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., 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 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 (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.
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), and forecast and assimilation of
the stratospheric ozone (O3) (Flemming et al., 2011; Lefever et
al., 2014), tropospheric carbon monoxide (CO) (Elguindi et al., 2010)
and O3 (Ordóñez et al., 2010). Coupled system IFS-TM5 has
been used in a case study on a period with intense biomass burning in Russia
in 2010 (Huijnen et al., 2012). Nevertheless, the coupled approach has
limitations such as the need for interpolation between the IFS and CTM model
grids and the duplicate simulation of transport processes. Furthermore, its
computational performance is often not optimal as it can suffer from load
imbalances between the coupled components.
Consequently, modules for atmospheric chemistry and related physical
processes have now been integrated on-line in the IFS, thereby complementing
the on-line integration strategy already pursued for aerosol and greenhouse
gases in IFS. The 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 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 preparation of CAMS.
Including chemistry modules in general circulation models (GCM) to simulate
interaction of stratospheric O3 (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.
Fiore et al., 2012). In the UK Met Office's Unified Model (UM), stratospheric
chemistry (Morgenstern et al., 2009) and tropospheric chemistry (O'Connor et
al., 2014) can be simulated together with the GLOMAP mode aerosol scheme
(Mann et al., 2010). Examples of the on-line integration of chemistry modules
in global circulation models with focus on NWP are GEM-AQ (Kaminski et al.,
2008), GEMS-BACH (Ménard et al., 2007) and GU-WRF/Chem (Zhang et al.,
2012). Savage et al. (2013) evaluate the performance of air quality forecast
with the UM on the regional scale. Baklanov et al. (2014) give a
comprehensive overview of on-line coupled chemistry–meteorological models
for regional applications.
C-IFS is intended to run with several chemistry schemes for both the
troposphere and the stratosphere in the future. Currently, only the
tropospheric chemical mechanism CB05 originating from the TM5 CTM (Huijnen et
al., 2010) has been thoroughly tested. For example, C-IFS (CB05) has been
applied to study the HO2 uptake on clouds and aerosols (Huijnen et
al., 2014) and pollution in the Arctic (Emmons et al., 2014). The
tropospheric and stratospheric scheme RACMOBUS of the MOCAGE model (Bousserez
et al., 2007) and the MOZART 3 chemical scheme as well as an extension of the
CB05 scheme with the stratospheric chemical mechanism of the BASCOE model
(Errera et al., 2008) have been technically implemented and are being
scientifically tested. Only C-IFS (CB05) is the subject of this paper.
Each chemistry scheme in C-IFS consists of the specific gas-phase chemical
mechanism, multi-phase chemistry, the calculation of photolysis rates and
upper chemical boundary conditions. Dry and wet deposition, emission
injection and parameterisation of lightning NO emissions as well as
transport and diffusion are simulated by the same approach for all chemistry
schemes. Likewise, emissions and dry deposition input data are kept the same
for all configurations.
The purpose of this paper is to document C-IFS and to present its model
performance with respect to observations. Since C-IFS (CB05) replaced the
current operational MACC model system for reactive gases (IFS-MOZART) both in
data assimilation and forecast mode, the evaluation in this paper is carried
out predominantly with observations that are used for the routine evaluation
of the MACC II system. The model results are compared (i) with a MOZART
stand-alone simulation, which is equivalent to a IFS-MOZART simulation, and
(ii) with the MACC re-analysis (Inness et al., 2013), which is an application
of IFS-MOZART in data assimilation mode. All model configurations used the
same emission data. The comparison demonstrates that C-IFS is ready to be
used operationally.
The paper is structured as follows. Section 2 is a description of the C-IFS,
with the focus on the newly implemented physical parameterisations and the
CB05 chemical mechanism. Section 3 contains the evaluation with observations
of a 1 year simulation with C-IFS (CB05) and a comparison with the results
from the MOZART run and the MACC re-analysis. The paper is concluded with a
summary and an outlook in Sect. 4.
Description of C-IFS
Overview of C-IFS
The IFS consists of a spectral NWP model that applies the semi-Lagrangian
(SL) semi-implicit method to solve the governing dynamical equations. The
simulation of the hydrological cycle includes prognostic representations of
cloud fraction, cloud liquid water, cloud ice, rain and snow (Forbes et al.,
2011). The simulations presented in this paper used the IFS release CY40r1.
The technical and scientific documentation of this IFS release can be found
at
http://www.ecmwf.int/en/forecasts/documentation-and-support/changes-ecmwf-model/cy40r1-summary/cycle-40r1. Changes in the operational model are documented at
https://software.ecmwf.int/wiki/display/IFS/Operational+changes.
At the start of the time step, the three-dimensional advection of the
tracers mass mixing ratios is simulated by the SL method as described in
Temperton et al. (2001) and Hortal (2002). Next, the tracers are vertically
distributed by the diffusion scheme (Beljaars and Viterbo, 1998) and by
convective mass fluxes (Bechtold et al., 2014). The diffusion scheme also
simulates the injection of emissions and the loss by dry deposition (see
Sect. 2.4.1). The output of the convection scheme is used to calculate
NO
production by lightning (see Sect. 2.4.3). Finally, the sink and source
terms due to chemical conversion (see Sect. 2.5), wet deposition (see
Sect. 2.4.2) and prescribed surface and stratospheric boundary conditions
are calculated (see Sect. 2.5.2).
The chemical species and the related processes are represented only in
grid-point space. The horizontal grid is a reduced Gaussian grid (Hortal and
Simmons, 1991). C-IFS can be run at varying vertical and horizontal
resolutions. The simulations presented in this paper were carried out at a
T255 spectral resolution (i.e. truncation at wave number 255), which
corresponds to a grid box size of about 80 km. The vertical
discretisation uses 60 levels up to the model top at 0.1 hPa
(65 km) in a hybrid sigma-pressure 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 in height.
The modus operandi of C-IFS is one of a forecast model in a NWP framework.
The 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
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 NWP applications of IFS. The advection is simulated with
a three-dimensional semi-Lagrangian advection scheme, which applies a
quasi-monotonic 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 (2011). A
proportional mass fixer 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 deposition 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). The scheme considers
deep, shallow and mid-level convection. Clouds are represented by a single
pair of entraining/detraining plumes which determine the updraught and
downdraught mass fluxes (http://old.ecmwf.int/research/ifsdocs/CY40r1/
in Physical Processes, Chapter 6, pp. 73–90). Highly soluble species such as
nitric acid (HNO3), hydrogen peroxide (H2O2) and
aerosol precursors are assumed to be scavenged in the convective rain
droplets and are therefore excluded from the convective mass transfer.
Annual emissions from anthropogenic, biogenic and natural sources
and biomass burning for 2008 in Tg for a C-IFS (CB05) run at T255
resolution. Anthropogenic NO emissions contain a contribution of
1.8 Tg aircraft emissions and 12.3 Tg (5.7 TgN)
lightning emissions (LiNO) is added in the biomass burning columns.
Species
Anthropogenic
Biogenic and natural
Biomass burning
CO
584
96
328
NO
70+1.8
10
9.2+12.3 (LiNO)
HCHO
3.4
4.0
4.9
CH3OH
2.2
159
8.5
C2H6
3.4
1.1
2.3
C2H5OH
3.1
0
0
C2H4
7.7
18
4.3
C3H8
4.0
1.3
1.2
C3H6
3.5
7.6
2.5
Parafins (TgC)
31
18
1.7
Olefines (TgC)
2.4
0
0.7
Aldehydes (TgC)
1.1
6.1
2.1
CH3COCH3
1.3
28
2.4
Isoprene
0
523
0
Terpenes
0
97
0
SO2
98
9
2.2
DMS
0
38
0.2
NH3
40
11
6.2
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 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 kgkg-1. For the
majority of the species the contribution of the negative fixer was below
0.1 % of the dominating source or sink term. The contribution was of the
order of 1 % for nitrogen species such as NO, N2O5 as
well as up to 3 % for highly soluble species such HNO3,
HO2, NO3_A. Large gradients of NOx at the
terminator in the stratosphere as well as intensive wet deposition were the
reasons for the increased occurrence of projected negative concentrations.
Emissions for 2008
The anthropogenic surface emissions were given by the MACCity inventory
(Granier et al., 2011) and aircraft NO emissions of a total of ∼0.8 TgNyr-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 (Granier et al.,
2005; Olivier et al., 2003). The biogenic emissions were simulated off-line
by the MEGAN2.1 model (Guenther et al., 2006). The anthropogenic and natural
emissions were used as monthly means without accounting for the diurnal
cycle. Daily biomass burning emissions were produced by the Global Fire
Assimilation System (GFAS) version 1, which is based on satellite retrievals
of fire radiative power (Kaiser et al., 2012). The actual emission totals
used in the T255 simulation for 2008 from anthropogenic and biogenic sources
and biomass burning as well as lighting NO are given in Table 1.
Physical parameterisations of sources and sinks
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 of the vegetation and the presence
of intercepted rain water. Dry deposition plays an important role in the
biogeochemical cycles of nitrogen and sulfur, and it is a major loss process
of tropospheric O3. Modelling the dry deposition fluxes in C-IFS is
based on a resistance model (Wesely, 1989), which differentiates the
aerodynamic, the quasi-laminar and the canopy or surface resistance. The
inverse of the total resistance is equivalent to a dry deposition velocity
VD.
The dry deposition flux FD at the model surface is calculated
based on the dry deposition velocity VD, the mass mixing ratio
Xs and air density ρs at the lowest model level s, in the
following way:
FD=VDXsρs.
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 Xs.
The dry deposition velocities were calculated as monthly mean values from a
1 year simulation using the approach described in Michou et al. (2004). It
used meteorological and surface input data such as wind speed, temperature,
surface roughness 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 uppermost surface layer. Together with the cuticular and mesophyllic
resistances this is combined into the leaf resistance according to Wesely
(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 O3 and sulfur dioxide (SO2) dry
deposition velocities can be up to 4 times 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
O3.
Table A4 (Supplement) contains annual total loss by dry deposition and is
expressed as a lifetime 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 SO2 and ammonia
(NH3), as the respective lifetimes were 1 day to 1 week. For
tropospheric O3, the respective globally averaged timescale is
about 3 months. Because dry deposition occurs mainly over ice-free land
surfaces, the corresponding timescale is at least 3 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.
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 account for the simulation of wet deposition. The IFS cloud
scheme provides information 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 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 HNO3 because the species is only removed from the 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 to the assumption that there
is instantaneous mixing within the grid box on the timescale of the model
time step. As discussed in Huijnen et al. (2014), 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 (Jacob et al., 2000; Liu et al., 2001). In contrast to Jacob et
al. (2000), tracers 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.
The input fields to the wet deposition routine are the following prognostic
variables, calculated by the IFS cloud scheme (Forbes et al., 2011): total
cloud and ice water content, grid-scale rain and snow water content and cloud
and grid-scale precipitation fraction as well as the derived fluxes for
convective and grid-scale precipitation fluxes 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 ms-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 calculated using Henrys law equilibrium or assuming that
70 % of aerosol precursors such as sulfate (SO4), NH3
and nitrate (NO3) is dissolved in the droplet. The effective Henry
coefficient for SO2, which accounts for the dissociation of
SO2, 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 A1 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 1.0
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 HNO3 and
0.6 for H2O2 (von Blohn, 2011). In ice clouds only,
H2O2 (Lawrence and Crutzen, 1998) and HNO3 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 (Jacob et al., 2000). Wash-out is either mass-transfer or
Henry-equilibrium limited. HNO3, aerosol precursors and 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.
Table A5 (Supplement) contains total loss by wet deposition and is expressed
as a timescale in days based on the tropospheric burden. For aerosol
precursors nitrate, sulfate and ammonium, HNO3 and
H2O2 wet deposition is the most important loss process, with
respective timescales of 2–4 days.
NO emissions from lightning
NO emissions from lightning are a considerable contribution to the
global atmospheric NOx budget. Estimates of the global annual
source vary between 2 and 8 TgNyr-1 (Schumann and Huntrieser,
2007). 5 TgNyr-1 (10.7 TgNOyr-1) is the most
commonly assumed value for global CTMs, which is about 6–7 times the value
of NO emissions from aircraft (Gauss et al., 2006), 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
O3 budget and hence the OH–HO2 partitioning
(DeCaria et al., 2005).
Flash density in flashes(km-2yr-1) from the IFS
input data using the parameterisation by Price and Rind (1992) (left), Meijer
et al. (2001) (middle) and observations from the LIS OTD database (right).
All fields were scaled to an annual flash density of 46 fls-1.
The parameterisation of the lightning NO production in C-IFS consists
of estimates 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 flash-rate densities using the
following input parameters: (i) convective cloud height (Price and Rind,
1992) or (ii) convective precipitation (Meijer et al., 2001).
The parameterisations 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 parameterisations were derived from field studies and depend on the model
resolution. With the current implementation of C-IFS (T255L60), the global
flash rates were 26 and 43 flashes per second 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., 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 Africa was well
reproduced by both parameterisations, but the schemes produce an exaggerated
maximum over tropical South America. The lightning activity over the United
States was underestimated by both parameterisations. The parameterisation 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 is followed 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 fourth-order
function of cloud height above freezing level.
The vertical distribution of the NO release is of importance for its impact
on atmospheric chemistry. Many CTMs use the suggestion of Pickering et
al. (1998) of a C-shape 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 approach 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, using the parameterisation of Meijer et al. (2001) based
on convective precipitation, were 4.9 TgNyr-1 at T159
resolution and 5.7 TgNyr-1 at T255 resolution.
CB05 chemistry scheme
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) 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 (CH3OH), ethane
(C2H6), propane (C3H8), propene (C3H6)
and acetone (CH3COCH3) has been introduced as described in
Williams et al. (2013). The isoprene oxidation has been modified motivated by
Archibald et al. (2010). Higher C3 peroxy radicals formed during
the oxidation of C3H6 and C3H8 were included
following Emmons et al. (2010).
The CB05 scheme is supplemented with chemical reactions for the oxidation of
SO2, di-methyl sulfide (DMS), methyl sulfonic acid (MSA) and
NH3, as outlined in Huijnen et al. (2014). For the oxidation of
DMS, the approach of Chin et al. (1996) is adopted. Table A1 (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 either Sander et al. (2011) or Atkinson et al. (2004, 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. C2 and C3 aldehydes, in that group. An
overview of all gas-phase reactions and reaction rates as applied in this
version of C-IFS can be found in Table A2 (Supplement).
For the loss of trace gases by heterogeneous oxidation processes, the model
explicitly accounts for the oxidation of SO2 in cloud through
aqueous-phase reactions with H2O2 and O3, depending on
the acidity of the solution. The pH is computed from the SO4, MSA,
HNO3, NO3_A, NH3 and NH4
concentrations, as well as from a climatological CO2 value. The pH,
in combination with the Henry coefficient, defines the fraction of sulfate
residing in the aqueous phase, compared to the gas-phase concentration
(Dentener and Crutzen, 1993). The heterogeneous conversion of
N2O5 into HNO3 on cloud droplets and aerosol particles
is applied with a reaction probability (γ) set to 0.02 (Evans and
Jacob, 2005). The surface area density is computed based on a climatological
aerosol size distribution function, applied to the SO4, MSA and
NO3_A aerosol, as well as to clouds assuming a droplet size of
8 µm.
Photolysis rates
For the calculation of photo-dissociation rates, an on-line parameterisation
for the derivation of actinic fluxes is used (Williams et al., 2012). It
applies a modified band approach (MBA), which is an updated version of the
work by Landgraf and Crutzen (1998), tailored and optimised for use in
tropospheric CTMs. The approach uses seven 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
seven pre-defined band intervals based on the two-stream solver of Zdunkowski
et al. (1980).
The optical depth of clouds is calculated based on a parameterisation
available in IFS (Slingo, 1989; Fu et al., 1998) 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.
In total, 20 photolysis rates are included in the scheme, as given in
Table A3 (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., 1993). This solver was 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 iterations 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 model levels, where the
perturbations due to emissions can be large.
Stratospheric boundary conditions
The modified CB05 chemical mechanism includes no halogenated species and no
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 O3, methane
(CH4), and HNO3 are needed to capture the influence of
stratospheric intrusions on the composition of the upper troposphere.
Stratospheric O3 chemistry in C-IFS (CB05) is parameterised by the
Cariolle scheme (Cariolle and Teyssèdre, 2007). Chemical tendencies for
stratospheric and tropospheric O3 are merged at an empirical
interface of the diagnosed tropopause height in IFS. Additionally,
stratospheric O3 in C-IFS can be nudged to O3 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 temperature gradient.
Stratospheric HNO3 at 10 hPa is controlled by a climatology
of HNO3 and O3 observations from the Microwave Limb
Sounder (MLS) aboard the Upper Atmosphere Research satellite (UARS).
HNO3 is set to according to the observed
HNO3–O3 ratio and the simulated O3
concentrations. Furthermore, stratospheric CH4 is constrained by a
climatology based on observations of the Halogen Occultation Experiment
instrument (Grooß and Russel, 2005), at 45 and at 90 hPa in the
extra-tropics, which implicitly accounts for the stratospheric chemical loss
of CH4 by OH, chlorine (Cl) and oxygen
(O1D) radicals. It should be noted that the surface concentrations
of CH4 are also fixed in this configuration of the model.
Gas–aerosol partitioning
Gas–aerosol partitioning is calculated using the Equilibrium Simplified
Aerosol Model (EQSAM, Metzger et al., 2002a, b). The scheme has been
simplified so that only the partitioning between HNO3 and the
nitrate aerosol (NO3-) and between NH3 and the
ammonium aerosol (NH4+) is calculated. SO42- 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
parameterisations 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 parameterisations is that 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:
(NH3)g+(HNO3)g↔(NH4NO3)s
(NH4NO3)s+(H2O)g↔(NH4NO3)aq+(H2O)aq
(NH4NO3)aq+(H2O)g↔(NH4+)aq+(NO3-)aq+(H2O)aq
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 tracer mass, the global integral of the total
surface emissions, integrated wet and dry deposition fluxes, chemical
conversion, as well as elevated atmospheric emissions and the contributions
of prescribed upper and lower vertical boundary conditions for CH4
and HNO3. A time-invariant pressure-based tropopause definition,
which varies with latitude, is used to calculate the tropospheric mass. To
monitor the numerical integrity of the scheme, the contributions of the
corrections to ensure positiveness and global mass conservation are
calculated. Optionally, more detailed diagnostics can be requested that
includes photolytic loss and the loss by OH for the tropics and
extra-tropics.
A detailed analysis of the global chemistry budget is beyond the scope of
this paper. Only a number of key terms for CO, O3 and CH4
are summarised here. They are compared with values from the Atmospheric
Composition Change: the European Network of Excellence (ACCENT) model
inter-comparisons of chemistry models by Stevenson et al. (2006) for
tropospheric O3 and by Shindell et al. (2006) for CO. A more
recent inter-comparison was carried out within the Atmospheric Chemistry and
Climate Model Intercomparison Project (ACCMIP) (Lamarque et al., 2013). The
ACCMIP values have been taken from Young et al. (2013) for tropospheric
O3 and from Voulgarakis et al. (2013) for CH4. It should
be noted that the values from these inter-comparisons are valid for
present-day conditions, but not specifically for 2008. A further source of
the differences is the height of the tropopause assumed in the calculations.
Overall, the comparison showed that the C-IFS (CB05) is well within the range
of the two multi-model ensembles.
The annual mean of the C-IFS tropospheric O3 burden was
390 Tg. The values are at the upper end of the range simulated by the
ACCENT (344±39 Tg) and the ACCMIP (337±23 Tg)
models. The same holds for the loss by dry deposition, which was
1155 Tgyr-1 for C-IFS, 1003±200 Tgyr-1 for
ACCENT and in the range 687–1350 Tgyr-1 for ACCMIP. The
tropospheric chemical O3 production of C-IFS was
4608 Tgyr-1 and loss 4144 Tgyr-1, which is for both
values at the lower end of the range reported for the production (5110±606 Tgyr-1) and loss (4668±727 Tgyr-1) for
the ACCENT models. The comparatively simple treatment of volatile organic
compounds in CB05 could be an explanation for the low O3 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 than the ACCENT mean (345, 248–427 Tg). The total
CO emissions in 2008 were 1008 Tg, which is in line with the
number used in ACCENT (1077 Tgyr-1) but lower than the estimate
(1550 Tgyr-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 was 1434 Tgyr-1, which is very
close to the ACCENT multi-mean of 1505±236 Tgyr-1. The
chemical CO loss in C-IFS was 2423 Tg and the loss by dry
deposition 24 Tg.
The annual mean CH4 total and tropospheric burdens of C-IFS (CB05)
are 4874 and 4271 Tgyr-1, respectively. The global chemical
CH4 loss by OH was 467 Tgyr-1. Following
Stevenson et al. (2006), this leads to a global CH4 lifetime
estimate of 9.1 years. This value is within the ACCMIP range of 9.8±1.6 years but lower than an observation-based 11.2±1.3 years estimate
by Prather et al. (2012). CH4 emissions were substituted by
prescribed monthly zonal-mean surface concentrations to avoid the long-spin
up needed by a direct modelling of the CH4 surface fluxes. The
CH4 surface concentrations were derived from a latitudinal
interpolation of observations from the South Pole, Cape Grim, Mauna Loa, Mace
Head, Barrow and Alert stations as discussed in Bândă et
al. (2015). The
resulting CH4 flux was 488 Tgyr-1, which is of similar
size as the sum of current estimates of the total CH4 emissions of
500–580 Tgyr-1 and the loss by soils of
30–40 Tgyr-1 (Forth Assessment Report by IPCC
http://www.ipcc.ch/publications_and_data/ar4/wg1/en/ch7s7-4-1.html#ar4top).
Evaluation with observations and comparison with the
IFS-MOZART coupled system
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 is to show how C-IFS (CB05) performs relative to
the MOZART-3 coupled CTM (Kinnison et al., 2007), which has been running in
the IFS-MOZART coupled system 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 O3, CO, nitrogen dioxide (NO2),
SO2 and 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.
The MACC re-analysis (REAN) and the corresponding MOZART (MOZ) stand-alone
run have already been evaluated with observations by Inness et al. (2013).
Furthermore, 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
O3 by Im et al. (2014). C-IFS (CB05) has been already evaluated
with a special focus on hydroperoxyl (HO2) in relation to CO
in Huijnen et al. (2014). The performance of an earlier version of C-IFS
(CB05) in the Arctic was evaluated and inter-compared with CTMs of the
POLARCAT model intercomparison Project (POLMIP) by Monks et al. (2014) for CO
and Arnold et al. (2014) for reactive nitrogen. The POLMIP inter-comparisons
show that C-IFS (CB05) performs within the range of state-of-the-art CTMs.
Summary of model runs set-up
C-IFS (CB05) was run from 1 January to 31 December 2008 with a spin-up
starting 1 July 2007 at a T255 resolution (80km×80km) 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 O3 above the tropopause was
nudged to the MACC re-analysis.
MOZ is a run with the MOZART CTM at 1.1∘×1.1∘ (120×120 km) horizontal resolution using the 60 vertical levels of
C-IFS. The set-up of the MOZART model and the applied emissions and dry
deposition velocities were the same in MOZ and REAN. The most important
difference between MOZ and REAN is the assimilation of satellite retrieval of
atmospheric composition in REAN. Furthermore, REAN was produced with the
IFS-MOZART coupled system, whereas MOZ is a stand-alone system driven by the
meteorological fields of REAN. The latter is equivalent to a simulation of
IFS-MOZART without data assimilation of atmospheric composition. The
assimilated retrievals were CO and O3 total columns,
stratospheric O3 profiles and tropospheric NO2 columns.
No observations of atmospheric composition have been feed in to the MOZ run.
No observational information has been used to improve the tropospheric
simulation of the C-IFS run. Another difference between MOZ and REAN is that
the IFS diffusion and convection scheme, as used in C-IFS, controls the
vertical transport in REAN, whereas MOZART's generic schemes were used in the
MOZ run.
Ozonesonde sites used in the evaluation for different regions.
Region
Area S/W/N/E
Stations (number of observations)
Europe
35∘ N/20∘ W/60∘ N/40∘ E
Barajas (52), DeBilt (57), Hohenpeissenberg (126), Legionowo (48), Lindenberg (52), Observatoire de Haute-Provence (46), Payerne (158), Prague (49), Uccle (142) and Valentia Observatory (49)
North America
30∘ N/135∘ W/60∘ N/60∘ W
Boulder (65), Bratts Lake (61), Churchill (61), Egbert (29), Goose Bay (47), Kelowna (72), Stony Plain (77), Wallops (51), Yarmouth (60), Narragansett (7) and Trinidad Head (35)
Arctic
60∘ N/180∘ W/90∘ N/180∘ E
Alert (52), Eureka (83), Keflavik (8), Lerwick (49), Ny-Aalesund (77), Resolute (63), Scoresbysund (54), Sodankyla (63), Summit (81) and Thule (15)
Tropics
20∘ S/180∘ W/20∘ N/180∘ E
Alajuela (47), Ascension Island (32), Hilo (47), Kuala Lumpur (24), Nairobi (39), Natal (48), Paramaribo (35), Poona (13), Samoa (33), San Cristobal (28), Suva (28), Thiruvananthapuram (12) and Watukosek (19)
East Asia
15∘ N/100∘ E/45∘ N/142∘ E
Hong Kong Observatory (49), Naha (37), Sapporo (42) and Tateno Tsukuba (49)
Antarctic
90∘ S/180∘ W/60∘ S/180∘ E
Davis (24), Dumont d'Urville (38), Maitri (9), Marambio (66), Neumayer (72), South Pole (63), Syowa (41) and McMurdo (18)
MOZ, REAN and C-IFS used the same anthropogenic emissions (MACCity), biogenic
emissions (MEGAN 2.1; Guenther et al., 2006,
http://acd.ucar.edu/~guenther/MEGAN/MEGAN.htm) and natural emissions
from the POET project. The biomass burning emissions for MOZ and REAN came
from the Global Fire Emission Data version 3 inventory which was
redistributed according to fire radiative power observations used in GFAS.
Hence, the average biomass burning emissions used by MOZART (MOZ and REAN)
agree well with the GFAS emissions used by C-IFS, but they are not identical
in temporal and spatial variability.
Observations
The runs (C-IFS, MOZ, REAN) were evaluated with O3 observations
from ozonesondes and O3 and CO aircraft profiles from the
Measurement of Ozone, Water Vapour, Carbon Monoxide and Nitrogen Oxides by
Airbus in-service Aircraft (MOZAIC) program. Simulated surface O3,
CO, NO2 and SO2 fields were compared against
Global Atmospheric Watch (GAW) surface observations and additionally
O3 against observations from the European Monitoring and Evaluation
Programme (EMEP) and the European air quality database (AirBase). The global
distributions of tropospheric NO2 and HCHO were evaluated with
retrievals of tropospheric columns from Global Ozone 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. The observation error of the sondes is about ±5 % in the
range from 200 to 10 hPa and -7–17 % below 200 hPa
(Beekmann et al., 1994; Komhyr et al., 1995, and Steinbrecht et al.,
1998). The
number of soundings varied for the different stations. Typically, the sondes
are launched once a week but in certain periods such as during O3
hole conditions soundings are more frequent. Sonde launches were carried out
mostly between 9 and 12 h local time. 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. Tilmes et al. (2012)
suggest a further refinement of the North America region into Canada and the
eastern and western United States as well of the tropics into Atlantic
Africa, the equatorial Americas and the eastern Indian Ocean / western
Pacific based on the inter-comparison of ozonesonde observations for the
1994–2010 period. The results will also be discussed for the sub-regions and
figures will be presented in the Supplement.
The MOZAIC program (Marenco et al., 1998, and Nédélec et al., 2003)
provides profiles 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 in 2008 had a sufficient number of profiles: Windhoek (323),
Caracas (129), Hyderabad (125) and London–Gatwick (83) as well as North
American airports Atlanta (104), Portland (69), Philadelphia (65), Vancouver
(56), Toronto (46) and Dallas (43). The North American airports were
considered to be close enough to make a spatial average meaningful. Because
of the varying data availability the North American mean is dominated by the
airports in the eastern United States.
Apart from Frankfurt, typically two 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 two to six profiles available each day, mostly in the
morning and the later afternoon to the evening. At the other airports the
typical observation times were 06:00 and 18:00 UTC for Windhoek (±0 h local time), 19:00 and
21:00 UTC for Hyderabad (+4 h local time), 20:00 and 22:00 UTC
for Caracas (-6 h), 04:00 and 22:00 for London (±0 h)
and 19:00 and 22:00 (-5/6 h) for the North American airports. This
means that most of the observations were taken between the late evening and
early morning hours, i.e. at a time of increased stability and large
CO vertical gradients close to the surface. Only the observations at
Caracas (afternoon) and to some extent in Frankfurt represent a more mixed
day-time boundary layer. The modelled column profile was obtained at the
middle between the start and end times of the profile observation and no
consideration was given to the horizontal movement of the aircraft. The model
columns were interpolated in time between two subsequent output time steps.
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
difference between a high-resolution orography and the actual station height.
The data coverage for CO and O3 was global, whereas for
SO2 and NO2, 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 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 O3,
which is spatially well correlated (Flemming et al., 2005). The EMEP
observations and the rural Airbase O3 observations were used for
the evaluation over Europe.
Satellite retrievals
Satellite retrievals of atmospheric composition are more widely used to
evaluate model results. Satellite data provide good horizontal coverage but
have limitation with respect to the vertical resolution and signal from the
lowest atmospheric levels. Furthermore, satellite observations are only
possible at the specific overpass time, and they can be disturbed by the
presence of clouds and surface properties. Depending on the instrument type
global coverage is achieved in several days.
Day-time CO total column retrievals from MOPITT, version 6 (Deeter,
2013), and retrievals of tropospheric columns of NO2 (IUP-UB v0.7,
Richter et al., 2005) and of HCHO (IUP-UB v1.0; Wittrock et al., 2006) from
GOME-2 (Callies et al., 2000) have been used for the evaluation. The
retrievals were averaged to monthly means values to reduce the random
retrieval error.
Tropospheric ozone volume mixing ratios (ppb) over Europe (left) and
North America (middle) and East Asia (right) averaged in the pressure ranges
1000–700 hPa (bottom), 700–400 hPa (middle) and
400–200 hPa (top) observed by ozonesondes (black) and simulated by
C-IFS (red), MOZ (blue) and REAN (green) in 2008.
MOPITT is a multispectral thermal infrared (TIR)/near infrared (NIR)
instrument onboard the TERRA satellite with a pixel resolution of
22 km. TERRA's local equatorial crossing time is approximately
10:30 a.m. The MOPITT CO level 2 pixels were binned within
1×1∘ within each month. Deeter et al. (2013) report a bias of
about +0.08×1018 moleccm-2 and a standard deviation
(SD) of the error of 0.19×1018 moleccm-2 for the
TIR/NIR product version 5. This is equivalent to a bias of about 4 % and
a SD of 10 % respectively assuming typical observations of 2.0×1018 moleccm-2. For the calculation of the simulated
CO total column, the a priori profile in combination with the
averaging kernels (AK) of the retrievals was applied. They have the largest
values between 300 and 800 hPa. The AK have been applied to ensure
that the difference between retrieval and the AK-weighted model column is
independent of the a priori CO profiles used in the retrieval. One
should note however, that the AK-weighted column is not equivalent to the
modelled atmospheric CO burden anymore.
GOME-2 is a ultra violet-visible (UV-VIS) and NIR sensor designed to provide
global observations of atmospheric trace gases. GOME-2 flies in a
sun-synchronous orbit with an equatorial crossing time of 09:30 LT in
descending mode and has a footprint of 40×80 km. Here,
tropospheric vertical columns of NO2 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 NO2. and between 337 and
353 nm for HCHO (Vrekoussis et al., 2010). Second, the
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 database (Wang
et al., 2008) are
used here. Furthermore, retrievals are limited to maximum solar zenith angles
of 85∘ for NO2 and 60∘ for HCHO.
Uncertainties in NO2 satellite retrievals are large and depend on
the region and season. Winter values at middle 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 smaller than that of NO2, 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 without
applying AK to tropospheric vertical columns of NO2 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.
Tropospheric ozone volume mixing ratios (ppb) over the tropics
(left), Antarctica (middle) and the Arctic (right) averaged in the pressure
bands 1000–700 hPa (bottom), 700–400 hPa (middle) and
400–200 hPa (top) observed by ozonesondes and simulated by C-IFS
(red), MOZ (blue) and REAN (green) in 2008.
Annual cycle of the mean ozone volume mixing ratios (ppb) at rural
sites of the EMEP and AirBase database and simulated by C-IFS (red), MOZ
(blue) and REAN (green).
Diurnal cycle of surface ozone volume mixing ratios (ppb) over
Europe in winter (top, left), spring (top, right), summer (bottom, left) and
autumn (bottom, right) at the rural site of the EMEP and AirBase database and
simulated by C-IFS (red), MOZ (blue) and REAN (green).
CO total column retrieval (MOPITT V6) for April 2008 (top
left) and simulated by C-IFS (top right), MOZ (bottom left) and REAN (bottom
right); AK are applied.
CO total column retrieval (MOPITT V6) for August 2008 (top
left) and simulated by C-IFS (top right), MOZ (bottom left) and REAN (bottom
right); AK are applied.
CO volume mixing ratios (ppb) over Frankfurt (left),
London (middle) and North America (left, averaged over six airports) averaged
in the pressure bands 1000–700 hPa (bottom), 700–400 hPa
(middle) and 400–200 hPa (top) observed by MOZAIC and simulated by
C-IFS (red), MOZ (blue) and REAN (green) in 2008.
Tropospheric ozone
Figure 2 shows the monthly means of O3 volume mixing ratios in the
pressure ranges surface to 700 hPa (lower troposphere, LT)
700–400 hPa (middle troposphere, MT) and 400–200 hPa (upper
troposphere UT) observed by sondes and averaged over Europe, North America
and East Asia. Figure 3 shows the same as Fig. 2 for the tropics, Arctic and
Antarctica. A more detailed breakdown of North America (Canada, eastern and
western United States) and the tropics (Atlantic Africa, equatorial Americas
and eastern Indian Ocean/western Pacific) following Tilmes et al. (2012) is
presented in the supplement. The observations have a pronounced spring
maximum for UT O3 over Europe, North America and East Asia and a
more gradually developing maximum in late spring and summer in MT 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 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 O3
considerably but had only little influence in LT and MT. The overestimation
of MOZ in UT seems to be caused by increased stratospheric O3
rather than a more efficient transport as lower stratospheric O3
was overestimated in MOZ. Note that stratospheric ozone in C-IFS was nudged
to the MACC re-analysis (see Sect. 3.1) but good agreement of C-IFS with
observation in UT in all three regions is also present in a run without
nudging to stratospheric O3. It is therefore not only a consequence
of the use of assimilated observations in C-IFS (CB05).
Over North America the spring-time underestimation by C-IFS and MOZ is more
pronounced than over Europe. The underestimation occurs in all regions but
was largest in early spring over Canada. C-IFS also underestimates spring
ozone throughout North America in MT. LT summer-time ozone was overestimated
in North America by all models, in particular over the eastern United States.
The bias of C-IFS was the smallest in LT but, in contrast to MOZ and REAN,
C-IFS underestimates summer-time ozone in MT over the eastern United States.
The overestimation of UT ozone by MOZ was most pronounced in Canada.
In East Asia all runs overestimate by 5–10 ppb in LT and MT,
especially in autumn and winter. At the northern high latitudes (Fig. 3) the
negative spring bias appears in all runs in LT and only for C-IFS in MT. As
in the other regions, MOZ greatly overestimates UT O3.
Averaged over the tropics, the annual variability is below 10 ppb,
with maxima in May and 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. A more
detailed analysis for different tropical regions shows that the seasonality
is well captured by all models over Atlantic Africa, equatorial America and
the eastern Indian Ocean / western Pacific in all three tropospheric levels.
However, the strong observed monthly anomalies (an observation glitch by one
station) in equatorial America in March and September were underestimated by
up to 20 ppb in all tropospheric levels.
CO volume mixing ratios (ppb) over Caracas (left), Windhoek
(middle) and Hyderabad (right), averaged in the pressure bands
1000–700 hPa (bottom), 700–400 hPa (middle) and
400–200 hPa (top) observed by MOZAIC, and simulated by C-IFS (red),
MOZ (blue) and REAN (green) in 2008.
Time series of the median of weekly CO surface volume mixing
ratios (ppb) in Europe (13 GAW sites) and model results of C-IFS, MOZ and
REAN.
NO2 tropospheric column retrieval (GOME-2) for 2008 (top
left) and by C-IFS (top right), REAN (bottom right) and MOZ (bottom left).
Time series of area-averaged tropospheric NO2 columns
(1015 moleccm-2) from GOME-2 compared to model results of
C-IFS (CB05) (blue), MOZ (red) and REAN (green) for different regions.
Time series of the median of weekly surface NO2 volume
mixing ratios (ppb) in Europe (20 GAW sites) and model results of
C-IFS, MOZ and REAN.
HCHO tropospheric column retrieval (GOME-2) for 2008 (top left)
and by C-IFS (top right), REAN (bottom right) and MOZ (bottom left).
Time series of area-averaged tropospheric HCHO columns
(1016 moleccm-2) from GOME-2 compared to model results of
C-IFS, MOZ and REAN for different regions.
Time series of the median of weekly surface SO2 volume
mixing ratios (ppb) in Europe (21 GAW sites) and model results of
C-IFS, MOZ and REAN.
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 in LT than MOZ but had a larger negative bias in MT. The biggest
improvement in 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
O3, in particular during polar night when UV satellite observations
are not available, as already discussed in Flemming et al. (2011).
The ability of the models to simulate O3 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 seasons in Europe.
All runs underestimate monthly mean O3 in spring and winter and
overestimate it in late summer and autumn. The overestimation in summer was
largest in MOZ. The recently reported (Val Martin at al., 2014) missing
coupling of the leaf area index to the leaf and stomatal vegetation
resistance in the calculation of dry deposition velocities could be an
explanation of the MOZ bias. While the overestimation appeared also 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 suggest. The diurnal cycles 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 overestimations of the
summer and 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 also plays an important role in the case of
biomass burning and high anthropogenic emissions. The global distribution of
total column CO retrieved from MOPITT 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. Figures showing the corresponding biases can be found in
the Supplement. April and August have been selected because they are the
months of the NH CO maximum and minimum. C-IFS reproduced well the
locations of 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
at middle and high latitudes in the Southern Hemisphere (SH) were
underestimated. The same global gradients of the bias were found in MOZ and
REAN. 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 the NH and not predominantly 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 the NH in MOZ. Assimilating MOPITT (V4) into
REAN led to much reduced biases everywhere even though the sign of bias in
the NH, tropics and SH remained. In August, the NH bias is reduced, but the
hemispheric pattern of the CO bias was similar to 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 of the GFAS biomass
burning emissions in that area.
More insight into the seasonal cycle and the vertical CO distribution
can be obtained from MOZAIC aircraft profiles. CO profiles at
Frankfurt (Fig. 8, left) provide a continuous record with about two to six
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 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 seasonal variability of LT
CO from MOZAIC and the model runs in North America are 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, 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 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. Although 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 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.
The outcome of the comparison with LT CO from MOZAIC is consistent
with the model bias with respect to the GAW surface observations in Europe
(Fig. 10). The winter biases were larger than summer biases and MOZ showed
the largest underestimation. The GAW stations measuring CO are mostly
located on mountains in the Alpine region and typical annual biases were
about -5, -20 and -35 ppb for REAN, C-IFS and MOZ,
respectively. The negative biases of stations in flatter terrain such as
Kollumerward tended to be larger.
Nitrogen dioxide
The global maxima of NO2 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 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 NO2 in larger areas of medium
observed NO2 levels in Asia and central Africa as well as in the
outflow areas over the western Atlantic and western Pacific Ocean. This could
mean that NO emissions in the most polluted areas are too low but also
that the simulated lifetime of NO2 is too short. Furthermore, an
insufficient simulation of NOx reservoir species such as PAN and
the lack of alkyl nitrates in CB05 might be the reason for the
underestimation.
The validation of the seasonality of NO2 (Fig. 12) for different
regions and months shows that tropospheric NO2 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 evaluation 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 GOME-2 NO2 tropospheric columns over East Asia. The winter-time
values are only half of the observations, whereas in summer, models agree
well with observations. In southern Africa
(20/0∘ S/15/15∘ W), the models overestimate the increased
NO2 values in the biomass burning season by a factor of 2 but show
good agreement with observations in the rest of the year. The overestimation
during biomass burning events could be related to the assumed NO
emission factor.
HCHO
On the global scale, HCHO is mainly chemically produced by the
oxidation of isoprene and CH4. Isoprene is emitted by vegetation.
On the regional scale, HCHO emissions from anthropogenic sources,
vegetation and biomass burning also contribute to the HCHO burden.
The annual average of tropospheric HCHO retrieved from GOME-2 and from the
model runs is shown in Fig. 14. The observations show higher values in the
tropics and the NH and maxima in the rain forest regions of South America and
central Africa and in Southeast Asia. The simulated fields of the three runs
are very similar. C-IFS, MOZ and REAN reproduce the observed global patterns
but show a small but widespread underestimation in the NH extra-tropics and
in industrialised East Asia. On the other hand, HCHO is overestimated
in Indonesia. 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 Indonesia
(5∘ S/5∘ N/100/120∘ E) throughout the year. The
seasonality in southern Africa (not shown) and tropical South America
(10/5∘ S/73/35∘ W) is well captured, in particular by
C-IFS. All models also reproduced the observations rather well for the
eastern United States (30/40∘ N/90/75∘ W), but tend to
underestimate wintertime HCHO columns for this region.
Sulfur dioxide
SO2 was evaluated with available GAW surface observations from
central and eastern Europe. There were considerable differences in the
performance for individual stations often caused by local effects not
resolved by the models. To summarise the evaluation for SO2,
Fig. 16 shows the median of weekly observed and modelled time series. REAN
and MOZ greatly exaggerated the seasonal cycle since the values in winter
were up to 8 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 higher in
summer. Overall, the on-line integration of C-IFS showed lower SO2
biases.
Computational cost (BU) of 24 h forecasts of different
horizontal model resolutions (60 levels) and chemistry schemes of C-IFS,
IFS-MOZART and IFS, *not fully optimised.
Resolution
IFS-MOZART
C-IFS (MOZART)*
C-IFS (MOCAGE)*
C-IFS (CB05)
IFS
T159
205
56
147
20
6
T255
1200
–
–
55
12
T511
–
–
–
700
125
As no SO2 observations were assimilated in REAN and identical
SO2 emissions were used, the differences between the runs were
caused by differences in the simulation of vertical mixing, sulfur 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 in MOZART.
Computational cost
The computational cost is an important factor for the operational
applications in CAMS. The computational costs 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 IBM Power 7 super-computer. BUs
are proportional to the number of used central processing units (CPU) times
the simulation time.
The increase in cost because of the simulation of the CB05 chemistry with
respect to an NWP run is a factor of about 4 at resolutions T159
(110 km), T255 (80 km) and T511 (40 km). C-IFS (CB05)
is about 8 times more efficient than the IFS-MOZART coupled system at a T159
resolution and about 15 times more at a T255 resolution. This strong relative
increase in cost of IFS-MOZART is caused by the increasing memory
requirements of the IFS at higher resolution, or also in data assimilation
mode. However, there is insufficient parallelism in MOZART to exploit the
larger number of CPUs for speeding up the simulation of the coupled system.
C-IFS with the MOZART chemical mechanism, i.e. the same chemistry scheme as
in 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 RACMOBUS schemes
encompass a larger number of species and reactions and include a full
stratospheric chemistry scheme, which is missing in CB05. The overhead
because of the doubled number of advected species in C-IFS RACMOBUS and
MOZART is however small because of the efficiency of the SL advection scheme.
Summary and outlook
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 IFS-MOZART coupled system for forecast and
assimilation of reactive gases within the pre-operational Copernicus
Atmosphere Monitoring Service.
C-IFS applies the CB05 chemical mechanism, 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
Jacob (2000) and accounts for the 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 deposition fluxes are applied as surface
boundary conditions of the diffusion scheme. Lightning emissions of NO
can be calculated either by cloud height (Price and Rind, 1993) or by
convective precipitation (Meijer et al., 2001). The latter parameterisation
was used in this study. The 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, European surface observations of O3, CO,
SO2 and NO2, and global satellite retrievals of
CO, NO2 and HCHO. The evaluation showed that C-IFS
preforms better or with similar accuracy as MOZART and is mostly of a similar
quality as the MACC re-analysis. It should be noted that satellite retrievals
of CO, O3 and NO2 were assimilated into the MACC
re-analysis to improve the realism of the fields simulated by IFS-MOZART.
In comparison to MOZ, C-IFS (CB05) had smaller biases (i) for CO in
the Northern Hemisphere, (ii) for O3 in the upper troposphere and
(iii) for winter-time SO2 at the surface in Europe. Furthermore,
the diurnal cycle of surface O3, 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 in the chemical mechanism and the simulation of wet and dry
deposition. However, the improvements in SO2 and the diurnal cycle
of O3 are most probably caused by the more consistent interplay of
diffusion and sink and sources processes in the on-line integrated C-IFS.
There is still room for improvement of C-IFS (CB05). It underestimated
surface O3 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 NO2 over China as retrieved from the
GOME-2 instrument was 2 times higher than the fields modelled by C-IFS,
MOZART and the 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 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.
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 on-line calculated values. Furthermore, the impact of the simulated
O3 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 secondary aerosols. The calculation of
photolysis rates can account for the presence of aerosols, and HO2
uptake on aerosols can be simulated (Huijnen et al., 2014).
In summary, C-IFS is a new global chemistry weather model for forecast and
assimilation of atmospheric composition. C-IFS (CB05) has already been
successfully applied in data assimilation mode (Inness et al., 2015). C-IFS
offers improvements over the IFS-MOZART coupled system because (i) it
simulates several trace gas C-IFS (CB05)es with better accuracy, (ii) it is
computational several times more efficient in particular at high resolution
and (iii) it better facilitates the implementation of feedback processes
between gas-phase and aerosol processes as well as between atmospheric
composition and meteorology.
Code availability
The C-IFS source code is integrated into ECWMF's IFS code, which is only
available subject to a licence agreement with ECMWF. ECMWF member-state
weather services and their approved partners will get access granted. The IFS
code without modules for assimilation and chemistry can be obtained for
educational and academic purposes as part of the openIFS release
(https://software.ecmwf.int/wiki/display/OIFS/OpenIFS+Home). A detailed
documentation of the IFS code is available from
https://software.ecmwf.int/wiki/display/IFS/CY40R1+Official+IFS+Documentation.
The CB05 chemistry module of C-IFS was originally developed in the TM5
chemistry-transport model. Readers interested in the TM5 code can contact the
TM5 developers (http://tm5.sourceforge.net) or can go directly to the
TM5 wiki page, http://tm.knmi.nl/index.php/Main_Page.