The atmospheric chemistry box model CAABA/MECCA-4.0gmdd

. We present version 4.0gmdd 1 of the atmospheric chemistry box model CAABA/MECCA which now includes a number of new features: (i) skeletal mechanism reduction, (ii) the MOM chemical mechanism for volatile organic compounds, (iii) an option to include reactions from the Master Chemical Mechanism (MCM) and other chemical mechanisms, (iv) updated isotope tagging, and (v) improved and new photolysis modules (JVAL, graph search calculates a path interaction coefﬁcient (PIC) based on the product of direct interaction coefﬁcients along the path from target to species, where nodes represent species and weighted directed edges represent DICs. Finally, the overall interaction coefﬁcient (OIC) is the maximum of all PICs between target and species. It is calculated for all sample points and expressed as a value between 0 (unimportant) and 1 (important). For target species, OIC = 1 by deﬁnition. OIC values are only calculated for the full mechanism.


Introduction
A full description of the multi-purpose atmospheric chemistry box model CAABA/MECCA (Chemistry As A Boxmodel 10 Application / Module Efficiently Calculating the Chemistry of the Atmosphere) has already been published elsewhere (Sander et al., 2005(Sander et al., , 2011a. Here, we only present new features that have been implemented after version 3.0. Section 2 describes all changes related to the chemical mechanism of MECCA. In Sect. 3 we show several new options for calculating photolysis rate coefficients in the model. HONO has been added according to Bejan et al. (2006) and Chen et al. (2011). Finally, reactions of phenyl peroxy radicals with NO 2 yielding NO 3 have been added, consistent with Jagiella and Zabel (2007).
Oxidation of VOCs by O 3 and NO 3 is similar to that in the MCM. The oxidation by OH, however, significantly differs from the MCM treatment and therefore is detailed in the next section.

VOC reaction with OH
Reactions of OH with organic molecules can be either H-abstraction or OH-addition. If available, experimental rate coefficients are preferred and taken mostly from the IUPAC kinetic data evaluation (Atkinson et al., 2006). Unmeasured rate coefficients for the C 1 to C 5 species are estimated with a site-specific Structure-Activity Relationship (SAR) similar to the MCM, based on the work of Atkinson (1987) and Kwok and Atkinson (1995). The base rate coefficients for OH-addition to double bonds 5 are taken from the more recent SAR by Peeters et al. (2007). For the C 6 to C 11 species, the MCM rate coefficients are retained.
It is worth noting that the latter have no temperature-dependence and are only given at 298 K. The effect of neighbouring groups is expressed by substituent factors and is differentiated by functional group. Most substituent factors by Kwok and Atkinson (1995) are updated or calculated ex novo by computing the relative rate coefficient of OH with the simplest VOC bearing the substituent relative to the one of its parent compound. A clear limitation of this approach is that for OH-addition 10 no substituent effect on the branching ratios is considered. No rigorous evaluation of the SAR has been conducted and the estimation uncertainty is expected to be in the same range as for the SAR used by the MCM.
The general formulae for H-abstraction by OH are: where k p , k s , k t are the group rate coefficients for the hydrogens on the primary, secondary and tertiary carbon atoms, respectively, and F (X) is the factor for the substituent X.
The SAR for OH-addition to (poly)alkenes is based on the hypothesis that the site-specific rate coefficient depends solely on the stability of the radical product . Thus, rate coefficients for the formation of primary, secondary 20 and tertiary radicals are derived from the high-pressure limits for ethene, 2-butene and 2,3-dimethyl-2-butene, respectively. It's worth noting that for the tertiary radical formation, Peeters et al. (2007) used solely the rate coefficient for 2,3-dimethyl-2butene and not that for 2-methyl-2-butene minus that for the secondary radical.

RO 2 reaction with NO x
Reactions with NO are the dominant sink for RO 2 under polluted conditions. The RO 2 -size independent MCM rate coefficient 25 is used with the exception of CH 3 O 2 and CH 3 CH 2 O 2 , for which the IUPAC recommendations are followed (Atkinson et al., 2006). In general, the two possible reaction channels are considered: with α being the alkyl nitrate yield for the formation of alkyl nitrates, which curb tropospheric ozone production. Acyl RO 2

30
do not form nitrates. The CH 3 ONO 2 -yield is calculated according to Butkovskaya et al. (2012)  The constant yield of about 10 % ("old model") is used in the MCM. Flocke et al. (1998). The CH 3 CH 2 O 2 -yield is calculated according to Butkovskaya et al. (2010). For all other peroxy radicals the corresponding alkyl nitrate yields are calculated with the relationship by Arey et al. (2001), which depends on temperature, pressure and molecular size. However, the latter is represented not by the number of carbon atoms but by the number of heavy atoms (excluding the −OO moiety) according to Teng et al. (2015). The oxygen atom in β-carbonyl RO 2 is not counted. Due to disagreement in the literature, no dependence of α on the degree of RO 2 substitution (primary, secondary and tertiary) is considered. Reduction factors for β-and γ-carbonyl RO 2 are derived from Praske et al. (2015) and for bicyclic RO 2 from aromatics are derived from Elrod (2011). As an example, Figure 2 shows the predicted variable yield for the nitrate of the secondary hydroxy butyl peroxy radical. Formation and decomposition of many peroxy nitrates is considered. The equilibria of acyl peroxy nitrates with their parent RO 2 are represented as in the MCM but the JPL kinetic data (Burkholder et al., 2015) is used. Only three alkyl peroxy nitrates, The equilibrium reactions for the latter are taken from Tyndall et al. (2001), Sehested et al. (1998) and Kirchner et al. (1999). Reactions of peroxy radicals with NO 3 all produce the corresponding alkoxy radical and NO 2 : The temperature-independent rate coefficient of  (2003) and Boyd et al. (2003). The branching ratios of the OH-channel for β-carbonyl, alkoxy and bicyclic peroxy radicals are taken from Dillon and Crowley (2008), Orlando and Tyndall (2012) and Birdsall et al. (2010), respectively. A 10 % OH-yield for reactions of β-hydroxyl peroxy radicals is taken from the isoprene oxidation study of Liu et al. (2013), which is consistent 20 with the results of Groß (2013) and Paulot et al. (2009). The HO 2 reaction of the simplest acyl peroxy radical (CH 3 CO 3 ) has unique branching ratios as determined by direct OH and O 3 measurements (Groß et al., 2014). For all other acyl peroxy radicals the kinetic data for β-hydroxy acyl peroxy radicals, e.g. HOCH 2 CO 3 , are taken from Groß (2013) with the rate coefficient having the temperature dependence as recommended by IUPAC.
There is laboratory evidence for a non-negligible reaction of CH 3 O 2 with OH (Bossolasco et al., 2014): The lower limit of the rate coefficient 1.4×10 −10 cm −3 s −1 reported by Bossolasco et al. (2014) is used in MOM. This is consistent with the revised experimental value by the same lab (Assaf et al., 2016). The major reaction channel involving HO 2 elimination represents (80 ± 20) % and is set as the only channel (Assaf et al., 2017). The other possible channels are very uncertain and are therefore not included. Geosci. Model Dev. Discuss., https://doi.org /10.5194/gmd-2018-201 Manuscript under review for journal Geosci. Model Dev. Discussion started: 3 September 2018 c Author(s) 2018. CC BY 4.0 License.

RO 2 permutation reactions
The self and cross reactions of organic peroxy radicals are treated according to the permutation reaction formalism in the MCM (Jenkin et al., 1997). Every organic peroxy radical reacts in a pseudo-first-order reaction with a rate coefficient that is expressed as 5 where k self,RO2 = second-order rate coefficient of the self reaction of the organic peroxy radical, k self,CH3O2 = second-order rate coefficient of the self reaction of CH 3 O 2 , and [RO 2 ] = sum of the concentrations of all organic peroxy radicals. The formalism is a simplification of the approach by Madronich and Calvert (1990) under the assumption that the dominant co- The value of k self,CH3O2 is taken from the IUPAC recommendations. Expressions for k self,RO2 distinguish acyl from alkyl peroxy radicals. The latter are differentiated by the degree and kind of substituents close the −OO 10 moiety. The rate expressions are not from the MCM, except for β-hydroxyl radicals, and have a temperature dependence (Atkinson et al., 2006;Glover and Miller, 2005;Orlando and Tyndall, 2012).

Photo-induced reactions
The enhanced photolysis of carbonyl nitrates from isoprene is implemented according to Barnes et al. (1993) andMüller et al. (2014). The enhancement is applied to the J-values of nitrooxyacetone (NOA), nitrooxyacetaldehyde (NO3CH2CHO), 15 lumped nitrates of methyl ethyl ketone (LMEKNO3), nitrates of MVK and MACR and unsaturated C 5 -nitrooxyaldehyde from the isoprene + NO 3 reaction.
Keto-enol tautomerization of aldehydes induced by light absorption is implemented based on data for acetaldehyde (Clubb et al., 2012). The enols are in equilibrium with the corresponding aldehydes by HCOOH-catalysis (da Silva, 2010). Formic acid is then produced upon reaction of the enols with OH similarly to the simplest enol (So et al., 2014). Vinyl alcohol is also 20 produced in the photolysis of propanal.
HPALD and PACALD photolysis is according to Peeters et al. (2014) and Jenkin et al. (2015) and the subsequent photolysis of the resulting carbonyl enols (HVMK and HMAC) is treated according to Nakanishi et al. (1977) and Messaadia et al. (2015).

Other chemical mechanisms
In addition to the native chemistry mechanism of MECCA (available in the file gas.eqn), several other, independent mechanisms are now provided as well. These are decribed in the following sections. They can for example be used for mechanism intercomparison studies within the same CAABA box model. This approach ensures that any resulting differences come from the chemical mechanism, not from other parts of the model.

5
The chemical mechanisms CB05BASCOE and MOZART from the Copernicus Atmosphere Monitoring Service project (CAMS 42) have been converted to KPP format and introduced into MECCA. Both mechanisms have been compared to MOM, and the initial results are shown in Fig. 3.
In addition, the Jülich Atmospheric Mechanism (JAM002) is now also available within CAABA.

10
The CB05BASCOE scheme (Huijnen et al., 2016) is a merge of a tropospheric and stratospheric chemistry scheme. The tropospheric chemistry is based on the Carbon Bond mechanism 2005 (CB05, Yarwood et al., 2005). Here, a lumping approach is adopted 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 treatment of C 1 to C 3 species (Williams et al., 2013), SO 2 , dimethyl sulfide (DMS), methyl sulfonic acid (MSA) and ammonia (NH 3 ), as described by Huijnen et al. (2010). The reaction rates The stratospheric chemistry is based on that from the BASCOE (Belgian Assimilation System for Chemical ObsErvations) system (Errera et al., 2008) and is labelled "sb15b". 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 atmosphere (Brasseur et al., 2000;Chabrillat and Fonteyn, 2003). The list of species includes all the ozone-depleting 5 substances and greenhouse gases necessary for multi-decadal simulations of the couplings between dynamics and chemistry in the stratosphere, as well as the reservoir and short-lived species necessary for a complete description of stratospheric ozone photochemistry. Gas-phase and heterogeneous reaction rates are taken from the JPL evaluations 17 and 18 (Sander et al., 2011b;Burkholder et al., 2015). The merged reaction mechanism includes 99 species interacting through 211 gas-phase and 10 heterogeneous reactions. Details regarding its implementation and evaluation within the ECMWF Integrated Forecasting

MOZART
The tropospheric chemistry in MOZART is based on the MOZART-3 mechanism by Kinnison et al. (2007). It includes additional species and reactions from MOZART-4 (Emmons et al., 2010) and further updates from the Community Atmosphere Model with interactive chemistry, referred to as CAM4-chem (Lamarque et al., 2012). The chemical mechanism includes an 15 updated isoprene oxidation scheme and a better treatment of volatile organic compounds, with lumped species to represent large alkanes, alkenes and aromatic compounds as well as their oxidation products. Overall, it includes the degradation of C 1 , C 2 , C 3 , C 4 , C 5 , C 7 , and C 10 species. The heterogeneous chemistry in the troposphere is implemented according to the corresponding module from CB05BASCOE. MOZART includes the extended stratospheric chemistry discussed by Kinnison et al. (2007) with further updates from CAM4-chem (Lamarque et al., 2012;Tilmes et al., 2016). This includes detailed 20 gas-phase halogen chemistry of chlorine and bromine. The stratospheric chemistry accounts for heterogeneous processes on liquid sulfate aerosols and polar stratospheric clouds, following the approach of Considine et al. (2000). Overall, the MOZART mechanism includes 117 gas-phase species, 65 photolysis and 247 gas-phase reactions. Rate coefficients are taken from the JPL recommendations (Sander et al., , 2011b.

25
Version 2 of the Jülich Atmospheric Mechanism (JAM002) has been implemented in the ECHAM-HAMMOZ chemistryclimate model (Schultz et al., 2018). It is a blend of the stratospheric chemistry scheme of the Whole Atmosphere Chemistry Climate Model (WACCM, Kinnison et al., 2007)  In contrast to the MCM, JAM002 does not use a radical pool but instead follows the pathways of peroxy radical reactions with HO 2 , CH 3 O 2 , and CH 3 COO 2 (peroxy acetyl) as explicitly as possible. Inorganic tropospheric chemistry considers ozone,

MCM
The Master Chemical Mechanism (MCM) describes in detail the tropospheric degradation of more than a hundred VOCs (Jenkin et al., 1997;Saunders et al., 2003). It is widely used as the reference mechanism for modeling studies of atmospheric 10 processes. Although the standard organic chemistry mechanism in MECCA (MOM, described above) is sufficient for many model applications, a more explicit mechanism can be necessary when studying specific VOCs. For example, the fate of limonene (C 10 H 16 ) emitted from boreal forests is not included in the standard MECCA mechanism. To use the MCM reactions inside MECCA, the new tool xmcm2mecca has been added, which converts an extracted subset from the MCM web page 2 to a KPP equation file that is compatible with MECCA. The User Manual provides a detailed description of this new tool.

Skeletal mechanism reduction
In the area of fuel combustion research, chemical models require highly complex mechanisms to describe ignition, flame propagation, and other properties. In order to save computing time, several methods have been developed to create a simplified chemical mechanism (called skeletal mechanism), which still produces similar results as the full mechanism (e.g., Tomlin and Turányi, 2013). One of these methods is DRGEP (Directed Relation Graph with Error Propagation), which was intro- Targets: Important chemical species, for which the skeletal mechanism has to produce similar results as the full mechanism. 25 Sample points: A set of environmental conditions (temperature, pressure, concentrations of chemical species) simulated by the chemistry model.
Interaction coefficients (DIC, PIC, OIC): The importance of chemical species in a mechanism is defined in terms of several interaction coefficients. The direct interaction coefficient (DIC) describes the importance of one species for another, based on its normalized contribution to production/consumption rates through reactions involving both species. Then, a Table 1. Simplified example list of species with overall interaction coefficients (OICs). The full mechanism includes all species; the skeletal mechanisms s1, s2, and s3 only include species above a certain OIC threshold. Target species with OIC = 1 are always included. The color coding of the skeletal mechanism is used in Fig. 4.    Error δ skel : A normalized value describing the error when using a skeletal mechanism instead of the full mechanism. A skeletal mechanism is suitable if δ skel < 1 for all targets and sample points. To allow individual weighting, the calculation of δ skel depends on a target threshold AbsTol and a maximum acceptable relative tolerance RelTol, which are defined for all targets: 5 where x full and x skel are the mixing ratios calculated with the full and the skeletal mechanism, respectively.
OIC threshold ε ep : A chemical species is considered important if OIC(species) > ε ep . The final ε ep calculated by DRGEP is the maximum value for which δ skel < 1 still holds.
To test the skeletal mechanism generation, we chose HCHO, HO 2 , NO, O 3 , and OH as target species, allowing a relative tolerance of RelTol = 20 % for mixing ratios above a threshold of AbsTol = 1 pmol/mol. Thirty sample points were extracted subset describing terpene chemistry is shown in Fig. 4. The importance (OIC values) of a few selected species is shown in Tab. 1. Three skeletal mechanisms (s1, s2, s3) were generated, reducing the number of species from 663 in the full mechanism to 462, 429, and 411, respectively. The number of reactions was reduced from 2091 to 1444, 1320, and 1262, respectively.
The third skeletal mechanism (s3) was rejected because it did not fulfill the criterion δ skel < 1. Results obtained with the full mechanism and with s2 were compared in a global simulation, as described below in Sect. 4.2.

Kinetic and isotope tagging
We have updated the sub-submodel MECCA-TAG , which had been introduced in version 3.0 of CAABA.
Several improvements to the kinetic tagging technique were implemented. These new features include: -Selectable composition transfer mode: Depending on the research question, prescribed-, molecular-or element-weighted composition transfer may be selected. These modes determine the shares with which each reactant contributes to the 10 products in the tagged chemical reactions: according to user-specified weightings, proportional to the reacting molecules count, or following the given element (e.g., C or H) content, respectively. Whilst the latter mode is intrinsic to isotope tagging, the others may be used for custom tagging configurations, e.g., product yield calculations.
-Diagnostics for unaccounted production or loss of elemental composition: MECCA-TAG optionally adds passive diagnostic species to the tagged reactions with unbalanced transfer of the element of interest. This helps to quantify the 15 amount of atoms the chemical mechanism receives from or loses to "nothing", including the isotope composition of such mass-balance violations.
-The new "class shifting" tagging mode: This mode allows migration of molecules between the tagging classes in specified reactions, which allows quantifying various exchange processes in the mechanism. For instance, one can distinguish oxidation generations: in reactions with given oxidants the products become "promoted" to the tagging class of the 20 next oxidation generation. Another application of "class shifting" is quantifying the efficiency of recycling chains. In essence, such is the "online" implementation of the approach similar to that of Lehmann (2004), with the number of tagging classes defining the maximum of the recycling sequences it is possible to follow.
The range of MECCA-TAG applications was extended with new tagging setups/configurations: -Radiocarbon configurations, which facilitate simulating the 14 C content in a desired set of species, including the routines 25 for calculating abundances using conventional units like pMC (percent Modern Carbon). There are also some changes in the implementation and software requirements. There is no "doubling" mode anymore for evaluating the results of the optimized tagging. Performing kinetic tagging of the chemical mechanism with MECCA-TAG requires the Free Pascal Compiler (fpc 3 , version ≥ 2.6) at the time the xmecca script is run. The sub-submodel files are located in the mecca/tag/ directory of the distribution. The directory mecca/tag/cfg/ contains tagging configuration control files (*.cfg).
The option to tag a newly created chemical mechanism is available in the xmecca script (also via batch files). Further details 5 about the MECCA-TAG code development can be found in the file mecca/tag/CHANGELOG within the CAABA distribution.

Photolysis
CAABA contains several submodels which provide photolysis rate coefficients J, also called "J-values". The simple submodels READJ and SAPPHO have already been described by Sander et al. (2011a). READJ has not changed since version 3.0.
SAPPHO photolysis rates can now be scaled using a common enhancement factor "efact" for all photolysis rates. This has 10 for instance been used to simulate the very bright conditions within a cloud top (Heue et al., 2014). The updated and new photolysis submodels JVAL and RADJIMT are described in the sections below.

JVAL
The submodel JVAL inside the CAABA/MECCA model calculates J-values using the method of Landgraf and Crutzen (1998). It was first updated to the version described by Sander et al. (2014), and then additional changes were made. Many

RADJIMT
RADJIMT is a new submodel that provides dissociation and ionization rates due to absorption of light and energetic photoelectrons in the mesosphere and thermosphere (see Tab. 2). It is part of the upper atmosphere extension of MESSy initially described by Baumgaertner et al. (2013), which was partly based on the implementations from the middle and upper atmosphere model CMAT2 (Harris, 2001;Dobbin, 2005;Dobbin and Aylward, 2008). For upper atmosphere simulations with CAABA, MECCA Photodissociation and photoionization due to the absorption of solar X-ray, EUV, and UV radiation are calculated using fluxes from the SOLAR2000 empirical model (Tobiska et al., 2000), the GLOW model (Solomon et al., 1988), as well as data  Henke et al. (1993) and Fennelly and Torr (1992). Relative partitioning between the possible products of the ionization process are based on the model of Strickland and Meier (1982) and Fuller-Rowell (1993).
For solar zenith angles larger than 75 • , the atmospheric column of each absorbing species is calculated using an approximation of the Chapman grazing incidence function (Smith and Smith, 1972).
Reaction enthalpies in kJ/mol (exothermic chemical heating) are provided as a product of the relevant chemical reactions 5 when "set enthalpy=y" is defined in the MECCA batch file. Radiative heating and cooling is also calculated by the submodel (variable "heatrates").
As an example, we have performed simulations with CAABA using the MECCA and RADJIMT submodels. The mechanism was created using the batch file mtchem.bat, which selects reactions of the upper atmosphere labeled %Up. The model setup in caaba_mtchem.nml was used: The temperature was kept constant at 195 K, and the pressure was set to 0.5 Pa (approximately

DISSOC
The new MESSy submodel DISSOC is based on the photolysis scheme by Meier et al. (1982). Briefly, it calculates a table of the so-called enhancement factor, which is basically the ratio of the actinic flux at a specific location to the solar irradiance at the top of the atmosphere. The enhancement factor depends on the pressure level, solar zenith angle and wavelength. Input data are the solar irradiance at the top of the atmosphere, absorption cross sections, ozone and oxygen profiles. For the implementation 5 into global models, the input profiles are allowed to be latitude-dependent, which increases the dimensions of the enhancement factor table from three to four. Photolysis rates are calculated from the tabulated enhancement factor as a wavelength integral over the product with the absorption cross sections. The calculation is formulated in spherical geometry, such that it can be also applied to zenith angles above 90 • . Rayleigh scattering is calculated based on Nicolet et al. (1982). Absorption cross sections are taken from the current JPL recommendations (Burkholder et al., 2015). 10 The code was first implemented by Lary and Pyle (1991) and coupled to a stratospheric chemistry-box model (Müller et al., 1994). Becker et al. (2000) improved the treatment of the diffuse actinic flux and corrected an implementation error of Meier et al. (1982). The extension to the use of multiple latitudes was introduced within the development of the model CLaMS (McKenna et al., 2002). The possibility to calculate diurnally averaged photolysis rates was introduced for the simplified fast chemistry setup used in multi-annual CLaMS simulations (Pommrich et al., 2014). 15 In the current configuration, DISSOC determines the photolysis rates for 38 photolysis reactions that are primarily of rele- To facilitate the analysis of the scaling impact, jfac is now written to output. Scaling thresholds have been implemented to prevent artifacts that would occur when J(NO 2 , JVAL) is very small and the calculation of jfac approaches a division by zero.

10
In a standard global model simulation, the MESSy submodel MECCA contains one chemical mechanism that is used for all grid boxes. This ensures a consistent chemistry simulation from the surface to the upper atmosphere. However, in some cases, it may be preferable to allow different mechanisms in different boxes, e.g., terpene chemistry only in the troposphere and ion chemistry only in the mesosphere.
With the script xpolymecca, several independent chemical MECCA mechanisms can be produced. The first mechanism has 15 the name "mecca", as usual. Additional mechanisms are labeled with a three-digit suffix. For example, the code of mechanism 2 is contained in messy_mecca002_kpp.f90 and related files.
To select an appropriate mechanism at each point in space and time, the MESSy submodel CHEMGLUE has been written.
The name of the submodel was chosen because CHEMGLUE can also "glue" together different chemical mechanisms at the border where a chemical species is included in one mechanism but not in the other. CHEMGLUE defines the new channel object 20 "meccanum", which contains the mechanism number for each grid point. These values can either be selected statically, e.g., depending on the model level number or the sea-land fraction mask. Alternatively, a dynamic (time-dependent) selection based on chemical or meteorological variables is possible, e.g., pressure, temperature, or the concentrations of ozone or isoprene.
Note that even when different boxes of a global model simulation use different chemistry mechanisms, the set of tracers contains all species from all mechanisms for all boxes. 25 The implementation ensures binary identical results when one chemical mechanism ("mecca") is replaced by two identical copies of it ("mecca" and "mecca002").
For a more realistic test, we created two different chemical mechanisms for organics. In the first mechanism, only the oxidation of methane is considered, and all non-methane hydrocarbons are neglected. The second (FULL) contains the full set of MOM (Sect. 2.1) reactions. CHEMGLUE selects the second mechanism whenever the mixing ratios of organics are 30 above a threshold (isoprene > 100 pmol/mol, α-pinene > 100 pmol/mol, or toluene > 10 pmol/mol). To investigate how much CPU time can be saved and how much the simplification affects the results, we have performed global test simulations based on the ECHAM5/MESSy atmospheric chemistry (EMAC) model by Jöckel et al. (2016). The horizontal resolution was Geosci. Model Dev. Discuss., https://doi.org/10.5194/gmd-2018-201 Manuscript under review for journal Geosci. Model Dev.  Fig. 6. Overall, the agreement between the simulations is quite good, considering that the simplified mechanisms neglect many reactions. -We extended CAABA with parameters to optionally control the output step frequency (output_step_freq) and the output synchronization frequency (output_sync_freq). The first variable sets the frequency at which values are written to the Geosci. Model Dev. Discuss., https://doi.org/10.5194/gmd-2018-201 Manuscript under review for journal Geosci. Model Dev. lost.
-The treatment of humidity has been improved. Now specific as well as relative humidity (RH) are available throughout CAABA, and can be interconverted with generic conversion functions. Of the two, specific humidity is the more robust variable for humidity because the definition of RH can be based on either partial pressure or on specific humidity (Jacobson, 1999). There are various parameterizations for saturation water vapor pressure, and RH can be defined over 10 liquid surface even below 0 • C, if supercooling is allowed. Functions that use humidity as input (concentration of air, conversion between humidity and water vapor concentration) now use the unambiguous specific humidity. If necessary, it is derived from relative humidity taking all of the above considerations into account.
-For better model time control, two boolean namelist parameters have been introduced: l_groundhogday=T repeats a diurnal cycle while l_freezetime=T repeats a certain point in time, effectively freezing the solar zenith angle.

15
-The selection of various chemical species to define steady state has been simplified to allow for more flexibility in the criteria. The progress towards the defined steady state is now logged during CAABA runtime. Artifacts by species' concentrations close to zero are now prevented.
-The previous tcsh script performing multiple model simulations (Sander et al., 2011a) has been converted to python (multirun.py). It does not depend on the availability of the NetCDF operators (ncks etc.) anymore.

20
-Model results can now be visualized with python scripts using matplotlib. The previously used ferret scripts are still included but not actively supported anymore.
-Complex reaction mechanism can be interpreted as graphs, with species representing vertices and reactions representing edges. To visualize and analyze these graphs, the graph-tool software by Peixoto (2014) can now be used. For example, Fig. 4 was created with graph-tool.

25
-Rate coefficients have been updated to the latest JPL recommendations (Burkholder et al., 2015) and recent laboratory studies. A complete list of chemical reactions, rate coefficients, and references is available in the supplement (meccanism.pdf).
-The kinetic preprocessor KPP (Sandu and Sander, 2006) performs the numerical integration of the chemical reaction mechanism. It has been updated to the latest version 2.2.3, which contains a number of small fixes throughout the code 4 . 30 Geosci. Model Dev. Discuss., https://doi.org /10.5194/gmd-2018-201 Manuscript under review for journal Geosci. Model Dev. Discussion started: 3 September 2018 c Author(s) 2018. CC BY 4.0 License.
-The scripts check_eqntags.py and check_eqns.pl check the internal consistency of the chemical mechanism.
-Details of all new features have been added to the updated User Manual, which now also includes an index. Additional minor bug fixes can be found in the CHANGELOG file.

Summary and outlook
We have presented the current version of the atmospheric chemistry module MECCA-4.0gmdd, which is now available to the 5 research community. Based on the model development described in this paper, our upcoming goals are: -Reduce complex mechanisms to a size suitable for global model simulations.
-Perform a chemical mechanism intercomparison for MOM, CB05BASCOE, MOZART, JAM002, and the MCM.
-Advancing our understanding of the role of organic compounds on the tropospheric O x and HO x budgets.
-Investigate the multiphase chemical pathways leading to organic acids and aerosols.

10
-Simulate stratospheric isotope H exchanges between CH 4 and H 2 O.
-Implementation of additional photolysis modules (e.g., CLOUDJ, TUV) and comparison of the resulting J-values.
-Parallelization to distribute independent (e.g., Monte-Carlo or sensitivity) box model simulations on multiple cores.

Code and data availability
The CAABA/MECCA model code is available as a community model published under the GNU General Public License 5 . The   obtained with the FULL chemistry mechanism. The middle row compares POLY to FULL, and the bottom row compares SKEL to FULL.