ICON-ART 2.1: a flexible tracer framework and its application for composition studies in numerical weather forecasting and climate simulations

2.1 Technical description of the tracer framework

The technical framework of ICON-ART is based on Fortran 2008, which provides essential functionality for the new tracer framework. Classes allow to overcome the stringent matching of data types. Items only have to match a data type or any extension of this type when it is declared with CLASS(type) where type is a derived data type (Chapman, 2008). If the CLASS matches more than one data type, it is called polymorphic. Within ICON-ART, we are using these polymorphic objects to extend the existing tracer structure by additional metadata. Linked lists are another Fortran 2008 feature that is extensively used. Linked lists allow for new features like passing a reference to the next list element or adding and deleting elements.

In the scope of this paper, we are working with two different types of tracers: passive and chemical active ones. An example for a passive tracer is a constituent that does not have any impact on other tracers or the thermodynamics of the simulated system. Passive tracers have predefined sources or only fixed initial conditions. They change by transport and do not interact with other tracers or processes. Chemical active tracers experience sources and sinks while being transported and can participate in feedback processes.

Passive and chemical active tracers are distinguished by their associated metadata. In our example, passive tracers have no chemical loss term to be considered. In other words, the lifetime of passive tracers is infinite. Information about the lifetime is used for simple chemical tracers that experience an idealised loss while being transported. Of course, the information dealing with transport should be provided for both tracer types. A flexible aerosol dynamics module making use of the flexible tracer framework is currently undergoing development.

2.2 Storage of metadata information

We extend the existing ICON-ART functionality by the usage of a key-value storage. The storage is based on string comparisons using generic hash tables. For every tracer, a unique key of fixed size is defined. This represents the search key. The table is constructed like a dictionary. Tracer information can be looked up fast, using this search key. Hash tables provide the foundation for a generic and flexible concept to read and store metadata. At model initialisation, one key-value storage for each tracer is initialised by

CALL storage%init(lcase_sensitivity=.FALSE.)

In this case, the dictionary entries are not case sensitive. The storage unit has two fundamental subroutines; these are storage%put and storage%get. The first creates a new dictionary entry, and the second searches for a key and gives the connected entry. In the next step, this storage unit has to be filled. Accepted types of storage values are reals, integers or characters.

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE tracers SYSTEM "tracers_gp.dtd">

<tracers>
 <chemical id="TRO3">
  <mol_weight type="real"> 4.800E-2 </mol_weight>
  <lifetime type="real"> 2592000 </lifetime>
  <transport type="char"> stdchem </transport>
  <init_mode type="int"> 0 </init_mode>
  <unit type="char"> mol mol-1 </unit>
  <long_name type="char"> ozone </long_name>
 </chemical>
</tracers>

2.3 Construction of the tracer metadata structure

Figure 1Schematic describing the tracer framework and the tracer definition in ICON-ART.

Download

Within ICON-ART, the metadata, e.g. regarding transport or chemical properties, are read in using the XML interface. Afterwards, the information is stored in the ICON-ART tracer framework. The individual steps of the tracer transport simulation in ICON and ICON-ART are described in Rieger et al. (2015). Figure 1 shows the calling structure of the tracer framework in ICON-ART with regard to the tracer and metadata construction. All tracers within the ICON-ART tracer framework share a common structure like the following:

TYPE t_tracer_meta
...
END TYPE t_tracer_meta

In the case of chemical tracers, we want to provide additional information on, e.g. mol weight. The extension of the derived type looks like this:

 TYPE, EXTENDS(t_tracer_meta) :: t_chem_meta
  REAL(wp)    mol_weight
 END TYPE t_chem_meta

In this case, the mol_weight is mandatory information. Optional metadata are stored in as part of the basic structure t_tracer_meta. This container includes a key-value storage. This solution ensures that the metadata container stays as flexible as possible. The respective XML file is read in, and all information is processed and stored in the opt_meta container; see Fig. 1. In the example of mol_weight, this attribute becomes a real variable accessible in the code. It can be accessed by using, e.g.

SELECT TYPE(meta => info_dyn%tracer)
 CLASS IS (t_chem_meta)
  CALL meta%opt_meta%get('mol_weight', &
       & mol_weight, ierror)
END SELECT

Here, CALL meta%opt_meta%get(…) is the operation to access the key-value storage in the container.

The combination of linked lists, polymorphic classes and key-value storage allows the user to define tracers in a flexible way. Only those tracers given in an input file are read in, registered and appear in the modelling system. With this solution, the user is free to define any number of chemical or passive tracers which is only limited by the memory of the computing system.

2.4 Construction of XML tracer input file via MECCA

The XML file can either be edited manually by the user or generated automatically by an external program. The tracer files used for the experiments described in Sect. 5 are shown in Appendix A. In this section, we demonstrate the XML file generation by an external program, which also extends the functionality of ICON-ART. In addition to the existing lifetime-based chemistry approach (Rieger et al., 2015), a full gas-phase chemistry approach is added.

Construction of the full gas-phase chemistry approach is done using the comprehensive and flexible atmospheric chemistry module MECCA (Sander et al., 2011). MECCA provides a set of chemical reactions covering the troposphere as well as the stratosphere. The base version of MECCA is extendable by its own chemical reactions or an update of rate coefficients.

The MECCA preprocessing part has been extended by a routine constructing the XML file for the tracer module of ICON-ART. The numerical flexibility of MECCA is based on the KPP software (Sandu and Sander, 2006). KPP generates Fortran90 code which is used to solve the differential equations based on the given chemical reaction. This is the first step which is needed for KPP and thus for MECCA. In a second step, the numerical integrator is chosen. For our example, we use the Rosenbrock solver of the third order (Sandu et al., 1997). In the third step, we use the driver part of MECCA. The driver stands for the main program which calls the integrator, reads input datasets and writes the results in the original MECCA model. Within ICON-ART, we replaced this step by routines of ICON and ICON-ART. Only the integrator call element is maintained. A shell script, in combination with the AWK scripting language (Aho et al., 1987) processes all information and generates the respective files and routines. Technical work has been done to ensure that a dictionary is accessible for the translation between four-dimensional chemical tracers in ICON-ART and one-dimensional concentrations of chemical species in the KPP routines.

Additionally, other chemical reaction mechanisms provided by MECCA can be used within ICON-ART without changes in the program code itself. It is sufficient to read the new tracer XML file generated by the MECCA preprocessor. Finally, all standard reaction schemes provided by MECCA are accessible in ICON-ART by only using MECCA as an external preprocessor. The model has to be recompiled once, but the user does not have to perform changes in the program code. Files that are changed by the preprocessor are copied to the respective ICON directory automatically. Photolysis rates are calculated using an updated version of CloudJ7.3 (Prather, 2015).

In the scope of this paper, we use a chemical reaction mechanism based on the extended Chapman cycle to demonstrate the functionality of the ICON-ART gas-phase routine.

Table 1Summary of the chemical reactions used for the extended Chapman cycle simulation.

Download Print Version | Download XLSX

We defined the chemical reactions as shown in Table 1.

4.1 Linearised ozone algorithm

Ozone is one of the most important chemical species regarding radiative heating in the middle atmosphere. Thus, ozone concentrations have to be simulated reasonably to address questions about transport processes determined by stratospheric processes (Braesicke and Pyle, 2003, 2004). ICON-ART, in its standard NWP configuration, uses monthly climatological ozone values derived from GEMS climatology (Global and regional Earth-system (Atmosphere) Monitoring using Satellite and in situ data, Hollingsworth et al., 2008). The ozone climatology used in the heating rate calculations is independent from the year of simulation. Here, we move one step forward and use the ansatz for a simplified description based on McLinden et al. (2000). This ansatz can be understood as a first-order Taylor expansion of the stratospheric chemical rates. The ozone concentration tendency is linearised with respect to the local ozone mixing ratio, temperature and overhead ozone column density. The algorithm accounts for a more realistic vertical gradient than the provided climatology. For the troposphere, we assume a constant lifetime of 30 days.

The following differential equation describes the linearised approach:

Here, ξ describes the ozone volume mixing ratio, T the temperature in the respective grid box and $c_{{\mathrm{O}}_{\mathrm{3}}}$ the overhead ozone column. The term (P−L) describes the ozone tendency, with P the production term and L the respective loss term. Climatological values are indicated with the superscript ⁰. The partial derivative with evaluation at the respective climatological value is marked with a subscript ₀. The climatological values for P−L, $c_{{\mathrm{O}}_{\mathrm{3}}}$ and the respective derivatives are given by a look-up table.

With this ansatz, heterogeneous processes are not taken into account. The regular linearised ozone (Linoz) ansatz does not include the last term ($\frac{\mathrm{1}}{{\mathit{\tau }}_{\mathrm{psc}}}\cdot \mathit{\xi }$). To address the catalytic destruction by chlorine and bromine radicals in the presence of polar stratospheric clouds, the linearised ansatz is expanded by an additional loss term. Here, τ_psc represents the lifetime of ozone in the region where polar stratospheric clouds can potentially occur.

For the definition of τ_psc, we follow Sinnhuber et al. (2003):

with ϑ as the solar zenith angle.

4.2 Vortex tracer

In the region of the polar stratospheric vortex, temperatures below 195 K can be observed. This low temperature regime determines the development of polar stratospheric clouds. Chlorine and bromine activation takes place on the surfaces of polar stratospheric cloud particles. This activation leads to ozone destruction. In the past, it has been observed that air parcels within the polar vortex seem to be isolated from midlatitudes (Loewenstein et al., 1989; Russell et al., 1993), because the polar vortex edge can act as a transport barrier.

To investigate processes which are relevant in the region of the stratospheric polar vortex, we introduce a polar vortex tracer in ICON-ART. The processes of interest include transport processes of air parcels from within the vortex, exchange processes at the vortex edge and mixing processes. The vortex tracer is a passive tracer with no destruction over time. This means that the tracer is only affected by transport tendencies and has no interaction with other tracers. To define the edge of the polar vortex, the first approach is to use Ertel's potential vorticity (PV), given in $\mathrm{K}{\mathrm{m}}^{\mathrm{2}}{\mathrm{kg}}^{-\mathrm{1}}{\mathrm{s}}^{-\mathrm{1}}$, which is defined as follows:

with ζ the relative vorticity in s⁻¹, f the Coriolis parameter in s⁻¹, ρ the air density in kg m⁻³, θ the potential temperature in K and z the geometric height in metres. The relative vorticity is calculated by

with $\widehat{\mathit{k}}$ the unit vector in the vertical and v the horizontal wind field in m s⁻¹.

On a timescale of weeks and on an isentropic surface, the PV is a conserved quantity (Nash et al., 1996). The maximum in the PV gradient defines the polar vortex edge. However, as this metric shows high variations in small horizontal regions, distinct maxima cannot always be defined. Therefore, we use a second metric following Nash et al. (1996). At the polar vortex edge, the horizontal (zonal) wind component maximises and provides a surrogate for the PV gradient. In the model, the following steps are performed separately for each hemisphere: first, the PV is calculated at every grid point. Between minima and maxima of PV, a sufficient number of equidistant intervals is defined. Then, the respective geographical area enclosed by PV isolines is calculated. The maximum westerly wind relative to the area under PV isolines is derived. Multiplication of both values, the maximum and the enclosed area, gives a reasonable constraint for the polar vortex edge boundary region. The described definition holds for the location of the Northern Hemisphere polar vortex. For the Southern Hemisphere, the maximum of the easterly winds gives the constraint for the location of the edge.

At the initialisation time step, the passive tracer is initialised to one within the boundaries of the vortex. Afterwards only transport processes are in action. This provides a useful metric to investigate transport processes in the region of the polar vortex boundary.

4.3 Age of air

The characterisation of stratospheric transport from observations is difficult, because the velocity of the meridional overturning (the so-called Brewer–Dobson circulation) cannot be accessed directly. However, the efficiency of the overturning can be estimated by investigating the trends of volume mixing ratios of some long-lived trace gases (Bönisch et al., 2011; Rosenlof, 1995). One of these species is sulfur hexafluoride (SF₆), which is an anthropogenically emitted chemical compound with a long lifetime of up to thousands of years (Ray et al., 2017). The increase of the middle atmosphere (up to 40 km) mixing ratio of SF₆ can be assumed as quasi-linear (Maiss and Levin, 1994). In addition, SF₆ has no significant chemical sink within the stratosphere. In the mesosphere, photolytic dissociation at the Lyman-α band and dissociative electron detachment should be taken into account (Ravishankara et al., 1993). With some assumptions, it is possible to calculate the age of air by using the SF₆ concentration at a specific layer and horizontal location, and comparing (matching) it to past tropical tropospheric values. Thus, the age of air represents the time that an air parcel needed to travel from the tropical tropopause to this particular location in the stratosphere. In the model, a simplified procedure is followed: lower boundary conditions are updated every integration time step to simulate a linear increase of SF₆ with time. The increase corresponds to an increase of the value of 1 year per year. Using the time lag technique (Reddmann et al., 2001; Schmidt and Khedim, 1991), for pressure levels above 950 hPa, one can calculate the age of air at this point by

with Δt_sim is the integration time step of the model given in seconds. After initialisation, the tracer ψ_age in units of seconds is transported. To avoid small fluctuations, the mean age of air is taken into account for further analysis. For the final calculation, the simulated values are merged with the simulated time to get the actual age of air (${{\mathit{\psi }}^{\prime }}_{\mathrm{age}}$) in seconds:

where $t_{\mathrm{sim}}^{*}$ is the simulated time and $t_{\mathrm{init}}^{*}$ the time of initialisation, both given in seconds.

4.4 Water vapour

Water vapour has a strong impact on the radiation budget of the atmosphere in the long-wave infrared spectrum. On the one hand, the infrared emission of water vapour leads to a cooling (loss in the thermal budget) of the atmosphere. Thus, the concentration of water vapour influences transport processes due to thermodynamic induced changes in the wind field. On the other hand, transport processes affect water vapour concentrations. The investigation of water vapour in models and observation allows studying global circulation patterns (Kley et al., 2000). The amount of water vapour entering the stratosphere depends on the temperature within the tropical tropopause layer (TTL). Above, the Brewer–Dobson circulation drives the upward transport in the tropics. Below, convective fluxes and slow ascend in the TTL determine vertical transport. The process of “freeze-drying” leads to dry air parcels entering the stratosphere, because ice crystals sediment out (Brasseur and Solomon, 2006). This process is included in the microphysical schemes of the NWP and the climate physics. Methane oxidation is a chemical source for stratospheric water vapour. Higher up, methane oxidation is a chemical source for stratospheric water vapour. However, the photodissociation of water, mainly located in the mesosphere, is an important sink for water vapour in the atmosphere.

We follow Dethof (2003) by using the following parameterisation:

with ${\mathit{\tau }}_{{\mathrm{CH}}_{\mathrm{4}}}$ the lifetime of methane in seconds, p the pressure and $\mathit{\alpha }=\frac{\mathrm{19}\ln \left(\mathrm{10}\right)}{\ln (\mathrm{20}{)}^{\mathrm{4}}}$. Based on Brasseur and Solomon (2006), the lifetime of water vapour (${\mathit{\tau }}_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}$) due to photodissociation can be calculated by

Taking both terms of production and loss into account, the volume mixing ratio of water vapour at time step t+Δt_sim can be calculated by

with ${\mathit{\psi }}_{{\mathrm{CH}}_{\mathrm{4}}}$ in mol mol⁻¹ as the methane tracer of ICON-ART.

Figure 3Antarctic total ozone column (DU) for a time sequence starting on 21 September 2002 finishing 1 October 2002 (left to right). Daily averages for two ICON-ART simulations for the Linoz chemistry scheme (a–d) and for the extended Chapman cycle chemistry (e–h). Total Ozone Mapping Spectrometer (TOMS) observations are shown in panels (i–l). The model was initialised on 20 September 2002.

Download

Studies have shown that observed mean-annual cooling trends in the tropical tropopause are larger than shown by model simulations, e.g. Shine et al. (2003). It can be seen that the large-scale dynamics in the Earth's atmosphere, tropical tropopause temperatures and lower stratospheric water vapour are closely linked to each other by complex feedback processes. By using the parameterisations described above in ICON-ART, we want to account for these important feedback mechanisms in a simple but reasonable way. To investigate the feedback processes of the ART water vapour tracer (including radiation), only the tendencies from the methane oxidation and photodissociation are taken into account. At every model time step, the water vapour mass mixing ratio is set to the value of the q_v tracer, which is affected by the microphysics routines of ICON. The q_v tracer is the standard water vapour tracer of ICON. Within the ICON-ART routine, the tendency due to methane oxidation and photodissociation can be added. In the last step, q_v is set to the value of ${\mathit{\psi }}_{{\mathrm{H}}_{\mathrm{2}}\mathrm{O}}$. The standard tracer q_v is overwritten before other processes like radiation or microphysics are active.

5.1 Ozone and vortex tracer

In 2002, an unusual split of the ozone hole was observed on 22 September 2002 and described by Newman and Nash (2005). The initiation of the splitting process is not fully understood yet. The split of the ozone hole had no chemical reason; instead, it is a dynamical change controlling atmospheric composition and in particular ozone distributions. Several studies, e.g. Matsuno (1971), have shown that a vortex split event can be caused by atmospheric interactions with upward propagating planetary waves. Sinnhuber et al. (2003) pointed out that the major stratospheric warming occurred far earlier than the normal final warming at the end of the ozone hole season. With ICON-ART, we are able to study the vortex split event of 2002 in a hindcast. First, we want to discuss the total ozone column simulated with ICON-ART. For the setup of the experiment, we use reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) – ERA-Interim (Dee et al., 2011) to initialise the meteorological variables (e.g. pressure, temperature, water vapour and horizontal wind fields). The hindcast is initialised on 20 September 2002, 00:00 UTC. The chosen grid is R2B6, corresponding to an approximate horizontal grid spacing of 40 km (Zängl et al., 2015). The model top is at 75 km with 90 vertical levels. The integration time step is 240 s, and the output time step is every hour. The simulated ozone column, interpolated on a regular latitude–longitude grid with a resolution of 0.5^∘, is shown in Fig. 3. The four columns show daily means of total ozone for the respective dates. The mean is taken over all 24 output steps starting at 00:00 UTC. For the initial values of ozone, the ERA-Interim data are used analogous to the meteorological data. The ozone hindcasts are performed with two different schemes: the modified Linoz scheme as described in Sect. 4.1 and a gas-phase algorithm (the extended Chapman cycle) with reactions described in Table 2.4. For the comparison between simulations and measurements, we are using satellite observations from the Total Ozone Mapping Spectrometer (TOMS) instrument (TOMS Science Team, 2016).

Figure 4Passive and active ozone tracers (mol mol⁻¹) at 50 hPa over Antarctica. Daily averages for two ICON-ART simulations for the Linoz chemistry scheme (a) and for the extended Chapman cycle chemistry (b) and the passive tracer (c) are shown.

Download

Figure 5Difference of the passive and active ozone tracers (mol mol⁻¹) at 50 hPa over Antarctica. Daily averages are shown for two ICON-ART simulations for the Linoz chemistry scheme (a) and for the extended Chapman cycle chemistry (b). For the loss rate of the Chapman cycle, the zero contour line is added. Shaded areas indicate temperature regions below 195 K.

Download

A total of 5 days after initialisation, the represented horizontal geometry of the total ozone column in both ICON-ART simulations was slightly different in comparison to the satellite observations. After more than 10 days of simulation, the shape of the vortex was still in good agreement with satellite observations. Within the polar vortex, the total ozone column reached values of about 200 DU in both, ICON-ART simulations and the satellite observations. At the end of the simulation, on 1 October 2002, the largest difference between the extended Chapman cycle simulation, using the full gas-phase algorithm, and the simplified Linoz simulation was found at 70^∘ S, 120^∘ W, outside of the polar vortex. The total ozone column reached values of 450 DU and above for the extended Chapman cycle simulation. For the Linoz simulation, ozone destruction was slightly higher and values up to 400 DU were reached. The comparison to the TOMS observation shows that this minimum fits to observation. However, the horizontal pattern slightly differs. The observation also shows small areas with up to 420 DU.

In order to illustrate the differences between the two idealised simulations (parameterised using Linoz; gas-phase chemistry only with an extended Chapman cycle), the total contribution of chemical tendencies is depicted in Figs. 4 and 5. We are using the differences between a passive (no chemical changes after initialisation) ozone tracer and the chemical ozone tracers for Linoz and extended Chapman cycle chemistry. The difference represents the chemical ozone loss.

The passive and the chemical ozone tracers are initialised identically. Technically, this is ensured by declaring the ozone tracers to be the same type as the chemical tracer but without chemical interactions. The corresponding XML entry can be found in Appendix A1. In Fig. 5, at 70^∘ S, 120^∘ W, on 1 October 2010, ozone loss is dominated by chemical loss. Ozone losses are higher in that region. The passive tracer, which is only affected by dynamical tendencies, has higher values than the chemical tracers; thus, ozone has been depleted by the chemical mechanisms. The maximum difference between passive and chemical tracers is up to 10 times higher for the Linoz simulation than for the extended Chapman cycle. For visualisation purposes, the contour line representing no loss (0 mol mol⁻¹) is added. This is to be expected, because the initial ozone distribution is not equilibrated with respect to the Linoz reference state (causing large tendencies) and we do not consider additional loss terms by heterogeneous processes or halogens in the extended Chapman cycle chemistry (limiting the range of chemical tendencies). However, we can focus on the general spatial structures and how transport has modified ozone distributions. Inside the polar vortex, on 1 October 2002, we model ozone production for both simulations. The chemical tracer in both simulations is increased with respect to the passive one. The increase is higher in the Linoz simulation than in the extended Chapman cycle. This implies that temperatures in that region are not low enough to trigger the heterogeneous destruction of ozone in the Linoz scheme. Outside the polar vortex, mainly on 25 September, high values of ozone loss can be observed for the Linoz simulation but not for the extended Chapman cycle. This is also caused by the difference in addressing heterogeneous destruction. Within the Linoz scheme, the loss term has been triggered and we can observe additional ozone loss. This feature is missing for the extended Chapman cycle chemistry.

In Fig. 6, the temporal evolution of the passive vortex tracer is depicted. The colour coding gives the values of the vortex tracer at an interpolated pressure level of 30 hPa at midnight on the given date. Again, the date of initialisation is chosen to be 20 September 2002 and within the boundaries of the vortex; the tracer is set to a value of 1. Here, only transport takes place. The horizontal spreading of the vortex tracer depicts the dynamical evolution of the vortex in the Southern Hemisphere. The horizontal spreading and steep tracer gradients correspond to the horizontal distribution of the total ozone column in Fig. 3. At the day of initialisation, the vortex is still intact, but in the following days the first observed major stratospheric warming in the Southern Hemisphere, e.g. (Newman and Nash, 2005), occurs. The massive outflow of vortex air masses, beginning on 24 September 2002, can be visualised by the vortex tracer distributions. This outflow is correlated to the increased dynamical impact on the vortex integrity. The vortex split occurred on 26 September 2002. The structure, represented by the spatial distribution of the vortex tracer, is nearly separated. On this day, the northernmost latitude of 30^∘ S is reached by a vortex filament. The vortex tracer allows to define regions of isolated air masses within the vortex, thus providing an insight into the chemical composition changes that are least affected by diffusion and mixing (e.g. McKenna et al., 2002 or Konopka et al., 2005).

Figure 6Daily snapshots at 00:00 UTC at 30 hPa of the passive vortex tracer in arbitrary units (upper left to bottom right). The passive tracer was set to 1 within the vortex boundary at the beginning of the integration.

Download

5.2 Feedback of ozone on radiation

In the previous section, we have shown that the Linoz configuration of ICON-ART provides good hindcasts on days to weeks. The limitation that the initial state should be equilibrated, mentioned above, does not matter after a spin-up period. Here, we will extend the time horizon considered to decadal integrations. In addition, we will illustrate how the optional composition of tracers with radiation feedback affects the system. With the flexible tracer structure, the user has the ability to switch on the radiative feedback by the tag

<feedback> 1 </feedback>.

Thus, no changes in the code have to be made by the user. Only the XML file has to be changed.

The ICON-ART simulation is configured as an AMIP (Atmospheric Model Intercomparison Project) like experiment (Gates et al., 1999). The boundary conditions used are summarised in Table 2. We set up two experiments on the R2B4 grid which corresponds to an effective horizontal grid spacing of 160 km. The integration time step is chosen to be 600 s. Output is written every 3 days and interpolated onto predefined pressure levels, corresponding to the standard ERA-Interim pressure levels. The first experiment uses an ozone climatology with monthly mean values based on Cionni et al. (2011). This simulation is called “control”. The feedback simulation uses the interactive ozone as well as the additional tendencies on water vapour, shown in Sect. 4.4. For all following diagnostics and discussions, the same two simulations are used. The results of the complete integration from 1979 to 2009 are found in Appendix B. In the following, analysis of the time span from 1990 to 2009 is shown. This avoids possible differences arising from changes in ozone depleting substances from 1979 to 1990 that are not considered in Linoz with time-invariant look-up tables.

Table 2Overview of the boundary conditions used in the AMIP-like experiments.

Download Print Version | Download XLSX

Figure 7Monthly averaged zonal means of ozone (kg kg⁻¹) at 50 hPa (shown twice) for the period from 1990 to 2009 (shaded). Contour lines represent the standard deviation of the monthly means. (a) Ozone climatology as used in the control simulation; (b) non-interactive Linoz ozone; (c) interactive Linoz ozone.

Download

Figure 7 shows the climatological ozone distribution at a pressure level of 50 hPa. Monthly averaged zonal means of ozone for the period of 1990 to 2009 are plotted twice. Figure 7a shows the ozone used in the control run (ctrl) (Table 2). Figure 7b shows the modelled ozone of the non-interactive Linoz simulation, where modelled temperatures are the same as in the control run. Figure 7c shows the temporal evolution of the zonal mean ozone in the feedback simulation. Additionally, contour lines are representing the interannual standard deviations of ozone in all panels. The black contour line represents a standard deviation of $\mathrm{1}\times {\mathrm{10}}^{-\mathrm{7}}$ kg kg⁻¹. Brighter colours present higher values of standard deviation with a spacing of $\mathrm{2}\times {\mathrm{10}}^{-\mathrm{7}}$ kg kg⁻¹.

A striking difference in the three ozone distributions shown in Fig. 7 is the duration of the ozone hole period indicated by the area that is shaded in dark blue. Modelled ozone (Fig. 7b, c) shows a much longer duration than the assumed background climatology. In addition, the duration of the ozone hole period increases slightly for the interactive integration. This is caused by the feedback of the modelled ozone, calculated with the Linoz parameterisation under consideration of the correction term for a shorter ozone lifetime due to the presence of polar stratospheric clouds. From October to December, very low ozone concentrations of about $\mathrm{1}\times {\mathrm{10}}^{-\mathrm{6}}$ kg kg⁻¹ can be seen in the panel for the non-interactive Linoz ozone (Fig. 7b). For the feedback simulation, low ozone concentrations in the Southern Hemisphere, in conjunction with low temperatures, occur until the end of spring. The feedback process stabilises the Southern Hemisphere polar vortex, prolonging its lifetime by delaying the final warming. The default climatology of the ICON-ART control simulation does not represent very low values of ozone concentrations. This misrepresentation has also been discussed by, e.g. Arblaster et al. (2014). Here, the authors point out that most models that are using a prescribed ozone climatology tend to underestimate the Antarctic ozone depletion. Thus, modelled ozone values are higher than indicated by observations between 1979 and 2007 (Hassler et al., 2013). Taking the standard deviation into account, the characteristics of the AMIP ozone climatology (Cionni et al., 2011) become clearer. The contour lines represent a semicircle each winter in the Southern Hemisphere that is aligned with the ozone concentration gradients. This symmetric and coherent pattern of the ozone minimum is most likely not a very realistic representation of variability on top of an ozone hole that is not deep enough. The standard deviation isolines for the modelled ozone are different and intersect the isolines of ozone concentrations, with higher variabilities at later times in the ozone hole period.

We note that the standard deviations shown in Fig. 7 should be interpreted differently between the control and feedback simulations of ICON-ART. This is due to the fact that in the control simulation a time-varying background climatology is used that has an imposed long-term trend, whereas in the feedback simulation, no trend is imposed, and the internal variability of ICON-ART is the main component of the ozone variability. To illustrate this difference, we summarise the results of a time series decomposition in Table 3.

We use the time series of the 60^∘ S zonal mean total ozone column of both ICON-ART simulations. For comparison, we analyse TOMS observational data (TOMS Science Team, 2016), also at 60^∘ S. The chosen latitude allows a continuous time series analysis. We separate the time series of the ICON-ART simulations in two periods: 1980 to 1997 and 1997 to 2009. The ICON-ART control simulation shows an ozone decline for the earlier period and a small ozone increase for the later period (by construction of the background climatology). The feedback simulation shows no pronounced trend (as expected; see above) in particular during the later period. Note that the RMSE (root mean square error) is higher in both periods for the feedback simulation compared to the ICON-ART control simulation. The higher RMSE is induced by substantial year-to-year meteorological variability. Both model time series can be compared to the shorter TOMS on NIMBUS 7 time series (1997 to 2005).

Table 3Results of the time series analysis for different time domains at 60^∘ S for the ozone total column. Results for both ICON-ART simulations and TOMS observation are shown. The early period is from 1980 to 1997 and the late period is from 1997 to 2009 (to 2005 for TOMS on NIMBUS 7). The the annual cycle (AC) minimum and maximum anomalies are derived from the full time series.

Download Print Version | Download XLSX

5.3 Temperature changes due to ozone feedback

In the previous section, we have shown that with the change to an interactive representation of ozone the duration of the southern polar vortex is increased. Here, we provide more details on the zonal mean temperature distributions in the control and feedback simulations.

The direct impact of ozone, already mentioned in Sect. 5.2, is also displayed in the seasonal variation of the zonal mean temperatures; see Fig. 8.

Figure 8Latitude–height cross sections of seasonal and zonal mean temperature (K) for ICON-ART simulations from 1990 to 2009. (a) Control run; (b) feedback simulation.

Download

Figure 9Latitude–height cross sections of seasonal and zonal mean temperature differences (K) for control minus feedback simulation, as shown in Fig. 8.

Download

Figure 10Latitude–height cross sections of seasonal and zonal mean temperature differences (K) for ERA-Interim minus feedback simulation.

Download

In the Northern Hemisphere winter (DJF), temperatures of 210 K are reached between 100 and 10 hPa in the tropics. For the season of June–July–August (JJA), the temperature minimum in the tropics is at 100 hPa with temperatures as low as 200 K in both ICON-ART simulations. In the Southern Hemisphere winter, ICON-ART control reaches temperatures of about 200 K and ICON-ART feedback temperatures lower than 195 K. The Northern Hemisphere summer is represented by temperatures higher than 260 K above 5 hPa.

Figure 9 shows the difference between control and feedback runs. In the Southern Hemisphere winter, the effect described in Sect. 5.2 can be observed. Due to the lower polar vortex temperature, differences up to 5 K occur. The control run shows warmer temperatures in the southern polar region (below 20 hPa). Above this altitude, the feedback run is warmer than the control run. This is due to the different ozone distributions in the Southern Hemisphere winter. Within the tropical stratosphere, temperature differences of about 2.8 K can be seen. The differences of zonal and seasonal mean temperatures between ERA-Interim and the ICON-ART feedback simulation are shown in Fig. 10. The feature of a long-lasting Southern Hemisphere polar vortex is present as well, seen in high temperature differences of up to 20 K from September to November (SON). In general, the ICON-ART control simulation shows warmer temperatures than the ICON-ART feedback simulation, except for high altitude ranges above 5 hPa.

The general structure is comparable to the results shown in the comparison studies of ECHAM5 (Roeckner et al., 2006). The difference between ERA-Interim and ICON-ART is increased in the Southern Hemisphere stratosphere. Nevertheless, the representation of the polar vortex seems to be more realistic in the ICON-ART feedback simulation than in the control simulation. In the vertical region around 50 hPa, the difference between ERA-Interim and the feedback simulation is about 15 to 20 K in the Southern Hemisphere and below 8 K in the tropics.

5.4 Zonal wind fields

Changes in temperature due to radiative feedback effects of ozone are also affecting the zonal wind structure. The zonal mean zonal wind is shown in Fig. 11. The left column shows the seasonal mean of the control ICON-ART simulation and the right column the feedback simulation. In both simulations, a strong eastward zonal wind with wind speeds up to 60 m s⁻¹ is reached in the Southern Hemisphere winter (JJA). The wind speed patterns in the tropical stratosphere also match the seasonal mean analysis.

Figure 12 shows the differences between the ERA-Interim and the results of the ICON-ART feedback simulation. Strong differences in the stratospheric zonal wind can be seen in the Northern Hemisphere winter. Here, ERA-Interim shows values up to 25 m s⁻¹ lower than in the ICON-ART simulation. Between 30 and 60^∘ N latitude above 20 hPa, the sign of the differences changes. Here, we observe stronger zonal wind speeds than in ERA-Interim. The overall patterns are similar to the differences shown in Roeckner et al. (2006).

Figure 11Latitude–height cross sections of seasonal and zonal mean zonal wind differences (m s⁻¹) for ICON-ART simulations from 1990 to 2009.

Download

Figure 12Latitude–height cross sections of seasonal and zonal mean zonal wind differences (m s⁻¹) for ERA-Interim minus feedback simulation.

Download

5.5 Water vapour

Here, we focus on the atmospheric water vapour tape recorder in ICON-ART. An atmospheric tape recorder can be defined as the vertical propagation of an anomaly that varies periodically in time with a tropospheric source (Gregory and West, 2002). The temporal and vertical distribution of the tropical stratospheric water vapour is a prominent example for an atmospheric tape recorder signal. The simulated stratospheric water vapour depends strongly on the temperatures that are encountered by an air parcel containing water vapour that is transported vertically from the troposphere upwards towards and through the tropopause (Schoeberl et al., 2012). The quantitative link between variations of tropical tropopause temperatures over decades and their influence on water vapour transfer into the stratosphere is still not fully understood (Rosenlof and Reid, 2008). It has been shown that stratospheric water vapour can have a strong impact on stratospheric climate (de F. Forster and Shine, 1999). Thus, the study of the water vapour tape recorder is an important tool for the further understanding of large-scale transport processes and climate change linkages.

The tape recorder anomalies are calculated relative to the annual mean water vapour in the tropics (5^∘ N–5^∘ S). For this study, we use the water vapour tracer, q_v. This tracer is the standard tracer of ICON itself. It is used in the radiation scheme and is not only transported but is also affected by the microphysical schemes. As in the previous section, ozone is calculated with the Linoz scheme and used interactively. The additional water tendencies by methane oxidation and photolysis is also included in the stratospheric water budget.

Figure 13Tropical (5^∘ N–5^∘ S) water vapour anomalies as monthly mean deviations (plotted twice) from the annual mean, averaged from 1990 to 2009, as a function of months and altitude.

Download

Figure 13 shows the stratospheric tape recorder. For the analysis, the years 1990 to 2009 are taken into account. We compare the ICON-ART results with the water vapour product from ERA-Interim. The calculated mean ERA-Interim tape recorder is shown in the bottom panel of Fig. 13.

The tape recorder signal for the ICON-ART control simulation shows lower values of anomalies down to $-\mathrm{7}\times {\mathrm{10}}^{-\mathrm{7}}$ kg kg⁻¹ in comparison to the feedback simulation (Fig. 13). For ERA-Interim, anomalies up to $-\mathrm{2}\times {\mathrm{10}}^{-\mathrm{7}}$ kg kg⁻¹ are diagnosed from February to June in the pressure range of 100 to 50 hPa.

The anomalies in the feedback simulation show higher absolute values. The dry anomalies in the control simulation, between April and June, are decreased in the feedback simulation. In addition, the positive (wet) anomalies from June to December, between 60 to 20 hPa, are decreased in the feedback simulation. The water vapour tape recorder shows less pronounced dry anomalies in the tropical stratosphere in the feedback simulation due to the additional source of water vapour by methane oxidation. Additionally, the analysis of the water vapour tape recorder is consistent with the results of the temperature differences, as seen in Fig. 9. Due to lower tropical tropopause temperatures in the feedback simulation, less water can enter the lower stratosphere. The annual mean temperature difference (not shown) shows that the control simulation is up to 2.8 K warmer than the feedback simulation between 100 and 50 hPa, in the tropics. The results correspond with the freeze-drying hypothesis explained above. In general, monthly mean anomalies are attenuated in the feedback simulation compared to the simulation using the standard ozone climatology, with in particular the winter months being more similar to ERA-Interim. Focusing on the strong positive anomaly between 100 and 50 hPa, from May to August, the explanation from above has to be extended. The annual mean shows a cold bias of the tropical tropopause. However, referring to Fig. 9, one can see that, for that time span, the sign of temperature changes. In the feedback simulation, higher temperatures in the tropical tropopause are reached. Thus, more water vapour can enter the stratosphere which is a possible explanation for the strong positive anomaly.

The tape recorder signal of ERA-Interim shows a maximum amplitude in the lower stratosphere between 100 and 60 hPa. In the ICON-ART control simulation, the lower stratospheric amplitude of the water vapour tape recorder is already smaller. With interactive ozone and water vapour, as in the ICON-ART feedback simulation, the lower stratospheric amplitude is attenuated further. However, a relative maximum above 20 hPa occurs. This is likely the result of an overestimated water vapour tendency provided by the methane oxidation, because methane is biased slightly high in the feedback simulation.

The slope of the latitude–height water vapour anomalies is nearly unchanged between non-interactive and interactive integrations. Thus, the velocity of the upward transport is largely unaffected by the inclusion of the radiative feedback. The result of both ICON-ART simulations in comparison to ERA-Interim is similar to the results presented in Jiang et al. (2015). Here, the authors combined measurements and simulations of water vapour from the Microwave Limb Sounder (MLS), GMAO Modern-Era Retrospective Analysis for research and Applications in its newest version (MERRA-2) and ERA-Interim and used them for a comparison of the water vapour tape recorder behaviour. The upward transport above the tropical tropopause of ERA-Interim is found to be faster than the transport diagnosed from MLS measurements. Thus, in the current configuration, ICON-ART produces similar ascent rates to ERA-Interim, which are likely faster than observed.

We have shown that the change between interactive and non-interactive integrations with respect to tropical ascent rates is small. However, some changes are clearly detectable and the important relationship between the tropical tropopause minimum temperatures and water vapour concentrations is qualitatively captured by the ICON-ART system. More comprehensive climate studies are in preparation.

5.6 Age of air

For the simulation of the age of air, we use the same setup as described in Sect. 5.3. As all other diagnostics, the age of air tracer is interpolated on a regular latitude–longitude grid with a horizontal resolution of $\mathrm{0.75}{}^{\circ }\times \mathrm{0.75}{}^{\circ }$ on predefined pressure levels.

Figure 14Latitude–height cross sections of seasonal and zonal mean ages of air (year) for ICON-ART simulations from 1990 to 2009. (a) Control run; (b) feedback simulation.

Download

Figure 15Monthly averaged zonal means of age of air (year) at 50 hPa (shown twice) for the period from 1990 to 2009 (shaded). Contour lines represent the standard deviation of the monthly means. (a) Control run; (b) feedback simulation.

Download

The tracer is initialised as described in Sect. 4.3. The control (ctrl) experiment is a simulation in which water vapour and ozone calculated within ICON-ART have no impact on the calculation of radiation. In the second experiment (feedback), the altered ozone and water vapour distributions of ICON-ART are coupled to the radiation routine. The first 11 years of each simulation are excluded in the analysis to prevent spin-up effects contaminating the result.

The ICON-ART modelled age of air is depicted in Fig. 14. The diagnostic of age of air can be seen in this case study as an important tool to analyse the feedback processes of greenhouse gases on transport processes in the Earth's atmosphere. It can be seen that the mean age of air is younger; thus, the upward transport has been faster in the control simulation. Due to upwelling transport processes in the tropics, the youngest air masses can be found there. The polar regions show the occurrence of older air masses up to an age of 4 years in both simulations. The asymmetry between Southern Hemisphere and Northern Hemisphere, induced by faster circulations in the Southern Hemisphere (Mahieu et al., 2014), is also well captured. One can clearly see the impact of ozone and water vapour on radiation and thus on transport processes. The age of air in the feedback simulation is up to 6 months older than the control simulation. Figure 15 shows the climatological mean age of air, in the same representation as shown in Fig. 7. Here, the temporal and zonal means of the age of air from 1990 to 2009 are taken at an altitude of 50 hPa. The standard deviation from the mean is represented by the contour lines. These lines represent the interannual variability. The absolute mean age of air is higher for the feedback simulation on both hemispheres. The band of low values in the tropics is narrowed for the feedback simulation. The values of standard deviation are comparable. However, in the region of the Southern Hemisphere polar vortex, from October to January, the standard deviation is higher for the feedback than for the control simulation. Since the polar vortex is stabilised by the ozone feedback, a different dynamical situation can be observed, influencing the interannual variability of the age of air. In comparison to other studies, which focus on other time spans (Brasseur and Solomon, 2006; Engel et al., 2009; Haenel et al., 2015; Stiller et al., 2012), ICON-ART shows an age of air which is too young compared to observations. But this behaviour has also been observed in other studies with different models, as described in, e.g. Monge-Sanz et al. (2007) or Hoppe et al. (2014). With this diagnostic, the general representation of stratospheric transport processes can be investigated further.