C-IFS-CB05-BASCOE: Stratospheric Chemistry in the Integrated Forecasting System of ECMWF

We present a model description and benchmark evaluation of an extension of the tropospheric chemistry module in the Integrated Forecasting System (IFS) of the European Centre for Medium-Range Weather Forecasts (ECMWF) with stratospheric chemistry, referred to as C-IFS-CB05-BASCOE (for brev ity here referred to as C-IFS-TS). The stratospheric chemistry originates from the one used in the Belg ian Assimilat ion System for Chemical ObsErvations (BASCOE), and is here combined with the modified CB05 chemistry module for the troposphere as currently used operationally in the 15 Copernicus Atmosphere Monitoring Service (CAMS). In our approach either the tropospheric or stratospheric chemistry module is applied depending on the altitude of each individual grid box with respect to the tropopause. An evaluation of a 2.5 year long C-IFS-TS simulat ion with respect to various satellite retrieval products and in-situ observations indicates good performance of the system in terms of stratospheric ozone, and a general improvement in terms of stratospheric composition compared to the C-IFS predecessor model version. Possible issues with transport processes in the stratosphere are identified. 20 This marks a key step towards a chemistry module within IFS that encompasses both tropospheric and stratospheric composition, and could expand the CAMS analysis and forecast capabilities in the near future.


Introduction
Existing earth observation systems in comb ination with global circulat ion models (GCMs) help to provide a better understanding of the Earth's at mospheric co mposition and changes therein (Hollingsworth et al., 2008). For the troposphere, 25 hemispheric transport and chemical conversion of atmospheric composition influences regional air quality (Pausata et al., 2012;Im et al., 2015, Marécal et al., 2015. Also, the amount of stratospheric ozone directly impacts the forecast capabilities of surface solar irrad iance (Qu et al., 2014), stressing the relevance of good stratospheric ozone forecasts. Stratospheric ozone further affects the chemical co mposition in the troposphere because of stratosphere -troposphere transport of ozone (Stevenson et al., 2006, Gaudel et al., 2015, and its radiative properties influencing the tropospheric photolysis rates. 30 Beyond such direct imp lications on the troposphere a comprehensive description of stratospheric co mposition allows a more complete understanding of processes taking place in the stratosphere, ranging fro m tracking the ozone hole (Lefever et al., 2015) and understanding the concentrations of ozone depleting substances (Chipperfield et al., 2015), to the assessment of dynamical effects such as the Quasi-Biennial Oscillation (QBO, Bald win et al., 2001), and fro m implications of sudden stratospheric warmings on circulat ion patterns (Manney et al., 2015) to general radiative feedbacks of o zone, water vapour and CO 2 on weather and climate (Solomon et al., 2010).
These aspects have long been studied in the framework of Chemistry Transport Models (CTMs) and, mo re recently, in 5 GCMs, see, e.g., the SPA RC Chemistry-Climate Model Validation Activity (CCM Val, 2010). In GCMs the role of stratospheric ozone chemistry on the tropospheric climate can explicit ly be studied (e.g. Scaife et al., 2011). But also meteorological models can benefit fro m having a good representation of the stratospheric composition and its variability, considering the rad iative effects and the resulting impact on stratospheric as well as tropospheric temperatures (Monge-Sanz et al., 2013), which beco mes relevant for tropospheric forecast skills on long -range to seasonal time scales (Maycock et al., 10 2011).
Within a series of MACC (Monitoring At mospheric Co mposition and Climat e) European research pro jects a global forecast and assimilation system has been built, which is the core for the global system of the Copernicus Atmosphere Monitoring Service, (CAMS, http://atmosphere.copernicus.eu ). In CAMS, forecasts of atmospheric composition are carried out ), wh ich benefit fro m assimilation of satellite retrievals 15 , to imp rove the initial conditions for composition fields in terms of reactive gases, aerosols and greenhouse gases. Here a tropospheric chemistry scheme has been embedded in ECMWF's Integrated Forecast System, referred to as Co mposition-IFS (C-IFS, . Even though the current operational version of C-IFS based on the Carbon Bond chemistry scheme (CB05) provides good model capability on tropospheric composition , the stratosphere is only realistically constrained in terms of ozone. This is because so far the model 20 ozone is based on a linear scheme (Cario lle and Tyssèdre, 2007) which is suitable owing to the data -assimilat ion capabilities of C-IFS of both total column and profile satellite retrievals Lefever et al., 2015).
Also it is recognized that the applicability of radiation feedbacks of trace gases, such as ozone and water vapour, as produced through CH 4 o xidation, are hampered by schemes that are based on linearizations (Cario lle and Morcrette, 2006;de Grandpré et al., 2009), This is due to the intrinsic dependencies to climatologies which are used to construct such schemes and hence 25 they may behave poorly in anomalous situations. Having full stratospheric chemistry available in the IFS therefore would not only allo w to study a wider range of at mospheric composition processes, but also a more independent assessment of radiation feedbacks on temperature, hence providing the potential for improvements in stratospheric and tropospheric meteorology. These considerations drive the need for extension of C-IFS with a module fo r stratospheric chemistry. For this we use the chemistry scheme fro m the Belg ian Assimilat ion System for Chemical ObsErvations (BASCOE), Errera et al. 30 (2008), which was developed to assimilate satellite observatio ns of stratospheric composition.
BASCOE is based on a Chemistry Transport Model (CTM ) of the stratosphere which is used to investigate stratospheric photochemistry (Theys et al., 2010;Muncaster et al., 2012). The assimilat ion system uses the 4D-VA R algorithm (Talagrand and Courtier, 1987) to produce reanalyses of stratospheric co mposition (Viscardy et al., 2010) wh ich co mpare favourably well with similar systems (Geer et al., 2006;Thornton et al., 2009) and facilitate detailed studies of transport proces ses in the stratosphere (Lahoz et al., 2011). The photochemistry module fro m the BASCOE-CTM was implemented into the Canadian assimilation system GEM, demonstrating the potential of a coupled chemical-dynamical assimilation system for stratospheric studies (de Grandpré et al., 2009;Rob ichaud et al., 2010). BASCOE has been used and evaluated within the framework of MACC as an independent system for the provision of Near Real-Time analyses of stratospheric o zone and for 5 the validation of the corresponding product by the main assimilation system (Lefever et al., 2015;.
The CB05 tropospheric scheme has been combined with the stratospheric scheme fro m BASCOE -CTM to form a single chemistry mechanis m that encompasses tropospheric and stratospheric chemistry throughout the atmosphere , here referred to as C-IFS-Atmos. However, this approach appears computationally expensive, due to the extended chemical mechanism.
Therefore we have developed an approach for an optimized merg ing of the CB05 tropospheric chemistry scheme and the 10 stratospheric chemistry scheme used in BASCOE within C-IFS. An assessment of the two chemistry schemes showed that there is only part ial overlap in t race gases and reactions that are essential in both regimes. For instance, 15 out of the fu ll list of 99 trace gases need to be treated in the chemical mechanisms for both troposphere and stratosphere. Also the modelling of the photolysis rates and heterogeneous reactions have been optimized for application in t roposphere and stratosphere separately. In this optimized approach we developed a flexible setup where -within a single framework-either the 15 tropospheric or stratospheric chemistry modules are address ed, referred to as C-IFS-TS. In this approach the parameterizations for the chemistry, including the respective chemistry mechanisms as optimized for troposphere and stratosphere separately, are retained.
In this paper we describe two merging approaches and provide benchmark evaluations of the C-IFS-Atmos and C-IFS-TS systems with focus on the stratospheric composition. The ancestor BASCOE-CTM is also included in the comparison 20 through a forward model run (without chemical data assimilat ion), in order to provide insight in the differences caused by the treatment of transport between C-IFS and BASCOE. The paper is organized as follows: In Sect 2 the chemistry modules for the stratosphere are described and the merging with the tropospheric scheme is explained.. Sect ion 3 provides details on the setup of the model runs, and the observational data used for the model evaluation. Section 4 provides a basic model evaluation of the system. We finalize this manuscript with conclusions and an outlook for further work. 25

Atmospheric chemistry in C-IFS
For general aspects related to chemistry modeling in C-IFS the reader is referred to . The meteorological model in the current version of C-IFS is based on IFS cycle 41r1 (ECMWF, 2015). The advection is simu lated with a three-d imensional semi-Lagrangian advection scheme, which applies a quasi-monotonic cubic interpolation of the departure values. 30 In the fo llowing two subsections we describe the C-IFS modules fo r the stratospheric (BASCOE-based) and tropospheric (CB05-based) chemistry parameterizations, continued by a section describing the merging procedure of these two modules to form the C-IFS-TS system. The full list of trace gases is given in Table A1 in the Appendix, including the domains where they are actively treated within the chemistry.

Stratospheric chemistry
Fro m the BASCOE system (Errera et al., 2008) the chemical scheme and the parameterization for Polar Stratospheric Clouds (PSC) has been imp lemented in C-IFS. The BASCOE chemical scheme used here is labelled "sb14a". It includes 58 species 5 interacting through 142 gas -phase, 9 heterogeneous and 52 photolytic reactions. This chemical scheme merges the reaction lists developed by Errera and Fonteyn (2001) to produce short -term analyses, with the list included in the SOCRATES 2-D model for long-term studies of the middle at mosphere (Brasseur et al., 2000;Chabrillat and Fonteyn, 2003). The resulting list of species (see Table A1) includes all the ozone-depleting substances and greenhouse gases necessary for mu lti-decadal simu lations of the couplings between dynamics and chemistry in the stratosphere, as well as the reservoir and short -lived 10 species necessary for a complete description of stratospheric ozone photochemistry.
Gas-phase and heterogeneous reaction rates are taken fro m JPL evaluation 17 (Sander et al., 2011) and JPL evaluation 13 (Sander et al., 2000), respectively. Lookup tables of photolysis rates were co mputed offline by the TUV package (Madronich and Flocke, 1999) as a function of log-pressure altitude, ozone overhead column and solar zenith angle. The photolysis tables used in chemical scheme sb14a are based on absorption cross -sections from JPL evaluation 15 . 15 The kinetic rates for heterogeneous chemistry are determined by the parameterization of Fonteyn and Larsen (1996), using classical expressions for the uptake coefficients on sulfate aerosols (Hanson and Ravishankara, 1994) and on Polar Stratospheric Clouds (PSCs) (Sander et al., 2000).
The surface area density of stratospheric aerosols uses an aerosol number density climatology based on SA GE-II observations (Hitch man et al., 1994). Ice PSCs are presu med to exist at any grid point in the winter/spring polar reg ions 20 where water vapour partial pressure exceeds the vapour pressure of water ice (Murphy and Koop, 2005).
Nitric Acid Tri-hydrate (NAT) PSCs are assumed when the nitric acid (HNO 3 ) partial pressure exceeds the vapour pressure of condensed HNO 3 at the surface of NAT PSC particles (Hanson and Mauersberger, 1988). The surface area density is set to 2×10 −6 cm 2 /cm 3 for ice PSCs and 2×10 −7 cm 2 /cm 3 for NAT PSCs. The sedimentation of PSC particles causes denitrification and dehydration. This process is approximated by an exponential decay of HNO 3 with a characteristic t ime-25 scale of 20 days for gridpoints where NAT particles are supposed to exist, and an exponential decay of HNO 3 and H 2 O with a characteristic time-scale of 9 days for gridpoints where ice particles are supposed to exist. Mass mixing ratios for N 2 O, CO 2 and a selection of anthropogenic and organic halogenic trace gases are constrained at the surface by a global mean constant value, Table 1. Assuming that trace gases are well mixed in the troposphere, this essentially serves as lower boundary conditions for the stratospheric chemistry. 30

Tropospheric chemistry
The tropospheric chemistry in the C-IFS is based on the CB05 mechanis m (Yarwood et al., 2005). It adopts a lu mping approach for organic species by defining a separate tracer species for specific types of functional groups. The scheme has been modified and extended to include an explicit treat ment of C1 to C3 species as described in W illiams et al., (2013), and SO 2 , d i-methyl sulphide (DM S), methyl sulphonic acid (MSA) and ammonia (NH3) (Hu ijnen et al., 2010). A coupling to the 5 MACC aerosol model is availab le (Huijnen et al., 2014), but not switched on for this study. The reaction rates follow the recommendations given in either Sander et al. (2011) or Atkinson et al. (2006) . The modified band approach (M BA), wh ich is adopted for the computation of photolysis rates (Williams et al., 2012), uses 7 absorption bands across the spectral range 202 − 695 n m. At instances of large solar zenith angles (71-85°) a different set of band intervals is used. In the MBA the radiative transfer calcu lation using the absorption and scattering components introduced by gases, aerosols and clouds is 10 computed on-line for each of the predefined band intervals. The co mplete chemical mechanism as applied for the troposphere is extensively documented in . There a specification of the emissions and deposition of tropospheric reactive trace gases is provided as well.

Merging procedures for the tropospheric and stratospheric chemistry
Here we investigate two options to merge tropospheric and stratospheric chemistry, as also summarized in Table 2. The 15 chemistry mechanis m for C-IFS-At mos is composed by simply co mbin ing the reaction mechanisms for troposphere and stratosphere into one large mechanism, removing reactions that are duplicated. In contrast to this model version here we propose an approach for a more efficient merging of the chemistry modules for the troposphere and stratosphere to form the C-IFS-TS system. Key chemical cycles differ between troposphere and stratosphere, hence allowing different chemical mechanis ms. For examp le, the o xidation of non-methane hydrocarbons (NMHC's) is essentially taking place in the 20 troposphere and represents an important driver for tropospheric O 3 production. The chemical evolution of PAN and organic nitrate can be neglected in the stratosphere. On the other hand, N 2 O and CFC's are essentially chemically inactive in the troposphere and will only be photolysed by UV radiation in the stratosphere. Therefore, the chemical reactions involving these gases do not need to be accounted for in the troposphere. . Also the parameterization of the photolysis rates leads to different requirements for the troposphere and stratosphere, as will be discussed in the next subsection. Finally the nu merical 25 solver of the chemical mechanism contributes substantially to the total costs of the model run in terms o f run -time, depending on the size of the reaction mechanis m. These elements have motivated us to divide the chemistry in the C-IFS-TS system into a tropospheric and stratospheric part. Note that there is only one set of transported atmospheric trace gases and only the position of the grid box above or below the tropopause determines if the tropospheric or stratospheric chemistry is applied. 30 The tropopause can be defined based on a different criteria. A co mmon approach is to use dynamical criterion such as the isentropic potential vorticity (e.g., Thuburn and Craig, 1997) but this fails in regions of small absolute vorticity, notably in the tropics. A definition based on the lapse rate (WMO, 1957) is an alternative, but may not be well defined in the presence of mu ltiple stable layers. We therefore choose to base our criterion on the chemical co mposition of the at mosphere considering that the tropopause is associated with sharp gradients in trace gases (e.g., Gaudel et al., 2015). Th is has the advantage that parcels with t ropospheric/stratospheric co mposition can be traced dynamically, and the most approp riate chemistry scheme can be adopted to it. In our simu lation we use a chemical defin ition of the t ropopause level, where 5 tropospheric grid cells are defined at O 3 <200 ppb and CO>40 ppb, for P > 40 hPa. With this definit ion the associated tropopause pressure ranges in practice between approx. 270 and 80 hPa for sub-tropics and tropics, respectively.
For both troposphere (CB05) and stratosphere (BASCOE) the nu merical solver is generated using the Kinetic Pre -Processor (KPP, Sandu and Sander, 2006) software. Specifically we adopt the standard four-stages, third order Rosenbrock solver (Rodas3). This is different fro m the Eulerian backward imp licit solver as used in , and is motivated by 10 the improved coding flexibility and accuracy.
Most of the gas phase reactions that take place both in the troposphere and stratosphere, such as NO x and HO x reactions, are simu lated in identical ways in both chemistry schemes. It is worth mentioning that the constituents O 1 D and O 3 P, produced fro m O 3 and O 2 photolysis, are not explicitly co mputed in the troposphere, as O 1 D and O 3 P are assumed to react with O 2 , O 3 and N 2 only. This is different for the stratosphere, where O 1 D and O 3 P are involved in many reactions . For trace gases 15 whose chemistry is currently neglected in the stratosphere (the NMHC's, PA N, Organic nitrate, SO 2 ) we adopt a 10-day decay rate to prevent their spurious accumulation in the stratosphere. Hence these losses are currently not accounted for in the stratospheric chemical mechanis ms and do not contribute either to the load of stratospheric aerosols. Note that tropospheric halogen chemistry, wh ich contributes to near-surface ozone depletion in spring-t ime polar region and to The four options to run this type of C-IFS experiments and the computational costs are given in Table 2. As compared to the 25 C-IFS-T experiments, the costs of running an experiment including full stratospheric chemistry with the C-IFS-TS system have increased by ~50%. Most of this increase is caused by the computation of the chemistry and not the tracer transport due to the efficiency of the semi-Lagrangian advection scheme for mu ltip le tracers. The C-IFS-At mos setup where tropospheric and stratospheric chemistry were merged into a single reaction mechanism, led to an increase in costs by ~50% co mpared to C-IFS-TS, indicat ing the benefit of having separate solver codes for tropospheric and stratospheric chemistry. The C-IFS-TS 30 implementation allows for an easy switch between system setups compared to the C-IFS-At mos imp lementation. For completeness also specifications of the BASCOE-CTM are p rovided in Table 2, wh ich is identical in terms of stratospheric chemistry parameterization co mpared to C-IFS-TS and C-IFS-S. Clearly the essential difference co mpared to the C-IFS setup refers to the fact that BASCOE is used here as a CTM, while C -IFS is a GCM. Most notably this implies a different advection treatment and a different horizontal grid (see s ection 3).

Merging photolysis rates
For parameterization of the photolysis rates the Modified Band Approach (MBA, Williams et al., 2012) and the lookup table approach (Errera and Fonteyn, 2001) are retained, see Table 3, as these have been optimized in the past for applications in 5 the troposphere and stratosphere, respectively. While for tropospheric conditions scattering and absorption properties of clouds and aerosol strongly impact the magnitude of photolysis rates and hence the local chemical co mposition, this is of less relevance in the stratosphere. Wavelengths shorter than 202 n m, on the other hand, are largely blocked by stratospheric ozone and oxygen and do not contribute to radiation in the troposphere (Williams et al., 2012). At higher alt itudes these short wavelengths contribute to the Chapman cycle and to the break down of CH 4 , CFC's and N 2 O either direct ly or through 10 oxidation by O 1 D. A lso the presence of sunlight at solar zenith angles (SZA) larger than 90° at high altitudes needs to be accounted for in the stratosphere due to the Earth's curvature. This plays a role in the timing of springtime ozone depletion in the polar lower stratosphere, but may be neglected in the troposphere. Table 4 lists the photolysis rates that are active both in the troposphere and stratosphere. Photolysis rates for reactions occurring both in the troposphere and stratosphere are merged at the interface, in order to ensure a smooth transition betwee n 15 the two schemes. This is done by an interpolation at four model levels around the interface level between both parameterizations, for SZA<85°. For larger SZA the original value for the photolysis rate is retained in case of stratospheric chemistry, while it is switched off for the troposphere.
Note that even though the reaction rates have been merged, the products from the same photolytic reactions are sometimes different as a consequence of the different reaction mechanisms between the troposphere and stratosphere. 20 An examp le of the merg ing strategy is given in Fig. 1. It shows that at the interface for J O 3 and J NO 2 on average a small increase of the merged photolysis rate is seen towards lower altitudes, with the switch to MBA in the troposphere, which is a consequence of the co mbination of d ifferences in the parameterizat ions. Even though such jumps are undesirable, no visible impact on local chemical co mposition was found, for any of the trace gases involved in both tropospheric and stratospheric chemistry, see also Figures S1-S3 in the Supplementary Material. Th is can be explained by the sufficiently small d ifference 25 in the photolysis rates at the merging altitude of the photolysis and chemistry schemes, combined with the sufficiently long lifetime of the affected trace gases.

Tracer transport settings
Tracer transport is treated identically for all ind ividual chemical trace gases. Since the semi-Lagrangian advection does not formally conserve mass de Grandpré et al., 2016) a global mass fixer is applied (Diamantakis 30 and Fle mming, 2014) to all but a few constituents, including NO, NO 2 and H 2 O. Rather than conserving mass during the advection step of the individual co mponents NO and NO 2 , this is enforced to a stratospheric NO x t racer defined as the sum of NO and NO 2 . While a chemical H 2 O trace gas is defined in the full atmosphere, in the troposphere H 2 O mass mixing ratios are constrained by the humid ity (q) simulated in the meteorological model in IFS and provide a boundary condition for water vapour in the stratosphere. Stratospheric H 2 O (i.e. above the tropopause level) is governed by chemical production and loss.
The global advection errors in H 2 O that essentially originate fro m the tropospheric part because by far most H 2 O mass is located in the t roposphere and the spatial gradients are much more pronounced. This should not affect the stratospheric H 2 O 5 mass budget, herefore the global mass fixer for the stratospheric H 2 O tracer has been switched off. Th is prevents spurious trends in stratospheric H 2 O co lu mns over the years (not shown), indicat ing that H 2 O mass conservation is well ensured in the stratosphere.

Model setup and observations used
We have executed runs with C-IFS-TS and C-IFS-Atmos for the period April 2008 until December2010. Stratospheric ozone 10 in C-IFS-TS is further compared to that of the C-IFS-T system . This uses the ECMWF standard linear o zone scheme (version 2a, Cariolle and Teyssèdre, 2007) in the stratosphere , wh ile stratospheric HNO 3 is constrained through a climatological ratio of HNO 3 /O 3 at 10 hPa .
We have initialized all C-IFS runs on 1 April 2008 using assimilated concentration fields fro m the BASCOE system in the stratosphere for this date. The horizontal resolution of these runs is T255 (i.e. appro x. 0.7° lon / lat) with 60 levels in the 15 vertical. Meteorology in the C-IFS runs is relaxed towards ERA-Interim.
Intercomparison of the runs C-IFS-TS and C-IFS-At mos aims to provide a justification of our approach to split the chemistry into two regions, wh ile interco mparison of C-IFS-TS with C-IFS-T can be used to identify the changes to stratospheric composition modelling between full stratospheric chemistry and the baseline approach with the linear ozone scheme.
The performance of the C-IFS runs has further been compared against the BASCOE-CTM (without chemical data 20 assimilation), using the same chemical mechanism and parameterizat ions for photolysis and heterogeneous chemistry as implemented in the C-IFS-TS. This serves as a model reference for the C-IFS implementation of stratospheric chemistry.
While C-IFS evaluates tracer transport on a reduced Gaussian grid, the BASCOE-CTM uses a regular lat itude-longitude grid. It is run here with a resolution of 1.125° lon / lat similar to the resolution chosen for C-IFS used, and on the same vertical grid of 60 levels. The BASCOE-CTM is driven by temperature, pressure and wind fields simu lated by the C-IFS 25 runs. However, wh ile BASCOE adopts a flu x-form advection scheme (Lin and Rood, 1996) the IFS uses the Semi-Lagrangian scheme for advection, accounts for vertical diffusion and includes a parameterization for convection (ECMWF, 2015). Using essentially the same dynamical fields together with an identical implementation of the chemistry code should allo w to identify differences due to the different transport schemes between C-IFS and the BASCOE-CTM. Co mmon chemical biases between both systems also point at issues in the chemical parameterization s such as reaction mechanism, 30 photolysis, heterogeneous chemistry and sedimentation.

Observational data used for validation
We evaluate the C-IFS runs in terms of stratospheric O 3 , NO 2 , N 2 O, CH 4 , H 2 O and HCl, and for this purpose use a range of observation-based products.
Model results are compared with retrievals (version 3) of O 3 , (Fro idevaux et al., 2008a), ClO (Santee et al., 2008), H 2 O (Read et al., 2007) and HCl (Froidevaux et al., 2008b)  Ozone Monitoring Instrument (OMI) observations. The satellite retrieval products used in the MSR are bias -corrected with respect to Brewer and Dobson Spectrophotometers to remove discrepancies between the different satellite data sets. The uncertainty in the product, as quantified by the bias of the observation -minus-analysis statistics, is in general less than 1 DU. 20 O 3 profiles are co mpared to ozonesonde data that are acquired fro m the World Ozone and Ultavio let radiat ion Data Centre (WOUDC). The precision of the o zonesondes is on the order of 5% in the stratosphere (Hassler et al., 2015), when based on electrochemical concentration cell (ECC) devices (~85% of all soundings). Larger random erro rs (5 -10%) are found for other sonde types, and in the presence of steep gradients and where the ozone amount is lo w. Sondes at 19, 12, 2 and 1 individual stations are used for the evaluation over northern hemisphere mid latitudes, tropics, southern hemisphere mid latitudes and 25 Antarctic, respectively.
Stratospheric NO 2 co lu mns are co mpared to observational data fro m the SCIAMACHY (Bovensmann et al., 1999) UV-VIS (ultravio let-visible) and NIR (near-in frared) sensor onboard the Envisat satellite. The satellite ret rievals are based on applying the Differential Optical Absorption Spectroscopy (DOAS) (Platt and Stutz, 2008) method to a 425-450 n m wavelength window. Stratospheric NO 2 colu mns from SCIAMACHY presented here are in fact total columns derived by 30 dividing retrieved slant co lu mns of NO 2 by a stratospheric air mass factor and contains data over the clean Pacific ocean (180°E -220°E) only (Richter et al., 2005). A lthough in this region the contribution of the troposphere to total colu mn NO 2 is small, stratospheric colu mn NO 2 fro m SCIAMACHY is still somewhat positively biased by a tropospheric contribution. However, stratospheric air mass factors for NO 2 are usually large compared to tropospheric ones, so that the uncertainty resulting from this should only have a minor impact on the data analysis presented in this study.
Monthly mean stratospheric NO 2 colu mns are associated with relative uncertainties of roughly 5-10% and an additional absolute uncertainty of 1×10 14 molec cm -2 . To account for d ifferences in observation and model output time, simulat ions are interpolated linearly to the equator crossing time of SCIAMACHY (10:00 LT). In addition, only model data for wh ich 5 satellite observations exist are included in the corresponding comparisons. ACE-FTS N 2 O observations between 6-30 km agree to with in 15% of independent observations, while above they agree to within ±4 ppbv (Strong et al., 2008). The uncertainty in ACE-FTS CH 4 observations is within 10% in the upper t roposphere -lower stratosphere, and within 25% in the middle and higher stratosphere up to the lower mesosphere (<60 km) (De

Mazière et al. 2008). 15
Three-hourly C-IFS and BASCOE-CTM output has been interpolated in space and time to match with any of these observations. overestimation of 50 DU (20%) is found. In the Tropics the bias is much smaller, with a slight underestimation (10 DU, 5%).

Model evaluation
In the Antarctic, the results are remarkably good during the ozone hole episodes but there is a serious overestimation developing from Feb ruary until the beginning of August, when it reaches 50 DU (30%) i.e. as much as in the Arctic. CIFS-Atmos and CIFS-TS provide very similar results over the full t ime period, suggesting that our approach to keep two different solvers in each region is valid for stratospheric ozone. Also after an initialization period of a few months the model runs do 25 not present any obvious drift, and the differences with BA SCOE-CTM are very s mall. This implies that differences due to the model configuration regard ing transport are not crucial for lower stratospheric ozone at these timescales. In the Tropics the C-IFS-TS and C-IFS-At mos results are slightly better than those with BASCOE-CTM, potentially due to the missing parameterization for convection. In the Antarctic, the parameterization of PSC leads to an overestimat ion of springtime ozone depletion while the Cariolle parameterization simu lates very well the lowest columnar values observed in September , 30 as discussed in more detail below. The recovery of ozone is overestimated by 20DU (10%) in December-January. While the amplitude of the annual cycle is overestimated above the Antarctic, its structure matches well the observations. An evaluation of O 3 total co lu mns (TC) against the MSR at various latitude bands is given in Figure S6  A closer analysis of the processes responsible for springtime polar ozone depletion is given in Fig. 6. In both the C-IFS-TS and C-IFS-At mos runs as well as BASCOE-CTM there is an HNO 3 deficit at the beginning of the winter. Denit rificat ion, which is not modelled in C-IFS-T, starts at the correct time in the models with stratospheric chemistry indicating that NAT PSC appear at about the right time. However, denit rificat ion proceeds more slowly and ends one month later than observed 30 by Aura-MLS. We attribute this shortcoming to the crude modelling of NAT PSC which does not calculate the amount of condensed nitric acid and water, keeps the surface area densities of PSC part icles fixed at an arbitrary value and parameterizes sedimentation through irreversible removal. Ch lorine activation starts at exactly the right time and is as strong in the C-IFS runs as in the Aura-MLS observations until the beginning of September, but starts decreasing afterwards while it lasts two mo re weeks in the observations. Hence the overestimat ion of ozone deplet ion during August and September in the models with e xp licit stratospheric chemistry is probably not due to an overestimation of chemical removal. This feature is more pronounced in CIFS-TS and CIFS-At mos than in the BASCOE-CTM, suggesting that it may be associated to differences in the modelling of transport.
The evaluation of the zonal mean ozone mixing ratios against MIPAS observations shows good general agreement, Fig. 7 The assessment of NO 2 against MIPAS daytime NO 2 observations, acquired by samp ling the ascending orbits fro m Envisat, shows good agreement with the models, although C-IFS-TS and C-IFS-Atmos tend to show a positive bias . The C-IFS-TS and C-IFS-At mos runs describe well the seasonal variation in zonal mean stratospheric NO 2 columns at different latitude bands, Fig. 8, with monthly mean biases with respect to the SCIAMACHY observations of less than 1 × 10 15 molec cm -2 in the tropics and at mid-latitudes. The positive bias is larger in C-IFS-At mos than C-IFS-TS. In contrast, poor performance can 15 be seen for C-IFS-T, due to the lack of stratospheric NO x chemistry in that version.
However, a positive NO 2 bias with respect to night-time MIPAS NO 2 observations appears larger for C-IFS-TS and C-IFS-Atmos than for the BASCOE-CTM (Fig . 7). In contrast, this figure also shows a negative bias in HNO 3 with respect to MIPAS observations in the BASCOE-CTM, and C-IFS-Atmos, and even more marked in the C-IFS-TS experiment. Even though a clear improvement compared to run C-IFS-T is found, further investigation is necessary to diagnose the origins of 20 the biases in night-time NO 2 above 10 hPa and in HNO 3 between 10 and 70 hPa. their long lifet imes these trace gases are good markers for the model ability to describe transport processes -i.e. not only the Brewer-Dobson circulat ion but also isentropic mixing, mixing barriers, descent in the polar vortex, and stratospheretroposphere exchange (Shepherd, 2007). Moreover, N 2 O is the main source of reactive nitrogen in the stratosphere while 25 CH 4 is one of the main precursors for stratospheric water vapour. The figure suggests reasonable profile shapes for both CH 4 and N 2 O in the upper stratosphere (10 h Pa and higher) where their abundance is more strongly influenced by chemical loss but at lower altitudes (100-10 hPa) C-IFS-TS and C-IFS-Atmos show larger discrepancies to the observations than the BASCOE-CTM run, with weaker vert ical grad ients in the tropics and SH-mid latitudes and a sharper gradient in the extratropical Northern Hemisphere. 30 This discrepancy cannot be due to different wind fields because the BASC OE-CTM experiment is driven by three-hourly output of the C-IFS experiment. We attribute it instead to the different nu merical schemes for advection and/or to differences in the representation of sub-grid transport processes in the GCM and in the CTM . Convection and diffusion are indeed explicit ly modelled in C-IFS but neglected in BASCOE CTM , which relies on the implicit diffusion properties of its flu x-stratospheric o zone is strongly determined by both chemistry and transport, t he transport issue indicated by fig. 9 could also contribute directly to the ozone biases seen below 10 hPa in Figures 3 and 4.   Fig. 10 shows a good consistency between H 2 O modelled by C-IFS-TS and the BASCOE-CTM results, albeit with a slight negative bias with respect to MLS observations above 5 hPa, and a positive bias around 30 h Pa in the tropics, associated 5 with corresponding biases in CH 4 . This figure also shows globally a good agreement between HCl modelled by C-IFS-TS and MLS observations, although with a positive bias of 0.8 ppbv confined in the region of ozone depletion above Antarctica.

Conclusions
We have presented a model description and benchmark evaluation of an extension of the C-IFS system with stratospheric ozone chemistry of the BASCOE model added to the already existing tropospheric scheme CB05. We refer to this system as 10 C-IFS-CB05-BASCOE, or C-IFS-TS in short. In our approach we have retained a separate treat ment for tropospheric and stratospheric chemistry, and select the most appropriate scheme depending on the altitude with respect to the tropopause level. This has the advantage that mechanisms which are optimized fo r tropospheric and stratospheric chemistry, respectively, can be retained, which also substantially reduces the computational costs of the chemical solver co mpared to an approach where all reactions are activated in the whole atmosphere, referred to as C-IFS-At mos. Also, it allows for an easy 15 switch between system setups. To avoid ju mps in t race gas concentrations at the interface the consistency in gas -phase reaction rates has been verified while the photolysis rates from the two parameterizations are in terpolated across the interface. We showed that differences between C-IFS-TS and C-IFS-At mos are overall s mall, hence our basic assumption to have different chemistry solvers for troposphere and stratosphere is valid for our applications.
An evaluation of a 2.5 year simu lation of C-IFS-TS indicates good performance of the system in terms of stratospheric 20 ozone, of similar quality as its ancestor BASCOE-CTM model results, and a considerable general improvement in terms of stratospheric composition co mpared to the C-IFS-T predecessor model version wh ich applied a linear o zone scheme in the stratosphere.
The O 3 partial colu mns  show biases mostly smaller than ±20 DU when co mpared to the Aura MLS observations. Also the profiles were generally well captured, and show an improvement with respect to the C-IFS-T linear 25 ozone scheme in the stratosphere over mid-latitudes. The depth and variability of the ozone hole over Antarctica is modelled well. While also the C-IFS-T shows a remarkably good agreement to the observations during the ozone hole episodes it develops a significant overestimation of the partial co lu mns during other months . The tropical maximu m of the mixing ratio, around 10 hPa, is the only stratospheric region where C-IFS-T agrees better all-year-long with observations. Also evaluation of other trace gases (NO 2 , HNO 3 , CH 4 , N 2 O, HCl) against observations derived fro m various satellite 30 retrievals (SCIAMACHY, ACE-FTS, MIPAS, M LS) illustrate the clear improvements obtained with C-IFS-TS co mpared to C-IFS-T, even though C-IFS-TS still suffers fro m positive biases in stratospheric NO 2 , whereas HNO 3 is b iased low. For the long-lived tracers CH 4 and N 2 O, larger errors with respect to limb -sounding retrievals were found between 10 hPa and 100 hPa than with the BASCOE-CTM, suggesting difficulties in representing slow transport processes . The BASCOE-CTM experiment shown here was driven by three-hourly wind fields output of the C-IFS experiments. Hence this discrepancy is due to a difference in the representation of the transport processes between the GCM and the CTM , i.e. the nu merical scheme used for advection (Semi-Lag rangian versus Flu x-form), the convection (parameterized in C-IFS but neglected in 5 BASCOE CTM) or the diffusion (parameterized in C-IFS but not explicitly considered in the CTM ). Hence, stratospheric transport in C-IFS will be an area for further evaluation and developments.
This benchmark model evaluation of C-IFS-TS marks a key step towards merging tropospheric and stratospheric chemistry within IFS, aiming at a possible configuration for daily operational forecasts of lower and middle at mospheric composition in the near future. Future work could focus on the following aspects: 10 -Chemical data-assimilation: in itial tests with data-assimilation of O 3 total colu mn and profile retrievals suggest that stratospheric ozone is successfully constrained in C-IFS-TS. Ho wever, observational constraints on other components driving o zone chemistry are currently lacking in the assimilat ion system. Our extension opens the possibility for assimilation of additional trace gases such as N 2 O and HCl. However, for the 4D-VA R assimilation of short-lived species such as NO 2 and ClO an adjoint chemistry module would likely be required as implemented the BASCOE DA system. 15 -Align ment of the reaction mechanism and photolysis rates: while at current stage the gas -phase and photolytic reaction rates of the parent schemes are retained, we foresee a further integration to ensure better align ment of the chemical mechanis ms. Especially the existing ju mps in photolysis rates as a consequence of the different parameterizations are not desirable, even though they are not harmfu l for model stability nor visib ly lead to any degradation in model performance.
The align ment in terms of gas -phase reaction rate expressions can be achieved by the introduction of the KPP solver in C-20 IFS, for both tropospheric and stratospheric chemistry, wh ich allo ws for a better traceable model develop ment than the hardcoded Euler Backward Integration solver as adopted in .
-Improvement of the representation of stratospheric sulphate aerosols and Polar St ratospheric Clouds: the current climatology for these aerosols, and parameterization for PSCs could easily be improved. While the current results are satisfactory for a general-purpose monitoring system, these improvements would especially allow better simulat ions of the 25 composition in in the polar lower stratosphere during springtime.
-Extension of tropospheric and stratospheric chemistry schemes: the availability of a co mprehensive set of trace gas fields allo ws for a relatively easy extension of the tropospheric reaction mechanism by including selective reactions originating fro m the stratospheric chemistry, and vice versa. Examples are the introduction of halogen chemistry in the troposphere (von Glasow and Crut zen, 2007), or SO2 conversion to sulphate aerosol in the stratosphere, relevant in case of strong volcanic 30 events (Bândă, et al., 2015).
-Optimization of solver efficiency: even though the use of KPP has simp lified the code maintenance and may result in a higher numerical accuracy of the solution, it also caused a considerable slow-down of the numerical efficiency as compared to the Euler Backward Integration solver, as that solver had been optimized for tropospheric ozone chemistry in C-IFS-chemical and physical conditions, and an optimizat ion of the automated solver code, which allows for a more efficient code structure (KP4, Jöckel et al., 2010).
In summary, the extension towards stratospheric chemistry in C-IFS broadens its ability for forecast and assimilat ion of stratospheric composition, which is beneficial to the monitoring capabilities in CAMS, and may also contribute to advances 5 in meteorological forecasting of the ECMWF IFS model in the future.

Code availability
The C-IFS source code is integrated into ECWMF's IFS code, wh ich is available subject to a lic ence agreement with ECMWF, see also  for details. The stratospheric chemistry module of C-IFS was originally developed in the framework of BASCOE. Readers interested in the BASCOE code can contact the developers through 10 http://bascoe.oma.be. 1.70E-10 3.30E-12 4.62E-12 9.08E-12 1.17E-12 5.44E-10 3.80E-4 Table 2. Number of trace gases, the chemistry scheme in troposphere and stratosphere, and corresponding number of reactions (gas-phase 5 / heterogeneous and photolytic), as well as specification of the circulation model and computational expenses of a one-month run on T255L60 in terms of system billing units (SBU) for various C-IFS model vers ions. For completeness also the BASCOE-CTM system is indicated.  Table 3. Parameterization of photolysis rates for troposphere (CB05-based) and stratosphere (BASCOE-based)