The Online Emission Module (OEM) was first developed for COSMO-ART and then adapted to ICON-ART and refactored for improved computational performance. OEM enables efficient processing of emissions which can be represented by adding up individual source categories, with the emissions from each category being fixed in space but varying over time. This holds for most applications requiring input from anthropogenic emission inventories such as the Emission Database for Global Atmospheric Research (EDGAR) or the regional inventory CAMS-REG . These inventories provide gridded emissions divided into individual source categories. An example for such a source classification is the Gridded Nomenclature For Reporting (GNFR), which distinguishes 12 source categories from public power to agricultural emissions . The main steps for integrating these inventories into the model are (i) re-mapping to the ICON grid, (ii) application of temporal scaling factors, (iii) vertical distribution of the emissions onto ICON's terrain-following hybrid sigma-z coordinates, and (iv) summing up over all source categories. In most model systems including WRF-Chem, CHIMERE or CAMx, e.g.,, these steps are performed externally using a pre-processing software such as HERMES , which generates a large number of files each containing the emission field representative of a given time interval (e.g. hourly), which are then read into the model during runtime. OEM, instead, allows reading the emissions together with all temporal and vertical scaling information only once during model initialization. The temporal and vertical scalings are then applied online during the simulation. This greatly reduces the need for data pre-processing and simplifies the setup of new simulations. As shown in , the additional time required for the online computations is fully compensated by the time saved through less frequent file access.

Inputs for OEM can be produced with the Python package emiproc , which is able to process multiple emission inventories to prepare the inputs for a range of atmospheric transport models including ICON-ART. The tool maps the emissions in a mass-conserving way onto the ICON grid and creates all additional inputs including source and country-specific temporal and vertical profiles and the corresponding country masks. The tool is also able to merge multiple inventories, for example to embed a high-resolution national inventory into a coarser global inventory.

To compute the emission of a species X at time t in a given grid cell, OEM performs the following operation 1EX(z,t)=∑s=0NsEX,s⋅wX,s(t)⋅vX,s(z), with EX,s the emission of species X from source category s in that grid cell from the inventory, wX,s(t) the temporal scaling factor at time t for source category s, and vX,s(z) the vertical scaling factor for source category s at vertical level z. Ns is the total number of categories. The temporal factor wX,s(t) is computed as the product of three different scaling factors describing diurnal, day-of-week, and seasonal variability 2wX,s(t)=wX,s,h(h(t))⋅wX,s,d(d(t))⋅wX,s,m(m(t)) with h(t) being the hour of the day, d(t) the day of the week, and m(t) the month of the year. Alternatively, a separate scaling factor can be defined for each hour of the year to represent, for example, heating emissions varying with outdoor temperatures. Furthermore, different temporal scaling factors can be provided for different countries (or regions) together with a country mask. Diurnal factors wX,s,h(h(t)) are computed with respect to local time.

In addition to temporal and vertical factors, it may be necessary to provide speciation factors, which describe the fractional contribution of individual model species X to the total emissions of a family of species X̃ in the inventory. Examples are non-methane volatile organic compounds (NMVOC) and nitrogen oxides (NO_x), for which inventories typically provide only the total emissions of the family but not of the individual compounds. Speciation factors cannot be supplied as input to OEM, but are dealt with by emiproc. Based on an emission field for the family X̃ and a set of speciation factors, emiproc generates and writes out emission fields for all model species X. XML tags and namelist settings for OEM are described in Appendix .

The ICON-ART model system incorporates a one-dimensional, sub-grid-scale plume-rise model developed by , similar to its implementation in COSMO-ART by . This plume-rise model is suited for applications with horizontal resolutions on the order of 10 to 100 km. For these applications, the model calculates plume height based on buoyancy, atmospheric stratification, and flow conditions, accounting for processes that occur on scales much smaller than the horizontal spacing of the ICON. The model uses an internal vertical grid spacing of 100 m with 200 vertical layers. Environmental conditions (pressure, humidity, temperature, wind speed) are provided by ICON and transferred to the plume-rise model as initial and environmental conditions for each active fire grid point to determine plume height.

Fire size and intensity, based on the ICON land use class, vegetation type and density, determine heat release and initial buoyancy. The lower boundary condition assumes a virtual buoyancy source below the surface, resulting in high vertical velocity. Final buoyancy is limited by turbulent and dynamic entrainment, with additional buoyancy from latent heat release during condensation. The plume top is defined where vertical velocity inside the plume drops below 1 m s⁻¹ after an equilibrium state between the surroundings and the heat source is achieved. Heat flux values, dependent on vegetation type, are taken from . They were corrected in comparison to the implementation in COSMO-ART such that the plume-rise model exactly reproduces the results in .

In addition to the original numerical solver introduced by , a more efficient and stable first-order implicit solver for the same equations was developed and implemented. The solver relies on a Godunov type scheme to solve the inviscid Burgers equation for the vertical velocity inside the plume and on upwind schemes for the other modeled plume variables such as temperature, specific humidity, specific cloud water, specific rain water, specific ice, horizontal entrainment velocity, and plume radius. An internal grid is not needed for this solver since it copes with unevenly spaced grids like the vertical atmosphere discretization in numerical weather forecast models such as ICON while at the same time allowing much longer plume internal time-steps. The solver comes with an a priori condition that determines whether a height calculation is needed or if the plume stays at the minimal emission height. The plume heights strongly vary with the time of day since a diurnal cycle function d is applied to fire intensity and size. The function proposed by and applied by in COSMO-ART is also used in ICON-ART. The plume-rise model returns plume bottom and top heights, with a parabolic emission profile f describing the vertical distribution of emissions. The emission rate E of a species in kgm-2s-1 is calculated for each grid cell based on height z and time t. 3EX(z,t)=MX(t)⋅d(t)⋅f(z,t) MX is the daily mean emission flux of species X from CAMS GFAS , d is the diurnal cycle, and f is the parabolic emission profile between upper and lower injection heights.

CAMS GFAS relies on Fire Radiative Power (FRP) from the NASA MOD14 product, which includes thermal radiation observations from the MODIS instrument . To address data gaps due to, e.g. cloud cover, fire data is assimilated using a Kalman filter and statistics. FRP density is updated based on previous and current observations, with sampling limited to four times per day to represent the diurnal fire cycle . GFAS data is provided at a resolution of 0.1°.

The rise of volcanic plumes during eruptions depends on both volcanic and atmospheric conditions. Volcanic conditions are the exit temperature, exit velocity, exit volatile fraction, and the vent diameter. The exit velocity and the vent diameter control the mass eruption rate (MER). The height of the plume depends on volcanic conditions due to effects on the MER as well as on the plume density and atmospheric conditions. Due to the complexity of plume dynamics, simple relationships e.g.,, which only depend on plume height and are often used in dispersion models, can lead to large uncertainties in emissions e.g.,.

Volcanic emissions in ICON-ART are calculated online using the 1-D volcanic plume model FPlume by , which considers the volcanic conditions as well as processes during the plume rise such as ambient air entrainment, plume bending due to wind, particle wet aggregation, energy supply due to water phase changes, and particle fallout and re-entrainment. described the coupling of FPlume with ICON-ART in detail. In short, FPlume requires atmospheric profiles for temperature, pressure, density, zonal and meridional wind speed, and specific humidity at the volcanic vent. In addition to meteorological data, FPlume needs estimates of the exit temperature, exit velocity, and exit volatile fraction, which depend on the type and setting of the volcano. FPlume can either calculate the MER based on a given height or the height based on a given MER. During every time step when the volcano is active, FPlume first calculates the plume properties, i.e., the total MER in the case of a given plume height or plume height in the case of a given MER. Second, the fraction of very fine ash, which is relevant for long-range transport, is determined based on the plume height and the total MER by using the relationship of . Third, very fine ash is emitted along a profile that has initially been defined by and applied by for the coupling of FPlume to NMMB-MONARCH-ASH transport model (Nonhydrostatic Multiscale Model on the B-grid – Multiscale Online Nonhydrostatic AtmospheRe CHemistry model – ASH). The distribution of the very fine ash mass into the ICON-ART modes is prescribed in the FPlume input file.

SO₂ can be emitted alongside with ash using the same timing and profile, however, the MER of SO₂ is prescribed in the FPlume input file, as FPlume does not differentiate between gaseous compounds and water vapor. extended the coupling of ICON-ART and FPlume by water vapor emission. Here, the MER of water vapor is calculated from the exit water mass fraction and the total MER and is emitted through the same profile as ash. This simplification neglects the entrainment of water vapor and was so far only tested for the water-rich 2022 Hunga eruption .

The advantages of the coupling are (1) a more accurate MER and therefore better agreement with observed mass column loadings as shown for the 2019 Raikoke eruption, Kuril Islands and (2) an easy consideration of eruption phases, which allows a comparison to observations in the near-field of complex volcanic emissions such as the 2021 La Soufrière eruption, St. Vincent .

For the emission of sea salt, ICON-ART offers two options. The first option is based on the parameterizations of , , and (hereafter referred to as MMS scheme) as described by . Another option is based on (hereafter referred to as G14 scheme) which is described below.

Numerous studies have shown that sea surface temperature (SST) significantly influences the emission rates of sea salt by affecting the physical properties of foam and droplet formation at the ocean surface. To develop a globally applicable source function that can also represent realistic sea salt concentrations in tropical regions, incorporating this temperature dependence is essential . To address this, we implemented an alternative scheme for sea salt emission in ICON-ART based on , which in addition to wind speed includes SST as a key parameter: 4dF(Dp,U10,T)dDp=Tw⋅(235⋅U103.5exp⁡-0.55ln⁡Dp0.12+0.2⋅U103.5exp⁡-1.5ln⁡Dp32+6.8×10-3⋅U103exp⁡-1ln⁡Dp302).

In this equation, F(Dp,U10,T) is the sea salt emission flux in particlesm-2s-1, Dp is the dry particle diameter in µm, U10 is the 10 m wind speed in m s⁻¹, T is the SST in K, and Tw is an empirical temperature correction factor (dimensionless), accounting for the effect of SST on sea salt emissions. This parameterization is applied to three sea salt modes with median diameters of 0.1, 3.0 and 30 µm, with standard deviations of 1.9, 2.0 and 1.7, respectively . This enhancement enables a more accurate representation of the regional and seasonal variability of sea salt concentrations, especially in tropical and subtropical ocean regions where conventional parameterizations often underestimate emissions. The new parameterization thus represents an important step towards a more physically consistent modeling of marine aerosol sources on a global scale. We note that the factor 10-3 in the last term of the equation is mistakenly missing in the original publication by and is corrected in Eq. ().

MMS and G14 sea-salt emission schemes differ not only in their whitecap formulations but also in their treatment of particle-size distributions and SST-dependent scaling . demonstrated that SST corrections can substantially reduce coarse-mode concentrations and may even have a larger impact than switching between source functions. They also found the strongest divergences for particles larger than PM_2.5, with SST effects further amplifying these differences. These insights highlight that the structural contrasts between MMS and G14 schemes, particularly the inclusion of SST dependence and the size-resolved flux formulation, can significantly influence emitted mass. While a quantitative evaluation is beyond the scope of this study, this context helps to clarify the expected behavior of the new G14 implementation.

DMS has multiple sources with one large contribution from oceanic biogenic emissions . As such, these emissions are highly dependent on the exchange between ocean and atmosphere and hence on wind speed. Therefore, an online calculation is needed to account for DMS emissions from the ocean.

provided a 1° × 1° monthly climatology of DMS ocean surface concentrations (Cw) based on measured data. These climatological values are converted in ICON-ART to the DMS emission flux FDMS using the following parameterization : 5FDMS=Cw⋅MDMS3.6×1020.222u10m2+0.333u10m.

In this equation, MDMS=6.21 gmol-1 is the DMS molar weight, Cw has the unit molL-1 and u10m is the wind speed at the altitude of 10 m above sea level in ms-1. The emission flux is added to the DMS tracer mixing ratio using the same procedure as for other emissions .

The pollen emission model is based on the EMPOL approach . The basic idea is that the emission process can be divided into two steps: (1) opening of the anthers and the accumulation of pollen in a reservoir; (2) release of pollen from the reservoir into the air, driven by turbulence.

The opening of the anthers is controlled by phenology, temperature and humidity. The phenology is modeled using a temperature sum approach, as described by . In general, warm and dry conditions favor opening of the anthers. To calculate the amount of pollen released into the reservoir, a plant distribution map is required. When turbulence is sufficiently strong, the pollen reservoir is emptied and the pollen becomes airborne. This process is parameterized using the Turbulent Kinetic Energy (TKE). Once airborne, pollen is treated as a passive tracer and can be removed by dry and wet deposition. Re-suspension of pollen is not considered by the model.

The pollen emission model was originally designed for COSMO-ART and later implemented in ICON-ART. Currently, implemented plant species include hazel, alder, birch, grasses and ambrosia. Each species is handled slightly differently within the model. Recent developments include an emission implementation for hazel and the integration of real-time pollen data. The approach is described by . The real-time pollen data is used in two ways. First, the start of the flowering season is set to the date when the first pollen are observed. Second, the emission flux is scaled so that the modeled concentrations match the observed values.

ICON-ART supports comprehensive and scalable gas-phase chemistry simulations through the integration of the atmospheric chemistry module MECCA (Module Efficiently Calculating the Chemistry of the Atmosphere), which is part of the CAABA (Chemistry As A Boxmodel Application) framework . MECCA provides a flexible and extensible platform for representing detailed chemical processes in the troposphere and stratosphere, including oxidation pathways, radical chemistry, and heterogeneous reactions.

The numerical integration of the chemical system is performed using the Kinetic Pre-Processor (KPP) , which generates optimized code for solving systems of ordinary differential equations representing chemical kinetics. ICON-ART uses the Rosenbrock solver within KPP for efficient and stable time integration, particularly suitable for stiff chemical systems.

Photolysis rates, a critical component of atmospheric chemistry, are computed using CloudJ , an advanced photolysis scheme that accounts for the effects of clouds, aerosols, and molecular absorption in a multi-wavelength approach. CloudJ can be used in offline or online configurations, depending on the application and computational cost considerations.

A key feature of the ICON-ART tracer framework is the use of MECCA as an external preprocessor, which enables users to define and compile custom chemical mechanisms tailored to specific scientific questions or case studies. This modular approach allows for the integration of predefined standard mechanisms as well as user-defined chemical schemes. The integration of MECCA into ICON-ART has been previously described in detail by . In this section, we summarize the available chemistry mechanisms currently implemented in ICON-ART and provide guidance on their configuration and application domains.

Polar stratospheric clouds (PSCs) are an essential part for polar ozone depletion e.g.,, and therefore have to be accounted for in atmopsheric chemistry models. In ICON-ART, the three major known types of PSCs are implemented. Nitric acid trihydrate (NAT) particles are calculated based on the scheme introduced by using a fixed measured size distribution where each size bin is transported as a passive tracer. Size changes are based on the saturation conditions for NAT particles, based on . The calculation of supercooled ternary solution (STS) particles uses the diagnostic scheme by with an adapted volume concentration . For ice particles, it has been found that lifting the altitude where the ICON microphysics are calculated to the lower stratosphere leads to realistic formation and transport of these particles . Further details can be found in where the scheme has also been applied to local grid refinements around the Antarctica.

It is crucial for an aerosol model to account for key dimensions of aerosol variability, such as chemical composition, size distribution, and mixing state . ART is capable of representing these aspects, allowing for a comprehensive description of aerosol interactions and impacts. It considers a diverse range of aerosol types, including carbonaceous particles, mineral dust, volcanic ash, sea salt, and water-soluble species such as sulfates, nitrates, and ammonium. Table  categorizes these aerosols based on their properties such as size modes, mixing states, and solubility. It distinguishes between externally and internally mixed aerosols, with some species exhibiting a mixed state depending on atmospheric conditions. The listed aerosols span a range of size modes from Aitken to giant, reflecting their diverse sources and transport characteristics. While sulfate, nitrate, ammonium, and water are fully or partially soluble and internally mixed, other aerosols, such as carbonaceous particles, dust, volcanic ash, and sea salt, tend to be less soluble and can exist in both external and internal mixtures. If the mixed model includes only soluble species, the composition is assumed to be volume-averaged. However, if insoluble species are present in the mixed mode, the model assumes a core-shell structure. This assumption is critical for the aerosol optical properties, which are addressed in Sect. .

The ICON-ART model considers aerosol dynamical processes using its AERODYN module , which simulates key processes such as nucleation, condensation, coagulation, and gas-aerosol partitioning. AERODYN allows to consider a flexible number of log-normal modes (up to 10) to represent various aerosol sizes (Aitken, accumulation, coarse) and three different mixing states (soluble, insoluble, mixed) as well as an insoluble giant mode, which is not involved in the aerosol dynamic processes. Table  summarizes all possible compounds in the different modes. As aerosol dynamics changes the sizes and mixing states of particles, two mechanisms shift particles to other modes: (1) a shift to larger modes, when a threshold diameter is exceeded, and (2) a shift to the corresponding mixed mode when the soluble components in the modes exceed a 5 % mass threshold .

Nucleation forms new soluble Aitken particles from gaseous H₂SO₄ increasing the zeroth and third moment of the Aitken mode. The parametrization used in AERODYN is based on . It calculates a critical H₂SO₄ concentration which depends on temperature and relative humidity and above which it is assumed that H₂SO₄ nucleates. In addition to forming new particles through nucleation, H₂SO₄ can condense in all modes (except the insoluble giant mode) leading to an increase of the third moment of a mode and a particle growth. The condensation is parameterized based on and was adapted in ART from .

Intermodal and intramodal coagulation can be activated between all aerosol modes. Coagulation increases the median diameter and reduces the total number concentration within the aerosol size distribution. The parameterizations for the coagulation rates of the zeroth and third moments are based on and references therein, particularly .

Although the coagulation matrix is flexible and can be user-defined, several key aspects should be considered regarding the assignment of particles to modes after coagulation:

In the case of intramodal coagulation, particles remain within the same mode.

The water uptake by aerosols in the atmosphere depends on the ambient relative humidity and the chemical composition of the aerosol. The chemical composition is determined by the molality, which is the amount of a component per kg of solvent. As molality or relative humidity increases, the liquid water content also increases. Conversely, crystallization can occur when the relative humidity falls below the efflorescence relative humidity (ERH). The relative humidity at which no further water uptake occurs depends on the aerosol components and their efflorescence properties. Therefore, determining the crystallization point for an internally mixed aerosol is significantly more complex compared to an externally mixed aerosol.

The thermodynamic model ISORROPIA-2 is used to calculate the water uptake by aerosols. However, this comes with computational overhead, which is undesirable, especially when focusing solely on water uptake on aerosols like sea salt. Due to the high hygroscopicity of sea salt aerosol, which increases aerosol particle mass and consequently affects processes such as removal mechanisms and optical properties, an alternative method for water uptake has been implemented in ICON-ART. This alternative method, previously applied in the COSMO-ART, is described in . In contrast to ISORROPIA-2, this approach bypasses full thermodynamic equilibrium calculations and instead uses an empirical hygroscopic growth formulation for sea salt. This greatly reduces computational cost while retaining the key impact of water uptake on particle mass and related aerosol processes.

As explained in Sect. , ART employs a parameterization for pollen emission (based on EMPOL-parameterization, ) and uses it to forecast alder, birch, grass and ragweed pollen at MeteoSwiss and DWD operationally. Due to their large size, pollen are not considered in the majority of processes in ICON-ART. Although they seem to be quite efficient in nucleating ice or as cloud condensation nuclei, they do not reach relevant altitudes in sufficiently large numbers to actually impact microphysical processes. In recent years, so-called subpollen particles (SPPs), released by those large pollen grains rupturing and bursting, have received increasing attention, since their significantly smaller size enables them to reach those altitudes in larger numbers. To reflect this increased interest, implemented a pollen bursting parameterization into ICON-ART based on physical assumptions, observations and processes (according to ), enabling it to emit SPPs.

The parameterization's driver is the turgor pressure that builds up inside each pollen grain. Once this turgor pressure reaches a pollen specific critical value that the pollen walls can no longer withstand the pollen bursts and releases SPPs into the atmosphere. The turgor pressure's temporal development is formulated to be: 15pTtm=pa+Δπ1-exp⁡-3Ekrρwρ0tm. In Eq. (), pT and pa are the turgor pressure and the ambient pressure, respectively, in Pa, tm is the model time step in s, Δπ is the difference in water potential in Pa, E is the compression module of water in N m⁻², k is the pollen grain's water permeability in m s⁻¹ Pa⁻¹, r is the pollen grain's radius in m and ρw and ρ0 are water and pollen density, respectively, in kg m⁻³.

A complete description of this parameterization is given in . reported that SPP concentrations (used only as INP, not as CCN) vary between 102×106 m⁻³, with a mean value of 4×103 m⁻³ (particularly relevant in warmer levels suitable for biological ice nucleation) while mean pollen concentration amount to 3.4×10-3 m⁻³.

ICON uses ecRad as the standard radiation scheme for numerical weather prediction . To account for ARI, ART computes the local radiative transfer parameters based on the aerosol optical properties – mass extinction coefficient, single scattering albedo, and asymmetry parameter – as well as the prognostic aerosol mass concentrations at each grid point and model level. These parameters are then passed as input to the radiation scheme .

In addition to the prognostic ARI calculations, ART includes forward operators to diagnose aerosol optical depth (AOD) and attenuated backscatter at different wavelengths. These diagnostics are derived by multiplying the prognosed aerosol mass concentrations with mass extinction and backscatter coefficients .

ART provides two approaches for specifying aerosol optical properties. In the first, these properties are precomputed offline and stored in lookup tables for use in both prognostic and diagnostic calculations . This method accounts for the variability in aerosol size distribution by applying polynomial fits to the optical properties . However, it is limited in capturing variations in particle composition and mixing state (e.g., coating), particularly when aerosol dynamics lead to the formation of internally mixed particles during simulations.

To address these limitations, the second approach computes aerosol optical properties online using MieAI , enabling a more flexible and physically consistent representation. Further details on both approaches are provided below.

To investigate aerosol–cloud interactions (ACI), ART is coupled with the two-moment cloud microphysics scheme of ICON , which provides a detailed and physically consistent representation of cloud processes. This scheme includes six hydrometeor categories – cloud droplets, ice crystals, rain, snow, graupel, and hail – and predicts both their mass and number concentrations through a set of prognostic budget equations. The evolution of liquid-phase hydrometeors is governed by processes such as condensational growth, autoconversion, accretion, self-collection, breakup, and freezing. In contrast, ice-phase hydrometeors are subject to diffusional growth, aggregation, self-collection, riming, secondary ice production (ice multiplication), and melting. The two-moment formulation enhances the model’s ability to capture the sensitivity of cloud development and precipitation formation to aerosol perturbations, making it well-suited for studies of ACI .

ART is currently designed as a submodel for ICON such that the development and organization of ART is independent from the base model in most aspects. Its coupling to ICON, however, is extensive, as Fig.  shows. Apart from several points within the time loop where ART provides additional processes, ART also influences some ICON processes directly. Beyond this process-based coupling, ART actively uses some structures introduced in the ICON code. The most prominent example for this is the tracer array, which ART extends to accommodate the additional tracers, whose set is defined by the user with ICON-ART's flexible tracer framework . Accessing ICON's tracer array allows the ART tracers to be treated the same way as the base ICON tracers with regard to processes like advection or turbulent diffusion. Additionally, ART uses and extends structures that represent tracer metadata and requires grid information provided by ICON.For the bulk of the communication between ICON and ART, a set of interface modules are used. The interfaces enable ICON to call process and other routines in ART altering the model state, and ART to retrieve crucial information and structures to fulfill its tasks. The separation of the different modules is based on the general stages of the model (examples: mo_art_init_interface.f90, mo_art_general_interface.f90 and mo_art_diagnostics_interface.f90) and the processes provided (examples: mo_art_sedimentation_interface.f90, mo_art_aerodyn_interface.f90, mo_art_coagulation_interface.f90 and mo_art_reaction_interface.f90). While previously the interfacing between ICON and ART was managed in the host model, it moved to the ART codebase in recent developments. This reduces the additional infrastructural burden when developing ART, since the ICON codebase is only altered if actually required.

ART showed great success using its flexible tracer framework introduced by in such a way that further development was conducted to extend this flexibility to processes. One of these processes is coagulation, which requires a coagulation-matrix describing potential mode shifts when coagulation occurs. This matrix is to be provided to ART via a corresponding XML-file, while also adding a new XML-tag for the tracers indicating whether or not they should be considered for coagulation. An important part for the consideration of aerosols and tracers is their emission into the atmosphere. Emission parameterizations were initially assigned hardcoded by their and the corresponding tracers' name. This put a strain to the actual flexibility achieved, so a new method was introduced. By storing information (like number of emission fluxes or sizes of the emitted material) about a specific emission parameterization and assigning substances to it in a new XML-structure, a mapping routine can map the emission fluxes to the correct mode of the specific substance before calling the emission parameterization itself. Also the new OEM uses XMLs to control its behaviour in ART. The usability of the underlying XML-structure for providing the model with information also goes beyond the actual tracers, modes and processes, since diagnostic variables can also be defined with it.

Since the development of ART has been largely driven by scientific motivations, particularly in the implementation of new processes and features, the resulting historically grown codebase contains numerous cross-dependencies between modules. This complexity makes it difficult to trace the code flow and hinders further development and maintenance. To address this issue, modularization efforts have been undertaken in recent years, streamlining the codebase, reducing inter-module dependencies, and providing a more intuitive interface. An additional benefit of these modularization efforts is the simplified parallelization and the potential to extend ART for use on new heterogeneous and future exascale HPC architectures.

Due to both its scientific design and historically evolved codebase, most processes, features, and modules have been developed exclusively for CPU usage. Although CPU parallelization is implemented and widely used across large parts of ICON, ART rarely exhibits comprehensive parallelization of its features. In recent years, the use of GPUs has gained increasing interest in scientific fields and is likely to continue doing so. Hence, more of the ART codebase needs to be ported to heterogeneous HPC architectures with GPUs to remain competitive and future-proof - due to its operational use at MeteoSwiss pollen-related ART-parts were already ported. As a first step, the online emissions module (OEM) and the vegetation photosynthesis and respiration model (VPRM) of ART have been refactored and ported using OpenACC compiler directives, following ICON's GPU porting approach in an attempt to accelerate the porting process while maintaining a unified codebase for both CPUs and GPUs. VPRM was introduced to enable the simulation of atmospheric carbon dioxide, which is not only affected by anthropogenic emissions but also by exchange with the biosphere. A first application of VPRM in ICON-ART was demonstrated by . Inverse emission estimation with ICON-ART as first demonstrated by and is now possible with a GPU-accelerated version of ICON-ART and has already been successfully tested and validated on Switzerland’s new HPC system, ALPS, leveraging NVIDIA’s Grace-Hopper superchips. A systematic quantitative evaluation of CPU versus GPU performance for ICON-ART with OEM, including detailed speedup analyses, is currently underway and will be presented in a forthcoming dedicated study.

In ICON-ART, a set of standard experiment configurations has been established to ensure consistent testing and validation across model releases. Each configuration, see Table , combines specific model components, tracer species, and physical or chemical processes relevant to targeted research or operational applications. For example, “Dust” focuses on mineral dust and its radiative feedback, while “Volcano” emphasizes volcanic ash and aerosol-radiation interactions. “Nat. Aero.” incorporates a wide range of aerosol tracers and processes without radiative coupling, providing a comprehensive base for sensitivity studies. “Pollen” addresses both pollen and subpollen particles, representing bio-aerosols, and their relevant processes and the chemistry experiments (“Simp. Chem.” and “Comp. Chem.”) are tailored for atmospheric composition forecasts, differing in the complexity of the chemistry scheme being used. “OEM” is configured for regional applications with full chemistry and online emission treatment. These standard configurations are fully supported, regularly maintained, and systematically tested with every ICON-ART release to ensure model integrity, reproducibility, and performance across use cases.

All standard experiments are also integrated into the workflow management solution auto-icon , which enables users to orchestrate a simulation workflow covering the retrieval of input data, compilation, and execution of the model as well as post-processing and visualization tasks. This integration enables a quick and smooth start for new users as well as easy-to-use and flexible testing and evaluation capabilities with the provided experiments during model development. In more complex model applications, auto-icon can significantly enhance throughput by executing all non-dependent tasks in parallel and wrapping individual tasks into larger allocations on the HPC system, which was shown to reduce the time-to-solution by 7 % or more .