989) using season and surface type specific parameters as referenced in Seinfeld and Pandis (1998). Dry deposition velocities have higher values during the day because of lower aerodynamic resistance and canopy resistance. Zhang et al. (2003) reported that averaged observed O 3 and sulphur dioxide (SO 2 ) dry deposition velocities can be up to 4 times 5 higher at day time than at night time. As this important variation is not captured with the monthly-mean dry deposition values, a ±50 % variation is imposed on all dry deposition values based on the cosine of the solar zenith angle. This modulation tends to decrease dry deposition for species with a night time maximum at the lowest model level and it increases dry deposition of O 3 . 10 Table S4 (Supplement) contains annual total loss by dry deposition and expressed as a life-time estimate by dividing by tropospheric burden for a simulation using monthly dry deposition values for 2008. Dry deposition was most effective for many species in particular SO 2 and ammonia (NH 3 ) as the respective lifetimes were one day to one week. For tropospheric O 3 the respective globally averaged time scale is about 15 3 months. Because dry deposition occurs mainly over ice-free land surfaces the corresponding time scale is at least three times shorter in these areas.

Wet deposition
Wet deposition is the transport and removal of soluble or scavenged constituents by precipitation. It includes the following processes: 20 in-cloud scavenging and removal by rain and snow (rain out).
-Release by evaporation of rain and snow.
-Below cloud scavenging by precipitation falling through without formation of precipitation (wash out).
It is important to take the sub-grid scale of cloud and precipitation-formation into ac- on the cloud and the precipitation fraction for each grid box. It uses a random overlap assumption (Jakob and Klein, 2000) to derive cloud and precipitation area fraction. The same method has been used by Neu and Prather (2012), who demonstrated the importance of the overlap assumption for the simulation of the wet deposition. The precipitation fluxes for the simulation of wet removal in C-IFS were scaled to be valid 5 over the precipitation fraction of the respective grid-box. The loss of tracer by rain-out and wash-out was limited to the area of the grid box covered by precipitation. Likewise, the cloud water and ice content is scaled to the respective cloud area fraction. If the sub-grid scale distribution was not considered in this way, wet deposition was lower for highly soluble species such as HNO 3 because the species is only removed from the 10 cloudy or rainy grid box fraction. For species with low solubility the wet deposition loss was slightly decreased because of the decrease in effective cloud and rain water. Even if wet deposition removes tracer mass only in the precipitation area, the mass mixing ratio representing the entire grid box is changed accordingly after each model time step. This is equivalent with the assumption that there is instantaneous mixing 15 within the grid-box at the time scale of the model time step. As discussed in , this assumption may lead to an overestimation of the simulated tracer loss.
The module for wet deposition in C-IFS is based on the Harvard wet deposition scheme Liu et al., 2001). In contrast to Jacob et al. (2000), trac-20 ers scavenged in wet convective updrafts are not removed as part of the convection scheme. Nevertheless, the fraction of highly soluble tracers in cloud condensate is simulated to limit the amount of tracers lifted upwards as only the gas phase fraction is transported by the mass flux. The removal by convective precipitation is simulated in the same way as for large-scale precipitation in the wet deposition routine. Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | at the grid cell interfaces. For convective precipitation a precipitation fraction of 0.05 is assumed and the convective rain and snow water content is calculated assuming a droplet fall speed of 5 m s −1 .
Wash-out, evaporation and rain-out are calculated after each other for large-scale and convective precipitation. The amount of trace gas dissolved in cloud droplets is 5 calculated using Henrys-law-equilibrium or assuming that 70 % of aerosol precursors such as sulphate (SO 4 ), NH 3 and nitrate (NO 3 ) is dissolved in the droplet. The effective Henry coefficient for SO 2 , which accounts for the dissociation of SO 2 , is calculated following Seinfeld and Pandis (1998, p. 350). The other Henry's law coefficients are taken from the compilation by Sander (1999) (www.henrys-law.org, Table S1 in the Supplement).
The loss by rain out is determined by the precipitation formation rate. The retention coefficient R, which accounts for the retention of dissolved gas in the liquid cloud condensate as it is converted to precipitation, is one for all species in warm clouds (T > 268 K). For mixed clouds (T < 268 K) R is 0.02 for all species but 1.0 for HNO 3 15 and 0.6 for H 2 O 2 (von Blohn, 2011). In ice clouds only H 2 O 2 (Lawrence and Crutzen, 1998) and HNO 3 are scavenged.
Partial evaporation of the precipitation fluxes leads to the release of 50 % of the resolved tracer and 100 % in the case of total evaporation (Jacobs et al., 2000). Washout is either mass-transfer or Henry-equilibrium limited. HNO 3 , aerosol precursors and 20 other highly soluble gases are washed out using a first order wash-out rate of 0.1 mm −1 (Levine and Schwartz, 1982) to account for the mass transfer. For less soluble gases the resolved fraction in the rain water is calculated assuming Henry equilibrium in the evaporated precipitation.   or 17 % of the total anthropogenic emissions. NO emissions from lightning play an important role in the chemistry of the atmosphere because they are released in the rather clean air of the free troposphere, where they can influence the O 3 budget and hence the OH-HO 2 partitioning (DeCaria et al., 2005). The parameterization of the lightning NO production in C-IFS consist of estimates 10 of (i) the flash rate density, (ii) the flash energy release and (iii) the vertical emission profile for each model grid column. The estimate of the flash-rate density is based on parameters of the convection scheme. The C-IFS has two options to simulate the flashrate densities using the following input parameters: (i) convective cloud height (Price and Rind, 1992) or (ii) convective precipitation (Meijer et al., 2001). 15 The parameterizations distinguish between land and ocean points by assuming about 5-10 times higher flash rates over land. Additional checks on cloud base height, cloud extent and temperature are implemented to select only clouds that are likely to generate lightning strokes. The coefficients of the two parameterizations were derived from field studies and depend on the model resolution. With the current implementation 20 of C-IFS (T255L60), the global flash rates were 26 and 43 flashes per seconds for the schemes by Price and Rind (1992) and Meijer et al. (2001), respectively. It seemed therefore necessary to scale the coefficients to get a flash rate in the range of the observed values of about 40-50 flashes per second derived from observations of the Optical Transient Detector (OTD) and the Lightning Imaging Sensor (LIS) (Cecil et al.,25 2012). Figure 1 shows the annual flash rate density simulated by the two parameterisations together with observations from the LIS/OTD data set. The two approaches show the main flash activity in the tropics but there were differences in the distributions over land and sea. The smaller land-sea differences of Meijer et al. (2001) agreed better with the observations. The observed maximum over Central African was well reproduced by both parameterizations but the schemes produce an exaggerated maximum over tropical South America. The lightning activity over the United States was under-5 estimated by both parameterisations. The parameterization by Meijer et al. (2001) has been used for the C-IFS runs presented in this paper. Cloud to ground (CG) and cloud to cloud (CC) flashes are assumed to release a different amount of energy, which is proportional to the NO release. Price et al. (1997) suggest that the energy release of CG is 10 times higher. However, more recent studies suggest a similar value for CG and CC energy release based on aircraft observations and model studies (Ott et al., 2010), which we follow in C-IFS. In C-IFS, CG and CC fractions are calculated using the approach by Price and Rind (1993), which is based on a 4th order function of cloud height above freezing level.
The vertical distribution of the NO release is of importance for its impact on atmo-15 spheric chemistry. Many CTMs use the suggestion of Pickering et al. (1998) of a Cshape profile, which peaks at the surface and in the upper troposphere. Ott et al. (2010) suggest a "backward C-shape" profile which locates most of the emission in the middle of the troposphere. The vertical distribution can be simulated by C-IFS (i) according to Ott et al. (2010) or (ii) as a C-shape profile following Huijnen et al. (2010). The ap-20 proach by Ott et al. (2010) is used in the simulation presented here. As lightning NO emissions occur mostly in situations with strong convective transport, differences in the injection profile had little impact.
As the lightning emissions depend on the convective activity they change at different resolutions or after changes to the convection scheme. The C-IFS lightning emissions were 4.9 Tg N yr −1 at T159 resolution and 5.7 Tg N yr −1 at T255 resolution.

Gas-phase chemistry
The chemical mechanism is a modified version of the Carbon Bond mechanism 5 (CB05, Yarwood et al., 2005), which is originally based on the work of Gery et al. (1989) with added reactions from Zaveri and Peters (1999) and from Houweling et al. (1998) 5 for isoprene. The CB05 scheme adopts a lumping approach for organic species by defining a separate tracer species for specific types of functional groups. The speciation of the explicit species into lumped species follows the recommendations given in Yarwood et al. (2005). The CB05 scheme used in C-IFS has been further extended in the following way: an explicit treatment of methanol (CH 3 OH), ethane (C 2 H 6 ), propane 10 (C 3 H 8 ), propene (C 3 H 6 ) and acetone (CH 3 COCH 3 ) has been introduced as described in Williams et al. (2013). The isoprene oxidation has been modified motivated by Archibald et al. (2010). Higher C 3 peroxy-radicals formed during the oxidation of C 3 H 6 and C 3 H 8 were included following Emmons et al. (2010). The CB05 scheme is supplemented with chemical reactions for the oxidation of SO 2 , 15 di-methyl sulphide (DMS), methyl sulphonic acid (MSA) and NH 3 , as outlined in . For the oxidation of DMS, the approach of Chin et al. (1996) is adopted. Table S1 (Supplement) gives a comprehensive list of the trace gases included in the chemical scheme. The reaction rates have been updated according to the recommendations given in 20 either Sander et al. (2011) or Atkinson et al. (2004Atkinson et al. ( , 2006. The oxidation of CO by the hydroxyl radical (OH) implicitly accounts for the formation and subsequent decomposition of the intermediate species HOCO as outlined in Sander et al. (2006). For lumped species, e.g. ALD2, the reaction rate is determined by an average of the rates of reaction for the most abundant species, e.g. C 2 and C 3 aldehydes, in that group. Introduction For the loss of trace gases by heterogeneous oxidation processes, the model explicitly accounts for the oxidation of SO 2 in cloud through aqueous phase reactions with H 2 O 2 and O 3 , depending on the acidity of the solution. In this version of C-IFS, heterogeneous conversion of N 2 O 5 into HNO 3 on cloud droplets and aerosol particles is applied with a reaction probability (γ) set to 0.02 (Evans and Jacob, 2005).

Photolysis rates
For the calculation of photo-dissociation rates an on-line parameterization for the derivation of actinic fluxes is used (Williams et al., 2012(Williams et al., , 2006. It applies a Modified Band Approach (MBA) which is an updated version of the work by Landgraf and Crutzen (1998), tailored and optimized for use in tropospheric CTMs. The approach 10 uses 7 absorption bands across the spectral range 202-695 nm. At instances of large solar zenith angles (71-85 • ) a different set of band intervals is used. In the MBA the radiative transfer calculation using the absorption and scattering components introduced by gases, aerosols and clouds is computed on-line for each of 7 pre-defined band intervals based on the 2-stream solver of Zdunkowski et al. (1980). 15 The optical depth of clouds is calculated based on a parameterization available in IFS (Slingo, 1989;Fu et al., 1996) for the cloud optical thickness at 550 nm. For the simulation of the impact of aerosols on the photolysis rates a climatological field for aerosols is used, as detailed in Williams et al. (2012). There is also an option to use the MACC aerosol fields. 20 In total 20 photolysis rates are included in the scheme, as given in Table S3 (Supplement). The explicit nature of the MBA implies a good flexibility in terms of updating molecular absorption properties (cross sections and quantum yields) and the addition of new photolysis rates into the model.

The chemical solver
The chemical solver used in C-IFS (CB05) is an Euler Backward Iterative (EBI) solver (Hertel et al., 1996). This solver has been originally designed for use with the CBM4 mechanism of Gery et al. (1989). The chemical time step is 22.5 min, which is half of the dynamical model time step of 45 min at T255 resolution. Eight, four or one itera-5 tions are carried out for fast-, medium-and slow-reacting chemical species to obtain a solution. The number of iterations is doubled in the lowest four models levels, where the perturbations due to emissions can be large.

Stratospheric boundary conditions
The modified CB05 chemical mechanism includes no halogenated species and no 10 photolytic destruction below 202 nm and is therefore not suited for the description of stratospheric chemistry. Thus realistic upper boundary conditions for the longer-lived gases such as O 3 , methane (CH 4 ), and HNO 3 are needed to capture the influence of stratospheric intrusions on the composition of the upper troposphere. Stratospheric O 3 chemistry in C-IFS (CB05) is parameterized by the Cariolle scheme 15 (Cariolle and Teyssèdre, 2007). Chemical tendencies for stratospheric and tropospheric O 3 are merged at an empirical interface of the diagnosed tropopause height in IFS. Additionally, stratospheric O 3 in C-IFS can be nudged to O 3 analyses of either the MACC re-analysis (Inness et al., 2013) or ERA interim (Dee et al., 2011). The tropopause height in IFS is diagnosed either from the gradient in humidity or the vertical 20 temperature gradient. Stratospheric HNO 3 at 10 hPa is controlled by a climatology of HNO 3 and O 3 observations from the Microwave Limb Sounder (MLS) aboard the Upper Atmosphere Research satellite (UARS). HNO 3 is set to according to the observed HNO 3 /O 3 ratio and the simulated O 3 concentrations. Further, stratospheric CH 4 is constrained by 25 a climatology based on observations of the Halogen Occultation Experiment instrument (Grooß and Russel, 2005), at 45 hPa and at 90 hPa in the extra-tropics, which GMDD 7, 2014 Tropospheric chemistry in the integrated forecasting system of ECMWF implicitly accounts for the stratospheric chemical loss of CH 4 by OH, chlorine (Cl) and oxygen (O 1 D) radicals. It should be noted that also the surface concentrations of CH 4 are fixed in this configuration of the model.

Gas-aerosol partitioning
Gas-aerosol partitioning is calculated using the Equilibrium Simplified Aerosol Model 5 (EQSAM, Metzger et al., 2002a, b). The scheme has been simplified so that only the partitioning between HNO 3 and the nitrate aerosol (NO − 3 ) and between NH 3 and the ammonium aerosol (NH + 4 ) is calculated. SO 2− 4 is assumed to remain completely in the aerosol phase because of its very low vapour pressure. The assumptions of the equilibrium model are that (i) aerosols are internally mixed and obey thermodynamic gas/aerosol equilibrium and that (ii) the water activity of an aqueous aerosol particle is equal to the ambient relative humidity (RH). Furthermore, the aerosol water mainly depends on the aerosol mass and the type of the solute, so that parameterizations of single solute molalities and activity coefficients can be defined, depending only on the type of the solute and RH. The advantage of using such parameterizations is that 15 the entire aerosol equilibrium composition can be solved analytically. For atmospheric aerosols in thermodynamic equilibrium with the ambient RH, the following reactions are considered in C-IFS. The subscripts g, s and aq denote gas, solid and aqueous phase, respectively:

Model budget diagnostics
C-IFS computes global diagnostics for every time step to study the contribution of different processes on the global budget. The basic outputs are the total and tropospheric GMDD 7, 2014 Tropospheric chemistry in the integrated forecasting system of ECMWF The comparatively simple treatment of volatile organic compounds in CB05 could be an explanation for the low O 3 production and loss terms. Stratospheric inflow in C-IFS, estimated as the residue from the remaining terms was 691 Tg and the corresponding value from the ACCENT multi-model mean is 552 ± 168 Tg. The annual mean total CO burden in C-IFS was 361 Tg, which is slightly larger 5 than the ACCENT mean (345 Tg, 248-427 Tg). The total CO emissions in 2008 were 1008 Tg which is in-line with the number used in ACCENT (1077 Tg yr −1 ) but lower than the estimate (1550 Tg yr −1 ) of the Third Assessment Report (Prather and Ehhalt, 2001) of the Intergovernmental Panel on Climate Change (IPCC), which also takes into account results from inverse modelling studies. The tropospheric chemical CO production 10 was 1434 Tg yr −1 , which is very close to the ACCENT multi-mean of 1505±236 Tg yr −1 . The chemical CO loss in C-IFS was 2423 Tg and the loss by dry deposition 24 Tg. The annual mean CH 4 total and tropospheric burdens of C-IFS (CB05) are 4874 and 4271 Tg yr −1 , respectively. The global chemical CH 4 loss by OH was 467 Tg yr −1 . Following Stevenson et al. (2006), this leads to a global CH 4 lifetime estimate of 9.1 yr. This 15 value is within the ACCMIP range of 9.8 ± 1.6 yr but lower than an observation-based 11.2 ± 1.3 yr estimate by Prather and Ehhalt (2012). CH 4 emissions were substituted by prescribed monthly zonal-mean surface concentrations to avoid the long-spin up needed by a direct modeling of the CH 4 surface fluxes. The resulting CH 4 flux was 488 Tg yr −1 , which is of similar size as the sum of current estimates of the total CH 4 20 emissions of 500-580 Tg yr −1 and the loss by soils of 30-40 Tg yr −1 (Forth Assessment Report by IPCC http://www.ipcc.ch/publications_and_data/ar4/wg1/en/ch7s7-4-1.html#ar4top). 25 The main motivation for the development of C-IFS is forecasting and assimilation of atmospheric composition as part of the CAMS. Hence, the purpose of this evaluation 7752 Introduction is to show how C-IFS (CB05) performs relative to the coupled CTM MOZART-3 (Kinnison et al., 2007), which has been running in the coupled system IFS-MOZART in pre-operational mode since 2007. C-IFS will replace the coupled system in the next update of the CAMS system. The evaluation focuses on species which are relevant to global air pollution such as tropospheric O 3 , CO, nitrogen doxide (NO 2 ), SO 2 and 5 formaldehyde (HCHO). The MACC re-analysis (Inness et al., 2013), which is an application of IFS-MOZART with assimilation of observations of atmospheric composition, has been included in the evaluation as a benchmark.

Evaluation with observations and comparison with the coupled system IFS-MOZART
The MACC re-analysis (REAN) and the corresponding MOZART (MOZ) stand-alone run have already been evaluated with observations by Inness et al. (2013). Further, the MACC-II sub-project on validation has compiled a comprehensive report assessing REAN (MACC, 2013). REAN has been further evaluated with surface observations in Europe and North-America for O 3 by Im et al. (2014). C-IFS (CB05) has been already evaluated with a special focus on hydroperoxyl (HO 2 ) in relation to CO in . The performance of an earlier version of C-IFS (CB05) in the Arctic was  20 C-IFS (CB05) was run from 1 January to 31 December 2008 with a spin up starting 1 July 2007 at a T255 resolution with 60 model levels in monthly chunks. The meteorological simulation was relaxed to dynamical fields of the MACC re-analysis (see Sect. 2.1). Likewise stratospheric O 3 above the tropopause was nudged to the MACC re-analysis. difference between MOZ and REAN is the assimilation of satellite retrieval of atmospheric composition in REAN. Further, REAN was produced with the coupled system IFS-MOZART whereas MOZ is a stand-alone driven by the meteorological fields of REAN. The latter is equivalent with a simulation of IFS-MOZART without data assimilation of atmospheric composition. The assimilated retrievals were CO and O 3 total Monitoring Experiment 2 (GOME-2). Measurements Of Pollution In The Troposphere (MOPITT) retrievals were used for the validation of the global CO total column fields.

In-situ observations
The ozonesondes were obtained from the World Ozone and Ultraviolet Radiation Data Centre (WOUDC) and from the ECWMF Meteorological Archive and Retrieval System. 5 The observation error of the sondes is about ±5 % in the range between 200 and 10 hPa and −7-17 % below 200 hPa (Beekmann et al., 1994;Komhyr et al., 1995;Steinbrecht et al., 1996). The number of soundings varied for the different stations. Typically, the sondes are launched once a week but in certain periods such as during O 3 hole conditions soundings are more frequent. Sonde launches were carried out 10 mostly between 09:00 and 12:00 LT. The global distribution of the launch sites is even enough to allow meaningful averages over larger areas such North-America, Europe, the Tropics, the Artic and Antarctica. Table 2 contains a list of the ozonesondes used in this study. The MOZAIC program (Marenco et al., 1998;Nédélec et al., 2003) provides pro- 15 files of various trace gases taken during commercial aircraft ascents and descents at specific airports. MOZAIC CO data have an accuracy of ±5 ppbv, a precision of ±5 %, and a detection limit of 10 ppbv (Nédélec et al., 2003). Since the aircraft carrying the MOZAIC unit were based in Frankfurt, the majority of the CO profiles (837 in 2008) were observed at this airport. A further 10 of the 28 airports with observations The North-American airports were considered to be close enough to make a spatial average meaningful. 25 Apart from Frankfurt, typically 2 profiles (takeoff and landing) are taken within 2-3 h or with a longer gap in the case of an overnight stay. At Frankfurt there were 2-6 profiles available each day, mostly in the morning and the later afternoon to the evening. and to some extent in Frankfurt represent a more mixed day-time boundary layer. The global atmospheric watch (GAW) program of the World Meteorological Organization is a network for mainly surface based observations (WMO, 2007). The data were retrieved from the World Data Centre for Greenhouse Gases (http://ds.data.jma. go.jp/gmd/wdcgg/). The GAW observations represent the global background away from the main polluted areas. Often, the GAW observation sites are located on mountains, which makes it necessary to select a model level different from the lowest model level for a sound comparison with the model. In this study the procedure described in Flemming et al. (2009b) is applied to determine the model level, which is based on the 15 difference between a high resolution orography and the actual station height. The data coverage for CO and O 3 was global, whereas for SO 2 and NO 2 only a few observations in Europe were available at the data repository.
The Airbase and EMEP databases host operational air quality observations from different national European networks. All EMEP stations are located in rural areas, while 20 Airbase stations are designed to monitor local pollution. Many AirBase observations may therefore not be representative of a global model with a horizontal resolution of 80 km. However, stations of rural regime may capture the larger scale signal in particular for O 3 , which is spatially well correlated (Flemming et al., 2005). Only the rural Airbase O 3 observations have been selected for the evaluation of the diurnal cycle. 5. This is equivalent to a bias of about 4 % and a SD of 10 % respectively assuming typical observations of 2.0 × 10 18 molec cm −2 . For the calculation of the simulated CO total column the averaging kernels (AK) of the retrievals were applied. They have the largest values between 300 and 800 hPa. At surface the sensitivity is reduced even though the combined NIR/TIR product has been used, which has a higher sensitivity 20 than the NIR and TIR only products. Applying the AK makes the difference between retrieval and AK-weighted model column independent of the a-priori CO profiles used in the retrieval. On the other hand, it makes the total column calculation dependent on the modelled profile. The AK-weighted column is not equivalent to the modelled atmospheric burden anymore, which needs to be considered for the interpretation of the 25 results. GOME-2 is a ultra violet-visibile (UV-VIS) and NIR sensor designed to provide global observations of atmospheric trace gases. GOME-2 flies in a sun-synchronous orbit with an equator crossing time of 09:30 LT in descending mode and has a footprint GMDD Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | of 40 km × 80 km. Here, tropospheric vertcial columns of NO 2 and HCHO have been computed using a three step approach. First, the Differential Optical Absorption Spectroscopy (DOAS; Platt, 1994) method is applied to measured spectra which yields the total slant column. The DOAS method is applied in a 425-497 nm wavelength window (Richter et al., 2011) for NO 2 . and between 337 and 353 nm for HCHO. Second, the 5 reference sector approach is applied to total slant columns for stratospheric correction. In a last step, tropospheric slant columns are converted to tropospheric vertical columns by applying an air mass factor. Only data with cloud fractions smaller than 0.2 according to the FRESCO cloud data base (Wang et al., 2008) are used here. Furthermore, retrievals are limited to maximum solar zenith angles of 85 • for NO 2 and 60 for HCHO. Uncertainties in NO 2 satellite retrievals are large and depend on the region and season. Winter values in mid and high latitudes are usually associated with larger error margins. As a rough estimate, systematic uncertainties in regions with significant pollution are of the order of 20-30 %. As the HCHO retrieval is performed in the UV part of the spectrum where less light is available and the HCHO absorption signal is 15 smaller than that of NO 2 , the uncertainty of monthly mean HCHO columns is relatively large (20-40 %) and both noise and systematic offsets have an influence on the results. However, absolute values and seasonality are retrieved more accurately over HCHO hotspots. For comparison to GOME-2 data, model data are vertically integrated to tropospheric 20 vertical columns of NO 2 and HCHO, interpolated to satellite observation time and then sampled to match the location of available cloud free satellite data, which has been gridded to match the model resolution. The resulting daily files are then averaged over months for both satellite and model data to reduce the noise. Figure 2 shows the monthly means of O 3 volume mixing ratios in the pressure ranges surface to 700 hPa (lower troposphere, LT) 700-400 hPa (middle troposphere, MT) and  Figure 3 shows the same as Fig. 2 for the Tropics, Arctic and Antarctica. The observations have a pronounced spring maximum for UT O 3 over Europe, North America and East Asia and a more gradually developing maximum in late spring and summer in MT 5 and LT. The LT seasonal cycle is well re-produced in all runs for the areas of the Northern Hemisphere (NH). In Europe, REAN tends to overestimate by about 5 ppb where the C-IFS and MOZ have almost no bias before the annual maximum in May apart from a small negative bias in spring. Later in the year, C-IFS tends to overestimate in autumn, whereas MOZ overestimates more in late summer. In MT over Europe C-IFS 10 agrees slightly better with the observations than MOZ. MOZ overestimates in winter and spring and this overestimation is more prominent in the UT, where MOZ is biased high throughout the year. This overestimation in UT is highest in spring, where it can be 25 % and more. These findings show that data assimilation in REAN improved UT O 3 considerably but had only little influence in LT and MT. The overestimation of MOZ 15 in UT seems to be caused by increased stratospheric O 3 rather than a more efficient transport as lower stratospheric O 3 was overestimated in MOZ. The good agreement of C-IFS with observation in UT in all three regions is also present in a run without nudging to stratospheric O 3 . It is therefore not a consequence of the use of assimilated observations in C-IFS (CB05). 20 Over North-America the spring time underestimation by C-IFS and MOZ is more pronounced than over Europe. C-IFS also underestimated MT O 3 observations in this period, whereas MOZ and REAN slightly overestimate. In East Asia all runs overestimate by 5-10 ppb in LT and MT especially in autumn and winter. In the northern high latitudes (Fig. 3) the negative spring bias appears in all runs in LT and only for C-IFS in 25 MT. As in the other regions, MOZ greatly overestimates UT O 3 .

Tropospheric ozone
Averaged over the tropics, the annual variability is below 10 ppb with maxima in May in September caused by the dry season in South-America (May) and Africa (September). The variability is well reproduced and biases are mostly below 5 ppb in the whole troposphere. Note that the 400-200 hPa range (UT) in the tropics is less influenced by the stratosphere because of the higher tropopause. C-IFS had smaller biases because of lower values in LT and higher values in MT and UT than MOZ.
Over the Arctic C-IFS and MOZ reproduce the seasonal cycle, which peaks in late spring, but generally underestimate the observations in LT. C-IFS had a smaller bias 5 in LT than MOZ but had a larger negative bias in MT. The biggest improvement of C-IFS w.r.t to MOZ occurred at the surface in Antarctica as the biases compared to the GAW surface observations were greatly reduced. Notably, the assimilation (REAN) led to increased biases for LT and MT O 3 , in particular during polar night when UV satellite observations are not available as already discussed in Flemming et al. (2011a). 10 The ability of the models to simulate O 3 near the surface is tested with rural AirBase and EMEP stations (see Sect. 3.2). Figure 4 shows monthly means and Fig. 5 the average diurnal cycle in different season in Europe. All runs underestimate monthly mean O 3 in spring and winter and overestimate it in late summer and autumn. The overestimation in summer was largest in MOZ. While the overestimation appeared also 15 with respect to the ozonesondes in LT (see Fig. 2, left) the spring time underestimation was less pronounced in LT.
The comparison of the diurnal cycle with observations ( Fig. 5) shows that C-IFS produced a more realistic diurnal cycle than the MOZART model. The diurnal variability simulated by the MOZART model is much less pronounced than the observations sug-20 gest. The diurnal cycle of C-IFS and REAN were similar. This finding can be explained by the fact that C-IFS and REAN use the IFS diffusion scheme whereas MOZART applies the diffusion scheme of the MOZART CTM.
The negative bias of C-IFS in winter and spring seems mainly caused by an underestimation of the night time values whereas the overestimation of the summer and 25 autumn average values in C-IFS were caused by an overestimation of the day time values. However, the overestimation of the summer night time values by MOZART seems to be a strong contribution to the average overestimation in this season.

Carbon monoxide
The seasonality of CO is mainly driven by its chemical lifetime, which is lower in summer because of increased photochemical activity. The seasonal cycle of the CO emissions plays also an important role in the case of biomass burning and high anthropogenic emissions. The global distribution of total column CO retrieved from MOPITT 5 and from AK weighted columns simulated by C-IFS, MOZ and REAN is shown for April 2008 in Fig. 6 and for August in Fig. 7. April and August have been selected because they are the months of the NH CO maximum and minimum. C-IFS reproduced well the observed global maxima in North-America, Europe and China as well as the biomass burning signal in Central Africa. However, there was a widespread underestimation of the MOPITT values in the NH, which was strongest over European Russia and Northern China. Tropical CO was slightly overestimated but more strongly over Southeast Asia in April at the end of the biomass burning season in this region. The lower CO columns in mid-and high latitudes in the Southern Hemisphere (SH) were underestimated. The same global gradients of the bias were found in MOZ and REAN. 15 The negative NH bias in April of MOZ is however more pronounced but the positive bias in the tropics is slightly reduced. The bias of MOZ seems stronger over the entire land surface in NH and not predominately in the areas with high emission. This is consistent with the finding of Stein et al. (2014) that dry deposition, besides underestimated emissions, contributes to the large negative biases in NH in MOZ. Assimilating MOPITT 20 (V4) in REAN led to much reduced biases everywhere even though the sign of bias in NH, Tropics and SH remained. In August, the NH bias is reduced but the hemispheric pattern of the CO bias was similar as in April for all runs. The only regional exception from the general overestimation in the tropics is the strong underestimation of CO in the biomass burning maximum in Southern Africa, which points to an underestimation 25 of the GFAS biomass burning emissions in that area. More insight in the seasonal cycle and the vertical CO distribution can be obtained from MOZAIC aircraft profiles. CO profiles at Frankfurt (Fig. 8, left)  record with about 2-6 observations per day. As already reported in Inness et al. (2013) and Stein et al. (2014), MOZ underestimates strongly LT CO with a negative bias of 40-60 ppb throughout the whole year. The highest underestimation occurred in April and May, i.e. at the time of the observed CO maximum. C-IFS CB05 also underestimates CO but with a smaller negative bias in the range of 20-40 ppb even though 5 it used the same CO emission data as MOZ. REAN has the lowest bias throughout the year but the improvement is more important in winter and early spring. The comparison over London, which is representative for 04:00 and 22:00 UTC leads to similar results as for Frankfurt (Fig. 8, middle). The outcome of the comparison with LT CO from MOZAIC is consistent with the model bias with respect to the GAW surface ob-10 servations in Europe, predominantely located in the Alpine region, shown in Fig. 10.
The seasonal variability of LT CO from MOZAIC and the model runs in North-America is very similar to the one in Europe (Fig. 8, right). The late winter and spring bias is slightly increased whereas the summer time bias was lower for all models. The surface bias in winter and spring of MOZ, C-IFS and REAN is about −50, −40 and −20 ppb 15 respectively. In the rest of the year REAN and C-IFS have a bias of about −15 ppb whereas the bias of MOZ is about twice as large.
MT CO was very well produced by REAN in Europe and North-America probably because MOPITT has the highest sensitivity at this level. The MT bias of C-IFS is about 75 % of the bias of MOZ, which underestimates by about 30 ppb. In the UT the 20 CO biases are for all models mostly below 10 ppb, i.e. about 10 %. C-IFS has overall the smallest CO bias whereas REAN tends to overestimate and MOZ to underestimate CO over Europe and North America. CO observed by MOZAIC over Windhoek (Fig. 9, middle) has a pronounced maximum in September because of the seasonality of biomass burning in this region. Al-25 though all runs show increased CO in this period, the models without assimilation were less able to reproduce the high observed CO values and are biased low up to 40 ppb in LT and MT. Biases were much reduced, i.e. mostly within 10 ppb, during the rest of the year. The assimilation in REAN greatly reduces the bias in the biomass burning period. In UT C-IFS had slightly smaller biases of about 10 ppb than MOZ and REAN. A less complete record of the seasonal variability is available for Caracas (Fig. 9, left). All models tend to underestimate UT and MT CO maxima in April by about 20 % but in contrast to Windhoek the C-IFS and not REAN has the smallest bias in LT. Hyderabad (Fig. 9, right) is the only observation site were a substantial overestimation of CO in 5 LT and UT is present even though the observations are in the range of 150-250 ppb, which is mostly higher than at any of the other airports discussed. All models overestimate the seasonality because of an underestimation in JJA and an overestimation during the rest of the year. 10 The global maxima of NO 2 are located in areas of high anthropogenic and biomass burning NO emissions. The global annual distribution of annual tropospheric columns retrieved from the GOME-2 instrument and simulated by the models is shown in Fig. 11. C-IFS, MOZ and REAN showed a very similar distribution, which can be explained by that fact that the same NO emission data were used in all runs. The global patterns of 15 the modelled fields resemble the observed annual patterns to a large extent. But the models tend to underestimate the high observed values in East-Asia and Europe and also simulate too little NO 2 in larger areas of medium observed NO 2 levels in Asia and Central Africa as well as in the outflow areas over the West-Atlantic and West Pacific Ocean. This could mean that NO emissions in the most polluted areas are too low but 20 also that the simulated lifetime of NO 2 is too short. The validation of the seasonality of NO 2 (Fig. 12) for different regions and months shows that tropospheric NO 2 columns over Europe, North America, South Africa and East-Asia are reasonably reproduced. The models tend to underestimate tropospheric columns over Europe in summer (see Table 2 for area descriptions). However, the eval- 25 uation with GAW surface stations mainly from Central and Eastern Europe (Fig. 13) revealed an overestimation by all models in winter and a small overestimation in summer for REAN and C-IFS. All runs significantly underestimate the annual cycle of the 7763 Introduction

HCHO
On the global scale HCHO is mainly chemically produced by the oxidation of isoprene and CH 4 . Isoprene is emitted by vegetation. On the regional scale HCHO emissions from anthropogenic sources, vegetation and biomass burning also contribute to the HCHO burden.

10
The annual average of tropospheric HCHO retrieved from GOME-2 and from the model runs is shown in Fig. 14 Figure 15 shows model time series of tropospheric HCHO against corresponding GOME-2 satellite retrievals for selected regions. The models underestimated satellite values over East-Asia especially in summer and overestimate HCHO columns for 20 Indonesia ( Fig. 16 shows the median of weekly observed and modelled time series. REAN 5 and MOZ greatly exaggerated the seasonal cycle since the values in winter were up to eight times larger than the median of the observations. The summer values of the two runs were about 50 % higher than the observations. C-IFS followed better the weak seasonality of the observations but suffered from a nearly constant bias of about 1 ppb (100 %), which was much smaller than the bias of REAN and MOZ in winter but slightly 10 higher in summer. Overall, the on-line integration of C-IFS showed lower SO 2 biases. As no SO 2 observations were assimilated in REAN and identical SO 2 emission were used, the differences between the runs were caused by differences in the simulation of vertical mixing, sulphur chemistry and wet and dry deposition in C-IFS and MOZART. The winter time bias of REAN and MOZ could be introduced by the diffusion scheme 15 in MOZART.

Computational cost
The computational cost is an important factor for the operational applications in CAMS. The computational cost of different configurations of IFS, C-IFS and IFS-MOZART are given in Table 3. Computational cost is expressed in billing units (BU) of the ECMWF 20 IBM Power 7 super-computer. BUs are proportional to the number of used Central Processing Unit (CPU) times the simulation time.
The increase of cost because of the simulation of the CB05 chemistry with respect to an NWP run is about a factor 4 at the resolutions T159 (110 km), T255 (80 km) and T511 (40 km). C-IFS (CB05) is about 8 times more efficient than the coupled system requirements of the IFS at higher resolution, or also in data assimilation mode. The additional resources allocated to the IFS are however mostly latent as the coupled MOZART model and the coupler software could not be made faster by using more resources. C-IFS with the MOZART chemical mechanism, i.e. the same chemistry scheme as in 5 IFS-MOZART, is about 2 times and C-IFS with RACMOBUS 7 times more costly than C-IFS (CB05) at a T159 resolution. Both the MOZART and the RACMOBUS schemes encompass a larger number of species and reactions and include a full stratospheric chemistry scheme, which is missing in CB05. 10 Modules for the simulation of atmospheric chemistry have been implemented on-line in the Integrated Forecasting System (IFS) of ECMWF. The chemistry scheme complements the already integrated modules for aerosol and greenhouse gases as part of the IFS for atmospheric composition (C-IFS). C-IFS for chemistry replaces the coupled system IFS-MOZART for forecast and assimilation of reactive gases within the 15 pre-operational Copernicus Atmosphere Monitoring Service. C-IFS applies the chemical mechanism CB05, which describes tropospheric chemistry with 55 species and 126 reactions. C-IFS benefits from the detailed cloud and precipitation physics of the IFS for the calculation of wet deposition and lightning NO emission. Wet deposition modelling is based on  and accounts for the 20 sub-grid scale distribution of clouds and precipitation. Dry deposition is modelled using pre-calculated monthly-mean dry deposition velocities following (Wesely, 1989) with a superimposed diurnal cycle. Surface emissions and dry depostion fluxes are applied as surface boundary condtions of the diffusion scheme. Lightning emissions of NO can be calculated either by cloud height (Price and Rind, 1993)  anthropogenic emissions were taken from the MACCity inventory and biomass burning emissions from the GFAS data set for 2008. An evaluation for the troposphere of a simulation in 2008 with C-IFS (CB05) and the MOZART CTM (MOZ) as well as with the MACC re-analysis (REAN) was carried out. The model results were compared against ozonesondes, MOZAIC CO aircraft profiles, 5 European surface observations of O 3 , CO, SO 2 and NO 2 and global satellite retrievals of CO, NO 2 and HCHO. The evaluation showed that C-IFS preforms better or with similar accuracy as MOZART and mostly of similar quality as the MACC re-analysis. It should be noted that satellite retrievals of CO, O 3 and NO 2 were assimilated in the MACC re-analysis to improve the realism of the fields simulated by IFS-MOZART. 10 In comparison to MOZ, C-IFS (CB05) had smaller biases (i) for CO in the Northern Hemisphere, (ii) for O 3 in the upper troposphere and (ii) for winter-time SO 2 at the surface in Europe. Further, the diurnal cycle of surface O 3 , tested with rural European Air quality observations, showed greater realism in the C-IFS simulation. As both models used the same emission data, the improvements can be explained by the differences 15 in the chemical mechanism and the simulation of wet and dry deposition. However, the improvements in SO 2 and the diurnal cycle of O 3 are most probably caused by the more consistent interplay of diffusion and sink and sources processes in the on-line integrated C-IFS.

Summary and outlook
There is still room for improvement of C-IFS (CB05). It underestimated surface O 3 20 over Europe and North America in spring and overestimated it in late summer and autumn. CO was still underestimated by C-IFS in particular in Europe and North America throughout the year but more in spring and winter, and in the biomass burning season in Africa. Winter time tropospheric NO 2 over China as retrieved from the GOME-2 instrument was two times higher than the fields modelled by C-IFS, MOZART and the 25 MACC re-analysis. Although only one chemical mechanism is described in the paper, C-IFS is a model that can apply multiple chemistry schemes. The implementation of the chemistry schemes of the CTMs MOCAGE and MOZART has technically been completed but further optimisation and evaluation is required. Both schemes offer a description of stratospheric chemistry, which is not included in the tropospheric scheme CB05. For this reason it is intended to combine the CB05 mechanism with the BASCOE stratospheric mechanism. An inter-comparison of the performance of the different chemical mechanism is planned.

5
It is foreseen to further improve the link between the physics and chemistry packages in IFS. For example, the detailed information from the IFS surface scheme will be utilised for the calculation of dry deposition and biogenic emissions. A first important step is to replace the climatological dry deposition velocities with-online calculated values. Further, the impact of the simulated O 3 fields, once the stratospheric chemistry is fully implemented, on the IFS radiation scheme and the corresponding feedback on the temperature fields will be investigated.
Another ongoing development is to link more closely the greenhouse gas, aerosol and gas-phase chemistry modules of C-IFS. Relevant chemical conversion terms can already be fed to the GLOMAP aerosol (Mann et al., 2010) module for the simulation of 15 secondary aerosols. The calculation of photolysis rates can account for the presence of aerosols, and HO 2 uptake on aerosols can be simulated .

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version