Coupling global chemistry transport models to ECMWF’s integrated forecast system

The implementation and application of a newly developed coupled system combin-ing ECMWF’s integrated forecast system (IFS) with global chemical transport models (CTMs) is presented. The main objective of the coupled system is to enable the IFS to simulate key chemical species without the necessity to invert the complex source 5 and sink processes such as chemical reactions, emission and deposition. Thus satellite observations of atmospheric composition can be assimilated into the IFS using its 4D-VAR algorithm. In the coupled system, the IFS simulates only the transport of chemical species. The coupled CTM provides to the IFS the concentration tendencies due to emission 10 injection, deposition and chemical conversion. The CTMs maintain their own transport schemes and are fed with meteorological data at hourly resolution from the IFS. The CTM used in the coupled system can be either MOZART-3, TM5 or MOCAGE. The coupling is achieved via the special-purpose OASIS4 software. The scientiﬁc integrity of the coupled system is proven by analysing the di ﬀ erence 15 between stand-alone CTM simulations and the tracer ﬁelds in the coupled IFS. The IFS concentration ﬁelds match the CTM ﬁelds for about 48 h with the biggest di ﬀ erences occurring in the planetary boundary layer (PBL). The coupled system is a good test bed for process-oriented comparison of the coupled CTM. As an example, the vertical structure of chemical conversion and emission injection is studied for a ten day period 20 over Central Europe for the three CTMs.


Introduction
Routine exploitation of space-born observations of the atmosphere has been a major contribution to the improvements in numerical weather prediction (NWP) over the last three decades. Inspired by the success of satellite data assimilation in NWP, the Introduction The idea of the coupled system is that the IFS computes only the transport of the reactive gases while the tendencies due to chemical conversion, deposition and emission injection are provided by one of the coupled CTM. The CTM itself is driven by meteorological data which are transferred at high temporal resolution from the IFS to the CTM. For example, the call of a subroutine for chemical conversion in an integrated 5 chemistry-global-circulation-model code is substituted in the coupled system by a call to the coupler software requesting the respective total tendencies from the CTM. The tendencies are then applied to the concentration fields in the IFS at every time step to account for the local concentration changes.
The motivation to build a coupled system, rather than directly integrating the respec- 10 tive modules in the IFS code is elaborated in Sect. 2.1. Key reasons were the flexibility to apply more than one CTM for the provision of sink and source terms and the reduced development effort. A potential problem of the coupled approach is that the chemistry and deposition tendencies applied to the IFS concentrations are calculated using the concentration fields 15 that were calculated in the coupled CTM, which uses its own transport, convection and diffusion scheme. The transferred CTM tendencies can therefore be dislocated from the IFS concentration fields to which they are being applied. The dislocation can occur because of (i) the transformation between the CTM and IFS model grids by the coupler software, (ii) the differences between the concentration fields of the CTM and of the Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion ferent application modes and the computational performance are described in Sect. 2. Section 3 comprises the test of the scientific integrity of the system, which investigates the impact of dislocation. Also included in this section is a comparison of the vertical structure and magnitude of tendencies due to emission injection and chemical conversion over Europe, which helps to give a better understanding of characteristics of the 5 three CTMs. A summary concludes the paper.
2 Description of the coupled system

Motivation for the design of a two-way coupled system
An extension of an earth-system model can follow two approaches: (i) directly integrating subroutines or modules in one unified model or (ii) coupling independent models by means of dedicated coupler software. Direct integration -often referred to as online coupling -is normally pursued when complex chemistry schemes are included in meteorological models. Examples of the on-line integration of chemistry modules in weather forecast models are GEM-AQ (Kaminski et al., 2008), GEMS-BACH (Ménard et al., 2007), WRF/Chem (Grell et al., 2005)  Interactive Discussion developed because of the envisaged high development cost to integrate and test complex chemical mechanisms as an integrated part of the IFS. The benefits from using ECMWF's operational data assimilation system and the associated infrastructure for observation processing would be difficult to keep if a new data assimilation system would be build around an existing CTM. Another advantage of the coupled system is 5 the possibility to easily couple different CTMs to the IFS and therefore to be more flexible in the choice of the applied chemistry schemes. A coupled system of independent components can also better benefit from the ongoing development work of the standalone versions of the CTMs since the CTMs stay independent models. Finally, this approach allows for different grid resolutions in the IFS and CTM so that computing 10 resources can be optimally used.
Although designed with the prospect of data assimilation, the coupled system can also be considered as an efficient way to provide meteorological parameters to a CTM at high temporal resolution without the need to access such fields from disk files. Furthermore, it is a research platform to (i) compare the vertical transport schemes of the 15 CTMs and that of the IFS, (ii) to inter-compare the chemical mechanisms of the CTMs by analyzing the tendency field due to chemistry (see Sect. 3.3) and (iii) to explore the impact of atmospheric composition on numerical weather prediction and its feedback to the tracer concentrations. 20 The coupled system is a three-dimensional two-way coupled system consisting of the IFS, one of the CTMs MOZART-3, TM5 and MOCAGE and the coupler software OA-SIS4. In the coupled system, the IFS simulates the advection, vertical diffusion and convection of selected chemical tracers (CO, NO x , HCHO, SO 2 and O 3 ) and applies tracer tendencies calculated by the coupled CTM to account for sink and source pro-25 cesses such as emission, chemical conversion and deposition. The prognostic tracer variables are also part of the control variables of the data assimilation mode in IFS. Figure 1 shows a schematic of the data flow in the coupled system. Every hour, the 768 Introduction Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion IFS provides meteorological fields to drive the CTMs and receives the CTM tendencies. Depending on the mode of operation (see below), concentration fields are exchanged from the IFS to the CTM or vice versa at the start of each coupled run in order to provide the initial conditions for the coupled run. The details of the application and formulation of the tendency terms are given in Sect. 2.3.

5
The choice of the exchanged meteorological fields depends on the requirements of the CTM. All CTMs receive fields of humidity, temperature, wind components, or divergence and vorticity in spectral representation, and sensible and latent heat flux. MOZART-3 and MOCAGE simulate their own hydrological cycle whereas TM5 also requires the IFS's precipitation and cloud fields, surface properties and convective mass fluxes.
The experiments with the coupled system are organized as a sequence of several 12 h runs in data assimilation mode or, in forecast mode, as runs up to four days starting every 24 h at 00:00 UTC. In data assimilation mode, the length of the coupled run is given by the length of the 4D-VAR assimilation window, which is normally 12 h. A more 15 detailed description of how the coupled system is used in data assimilation is given in Inness et al. (2009). In forecast mode, the meteorological fields in the IFS need to be initialized at least every 24 h with a meteorological analysis in order to avoid a drift from the observed state of the atmosphere.
At the start of each coupled run, the initial conditions of the coupled tracers in the IFS 20 and the CTM are set to the same values: either the CTM fields replace the IFS tracer initial conditions fields (CTM-IC mode) or the IFS tracer fields replace the respective initial conditions of the CTM (IFS-IC mode).
In CTM-IC mode, the CTM gets the whole set of initial conditions from the previous CTM run. In this configuration the concatenated CTM output of IFS-CTM is equivalent 25 to the normal continuous CTM off-line run, except for the higher exchange frequency of meteorological fields.
In IFS-IC mode, CTM fields are used as IFS initial conditions only at the very first coupled run. In all subsequent runs, the IFS's coupled tracers are initialized from the previous IFS run and the respective CTMs fields are replaced by the IFS fields. The un-coupled CTM species are initialised from the previous CTM run as in CTM-IC mode. The IFS-IC mode is applied in data assimilation because the IFS tracer fields contain the assimilated information of the observations. The IFS-IC mode can also be applied to impose vertical tracer profiles simulated by the IFS on the CTM.

Formulation of tendency terms and their application in the IFS
The exchange of concentration tendencies is a unique feature of the coupled system. The formulation of the tendency terms has to reflect (i) operator splitting and timestepping in both the CTMs and the IFS, (ii) relative size and spatial structure of the tendency fields, and (iii) the computational cost of the exchange.

10
The CTMs use an operator-splitting approach in which advection, chemistry, emission injection, diffusion and deposition are called in sequence, and the concentrations are updated directly within each operator subroutine. The IFS computes semilagrangian advection, diffusion and convection of the tracers based on unperturbed concentration field values from the previous time step (Beljaars at al., 2004) and up-15 dates the concentration values with the accumulated tendency of all sink and source processes at the end of the time step.
The total CTM tendencies T [kg kg −1 s −1 ] are given by the sum of chemical loss L C and production P C , production due to emission injection P E and loss L D due to deposition: The injection of surface emissions is integrated in the MOZART-3 diffusion scheme, whereas TM5 and MOCAGE distribute the injected mass in a fixed ratio over selected layers near the surface and apply their diffusion operator after the injection. P E is therefore, at least for MOZART-3, a combination of the emission injection and the tendencies 25 due to vertical turbulent diffusion. Since P E already contains the diffusion tendencies, its application in the IFS requires that the IFS diffusion must not be applied again to 770 Introduction

Tables Figures
Back Close

Printer-friendly Version
Interactive Discussion the respective tracer fields. In order to also use the IFS diffusion scheme for the tracer transport within the coupled system, the effective net surface flux Φ E−D from emissions and dry deposition is determined by calculating the total columns of the surface contribution of P E and the fraction of L D representing dry deposition. Φ E−D is then presented as a surface flux to the IFS diffusion scheme and the components P E and L D are excluded from T leaving T Air . Deposition L D and chemical loss L C are almost always proportional to the tracer concentration x. A relative formulation L=l x, i.e. a loss rate l , would have linked tendency and concentration values and would have helped to avoid negative concentrations after the application of the CTM tendencies in the IFS. However, it was decided against the 10 relative formulation of tendencies because (i) it would have been be more difficult to distinguish chemical loss and production from the output arguments of the chemistry routines, which directly only provide the total change, and (ii) because a separate interpolation of production and loss tendencies, which often almost compensate each other, could have caused imbalances when the two fields are combined again in the 15

IFS.
After consideration of the above arguments, it was decided to transfer and apply the process-specific tendencies of the CTM in one of the following two modes: 1. one three-dimensional tendency field T containing all sources and sinks as well as diffusion (total-tendency mode) 20 2. one three-dimensional tendency field T Air and the effective Φ E−D surface flux of emission and deposition (surface-flux mode)

CTM and IFS specifications
In the coupled system, the IFS runs with a T159 spectral resolution and the grid point space is represented by the reduced Gaussian grid (Hortal and Simmons, 1991), which Interactive Discussion a bi-linear mode is applied. The IFS runs -for most parts of the globe -at a finer horizontal resolution than the CTMs because this improves (i) the quality of the meteorological forecasts and (ii) the acceptance of high resolution observations within data assimilation mechanism.
The IFS and all CTMs use the same vertical discretization of 60 hybrid sigma-5 pressure levels, reaching up to 0.1 hPa. The use of an identical vertical structure in the IFS and the CTM avoids the need for vertical interpolation. The minimum coupling interval is 3600 s which is the largest acceptable time step for the IFS at a T159 resolution, and also the time step of some of the CTMs. An overview of the CTM resolution and parameterisations is given in Table 1.

Computational performance of the coupled system
A draw back of the coupled approach, in contrast to the online integration of chemistry modules in the IFS, is the reduced computational performance because of the longer CTM run-time and to a smaller extent also because the overhead introduced by the coupling. Whereas the IFS can be run highly efficient, the CTM run time on 15 ECMWFs high performing facility could not be reduced to match that of the IFS. The run time of the CTM determined therefore the run-time of the coupled system and caused considerable latency in particular in data assimilation applications. In data assimilation mode the IFS has to use a larger amount of resources because of memory constraints. Table 2 lists the run time of the coupled system for a 24 h forecast, for 20 a 12 h data assimilation cycle and the respective average turn over (simulation days per wall clock day). The turnover also reflects the time spend to archive the results and is highly depended on the supercomputer workload. Because of the long run-time MOCAGE was only applied for a couple of forecast runs. TM5s run has been improved to match that of MOZART-3 only in the last year of the projects. Therefore MOZART Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion 3 Scientific integrity of the coupled system

Comparing IFS-CTM concentrations with CTM concentrations
The application of the CTM tendencies to IFS tracers is an approximation because the underlying CTM concentrations could be dislocated from the concentration patterns in the IFS. This dislocation may occur because of (i) the horizontal interpolation from the 5 CTM to the IFS grid and (ii) the differences between the CTM and the IFS transport.
In the case of the coupled system, both the IFS and the CTM simulate atmospheric transport processes. Different advection schemes or spatial and temporal resolutions may lead to different concentration fields in the IFS and the CTM. The most severe consequence of the dislocated tendencies would be negative con- 10 centration values in the IFS because of unbalanced loss processes. The severity of the impact of the dislocation depends on the sensitivity of the sink and sources on the concentration itself, i.e. the speed of the chemical conversion and the intensity of the deposition.
In order to minimise the dislocation, the CTM and the IFS concentration fields have 15 to be made as similar as possible by not letting the transport schemes develop different concentration patterns and by periodically aligning the concentration fields in the IFS to the ones in the CTM or vice versa. This aligning is ensured by letting the coupled tracers in the IFS and the CTM start from the same initial conditions (see Sect. 2.2) either in CTM-IC mode or IFS-IC mode.

20
The integrity of the coupled system depends on whether the application of external tendency fields accounting for processes not included in the IFS (chemistry, emission and deposition) gives reasonable results for the forecast length. The objective is that the IFS is able to imitate the CTM concentration changes. Therefore, the development of the differences between the IFS tracer fields and their counterpart in the CTM in 25 coupled runs starting from the same MOZART-3 initial conditions was studied. The resemblance of the fields was carefully checked and no unreasonable features in the IFS fields were discovered. The only obvious problem occurred during an earlier attempt 773 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion to couple NO 2 rather than NO x . The stratospheric NO-NO 2 conversion at sunrise and sunset, which moves around the globe, could not satisfactorily be captured by the coupled system with an exchange frequency of one hour. To avoid this problem, it was decided to use NO x as a coupled species, which did not show the stripe-like undulation as a consequence of the constantly progressing terminator seen in the NO 2 -fields.

5
To demonstrate the accuracy of the tendencies application method, Fig. 2 shows exemplary time series of spatially averaged O 3 , CO, HCHO and NO x concentrations from the coupled system IFS-MOZART for model layer 55 (about 240 m above the surface) over Europe. Shown are the two modes of the tendency application (total-tendency and surface-flux mode, see Sect. 2.2) as well as no tendency application. When no source 10 and sink tendencies were applied the averaged IFS tracer quickly diverged from the CTM reference showing the general need for the tendency application also in a time scale of a few hours. When total tendencies were applied, the IFS could imitate the CTM up to a forecast length of 48 h. The differences were larger, in particular for primary species, when the IFS vertical diffusion scheme injected the effective surface flux 15 Φ E−D , indicating a stronger diffusion in the IFS.
To gain a more detailed overview of the ability of the IFS tracers to follow the CTM concentration fields, the relative difference between IFS and CTM fields were calculated for each model grid point and forecast hour in both modes of the tendency application (see Sect. 2.2). The relative differences were obtained by normalizing with 20 the range, i.e. the difference between maximum and minimum value of the CTM concentration in the respective atmospheric region because it prevents the normalisation with concentration values close to zero. Table 3 contains the percentage of grid points with relative differences lower than 1, 10 and 100%, discriminating between the PBL, the free troposphere and the stratosphere for the "surface-flux" mode. In the PBL the 25 differences in "total-tendency" mode were smaller than the differences in "surface-flux" because the CTMs diffusion tendencies are directly used in the IFS. The differences were of the same size in the rest of the atmosphere.
The discrepancy between the IFS and the CTM coupled tracers developed quickly 774 Introduction

Tables Figures
Back Close

Printer-friendly Version
Interactive Discussion after the first data exchange and increased from there onwards much more slowly with increasing forecast length. But even after 24 h the differences were less than 10% at more than 97% of the grid points for every species. When studying the more strict error limit of discrepancies less than 1%, only about 30% of the O 3 in the PBL could be simulated by the IFS with this accuracy, whereas for the other species 80-90% of 5 the grid points satisfied this criterion. The largest absolute differences occurred in the PBL, indicating the high variability in this part of the atmosphere because of emissions injection, diffusion and active chemistry.

Comparing IFS-CTM concentrations with observations
In this section it will only be studied if the dislocated CTM and the IFS fields compare 10 differently against observations. An comprehensive evaluation of the coupled system will be presented in follow-up papers. The performance of the coupled stand-alone CTMs has already been published in the papers cited in Sect. 1. The simulation of IFS-MOZART as part of a pre-operational forecast were compared against surface observation from the Global Atmopshere Watch network at the stations The RMSE of three-hourly data (see Fig. 6) was of very similar value for most of the stations, indicating no significant difference in the performance of the coupled system 20 compared to the CTM stand-alone run. The largest differences occurred at the station CPT where O 3 was better simulated by the IFS in the coupled system and CO better by the stand-alone CTM. Figures 7 and 8 show time series of the observation and the simulation of the coupled system and the MOZART-3 direct output at CPT. There was a minimal offset, which was small compared to the bias against the observations, be-25 tween the coupled system and the direct CTM output, which caused a higher or lower RMSE for CO and O 3 , respectively. The height of the peaks was higher in the coupled model output which seems to better match the observations. It was inferred that the 775 Introduction

Tables Figures
Back Close

Printer-friendly Version
Interactive Discussion differences between the IFS and CTM fields were mainly caused by the different horizontal grid partitioning in the IFS and the CTM, which attributes a different amount of emission in the grid box where the observation is located. Given the overall uncertainty of the concentration values and the anticipated changes due to data assimilation, it was concluded that IFS concentration fields were scientifi-5 cally sound since they resembled the CTM fields to a high degree.

Model diagnostics based on the tendency terms
Studying the source and sink tendencies from the emission injection and chemical conversion may help to gain more insight into the CTMs. In a case study, the tendency terms from the different source and sink processes were analysed with emphasis on 10 the troposphere over all 287 land points in Central Europe (42 for a ten day period in June 2004. Vertical profiles of the area-averaged concentration changes from each CTM were calculated for day (12:00 UTC) and night conditions (00:00 UTC) and, for display, normalised with the area-averaged concentrations. The "chemistry" profile includes the net 15 chemical conversion and the negligible contributions from wet deposition and air-borne emissions (T Air , see Sect. 2.3.) The "emission" profile comprises the three-dimensional tendencies due to emission injection, diffusion, convection and dry deposition, whose total columns were used to calculate the net surface flux Φ E−D in "surface-flux" mode. The sum of the two is equal to T in total-tendency mode. 20 The CO tendencies for emission injection and diffusion (Fig. 3, left) during the day showed that diffusion, despite CO emissions, leads to a concentration decrease close to the surface and an accumulation in the upper part of the boundary layer. The accumulation zone in MOZART-3 was largely confined to 900 hPa whereas the vertical transport in TM5 and MOCAGE reached higher levels, indicating more pronounced dif-25 fusion and convection. The stable conditions during the night caused increasing CO concentration values only in the lowest two to three model levels in all CTMs. The chemical conversion of CO (Fig. 3, right)  Interactive Discussion explains the absence of concentration changes during the night for all models. The relative concentration changes due to chemistry were smaller than the changes due to emissions and diffusion. However, chemical CO loss occurred throughout the vertical column of the troposphere. All models simulated the CO depletion due to oxidation with OH in the free troposphere. In spite of similar formulations of the chemical rate constant 5 for CO oxidation, the relative chemical tendency of CO among the three CTMs differs by more than a factor of two. MOCAGE showed the strongest chemical loss both in relative and absolute terms. A comparison of the OH concentrations of the three CTMs confirmed that MOCAGE's average OH concentrations were higher by about 0.05 ppt than TM5 and by 0.07 ppt than MOZART-3. The CTMs simulated a net chemical pro-10 duction of CO due to oxidation of volatile organic compounds in the PBL, which was smallest in TM5 and largest in MOZART-3. The vertical profile of the surface flux related tendencies for NO x resembled that of CO although the relative changes were about ten times larger (Fig. 4, left). The mixing of the emissions during the day was again limited to a shallower layer in MOZART-3 15 compared to the other CTMs. The chemistry (Fig. 4, right) caused a loss of NO x in the lower troposphere of up to 40% per hour during the day because of conversion into HNO 3 and PAN. Again MOCAGE simulated the strongest tropospheric NO x depletion during the day because of the higher OH concentrations. During the night, only TM5 and MOZART-3 computed tropospheric NO x loss, in the range of 10%, probably due 20 to the heterogeneous N2O5 uptake on clouds and aerosols, which is not included in MOCAGE.
The O 3 surface flux (Fig. 5, left) is caused by dry deposition at the surface. Compared to this large loss, the averaged diffusion did not contribute substantially to a systematic vertical concentration change in any of the CTMs. During the day, O 3 production oc-25 curred in the PBL of all CTMs, and O 3 loss occurred in the lowest layer during the night because of titration with NO, which was concentrated there. The comparatively low O 3 loss can be attributed to the fact that most of the titration took place before midnight. Only TM5 simulated reduced O 3 production in the lowest layer during the day, which GMDD Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion was probably related to the high NO x increase there.

Summary and conclusions
The design and the validation of a coupled system which links the ECMWF's Integrated Forecast and data assimilation System (IFS) to each of the three global CTMs, MOZART-3, TM5 and MOCAGE, are presented in this paper. The purpose of the cou-5 pled system is to enable the IFS to simulate global reactive gases in order to provide forecasts and analyses of atmospheric composition without directly integrating complex chemistry schemes, emission injection and deposition into the IFS. The coupled system is an alternative approach to the on-line integration of chemistry-schemes in meteorological models. The main motivation for developing the coupled system was the ease with which different chemical schemes could be tested, and the reduced development effort. The coupled system IFS-CTM can directly utilise the IFS 4D-VAR algorithm to assimilate observations of atmospheric trace gases such as CO, tropospheric and stratospheric O 3 , SO 2 , NO x and HCHO. This paper focuses on the ability of the coupled IFS to simulate sound concentration fields by comparing them to the 15 concentration fields of the coupled CTM, which they should closely resemble. In the coupled system, the CTM is driven by the meteorological data received from the IFS. The special characteristic of the coupled system is that the IFS receives either three-dimensional tendencies accounting for all source and sink processes or threedimensional tendencies due to chemistry and net surface fluxes accounting for emis-20 sion and dry deposition. The respective tendencies and fluxes are applied to the IFS concentration fields, whose transport has is modelled by the IFS.
To prove the validity of the coupled approach, the chemical tracers in the coupled system IFS-MOZART were compared with concentration fields from MOZART-3. Only small differences were found for a period of about 48 h. The largest differences oc-25 curred in the PBL. A comparison with observations from the GAW network showed that these small differences lead to sometimes slightly bigger and sometimes smaller GMDD Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion errors with respect to observations. The exchange of tendencies describing emission injection and deposition as well as chemical conversion is a special feature of the coupled system. The tendencies were used for a process-oriented inter-comparison of the three CTMs, MOZART-3, TM5 and MOCAGE, over Europe in June 2004. Averaged profiles for day and night conditions 5 showing the impact of the surface fluxes (i.e. emissions, deposition. diffusion) were compared with profiles showing the impact of the chemical conversion. NO x was the most variable species with average changes per hour of up to 30% of the concentration value for both chemistry and emission injection. The maximum relative O 3 changes were due to chemistry and reached up to 5% in the PBL. Emissions of CO caused an 10 increase of up to 3% and its chemical depletion was up to 1% of the concentration value per hour. Even with such a large variability, NO x could be reasonably well simulated with the coupled system. Despite the surface emission, diffusion caused a net loss close to the surface and an accumulation in the upper part of the PBL in all CTMs during the day. The day-time vertical mixing was shallowest in MOZART-3. The chemical 15 loss of CO and NO x linked to the reaction with the hydroxyl radical was highest in MOCAGE. In contrast MOCAGE's night time NO x depletion was much lower than that of the other two CTMs, with TM5 simulating the largest chemical NO x loss. MOCAGE's dry deposition of O 3 was confined to the lowest model layer and was stronger than in the other CTMs. 20 The coupled system IFS-MOZART-3 has been applied to provide forecasts of atmospheric composition since May 2007. The results of the O 3 and CO forecast are published daily at the GEMS web-site http://gems.ecmwf.int/d/products/grg/realtime/. Besides providing a global picture of atmospheric composition, the results of the NRT forecast are used as boundary conditions for European regional air quality models run 25 daily within the GEMS project.
In summary, the IFS tracer fields of the coupled system compared well with the corresponding CTM fields and with CO and O 3 observations. It can be concluded that the coupled system is a flexible and scientifically sound instrument for the forecast of   Louis (1979) for free troposphere Chemical mechanism JPL-03 and JPL-06 CBM4 scheme as described in REPROBUS (Lefèvre et al., 1994) (Sander et al., 2003 as described Houweling et al. (1998) Fig. 3. Profile of the averaged changes per hour in percent of CO over Europe due to emissions injection and deposition (left two panels) and due to chemistry (right two panels) from TM5, MOZART-3 (MOZ) and MOCAGE (MOC) at 12:00 and 24:00 UTC.