In HAM-lite, aerosols are represented as an ensemble of internally mixed, log-normal modes with prescribed sizes and compositions see. The size of a mode can fall within the nucleation, Aitken, accumulation, or coarse size range from below 0.005 µm to above 0.5 µm. And the composition of each mode can represent an arbitrary mixture of aerosol components. In this study, the default configuration includes mineral dust, sea salt, organic carbon, and black carbon, with all particles within a mode sharing the same composition . Depending on the application, the relative volume fractions of these species can vary, ranging from a pure dust mode to predominantly carbonaceous mixtures with minor sulfate contributions. An example configuration of aerosol composition and modal properties used in this study is summarized in Table .

The size distribution of aerosols is given by J log-normal aerosol modes: 1N(ln⁡r)=∑j=1JNj2πln⁡σjexp⁡-(ln⁡r-ln⁡r‾j)22ln⁡2σj, in which Nj is the number concentration, r‾j is the median radius, and σj is the geometric standard deviation of mode j .

In contrast to the HAM scheme, HAM-lite treats only the number concentrations as a prognostic variable, while the median radius (and composition) are prescribed. Consequently, particle properties, like density and hygroscopicity, can be precomputed as volume-weighted averages: 2ρj=∑k=1Kαj,kρk,   κj=∑k=1Kαj,kκk, where K is the number of species, αj,k is the volume fraction of species k in mode j, and ρk and κk are the density and hygroscopicity of species k. The wet radius and wet density are tabulated as a function of air temperature and relative humidity using growth factors from . The configuration of aerosol modes in HAM-lite is flexible and can be adapted to the research goals and available computational resources; additional components can be included by adding another mode. The optimal choice of prescribed aerosol properties may differ between global and regional configurations, depending on aerosol-specific transport, mixing, and aging characteristics across scales.

HAM-lite has been coupled to a new configuration of the ICON ESM designed for kilometer- and sub-kilometer-scale simulations, as described by . The aerosol modes of HAM-lite are added as prognostic tracers to the dynamical core of ICON, and the aerosol processes of HAM-lite are coupled with the atmospheric processes of ICON. The ICON model can operate in global and regional domains. In addition, it is possible to refine the resolution locally by adding nested domains, in which the nested domain feeds back to the outer domain via its boundaries.

To enable regional simulations with the ICON-HAM-lite model, we revised the handling of aerosol tracers as follows. First, we extended the tracer structures in HAM-lite by an additional dimension for the model domain or patch index, consistent with the existing ICON framework. Second, we integrated the aerosol tracers of HAM-lite into the reading routines for initial and lateral boundary conditions, allowing aerosol fields to be initialized and updated consistently with water tracers of ICON via the same data files. Third, we developed new preprocessing tools to extract the boundary conditions, anthropogenic emissions and surface properties from global, generic data sets. The boundary conditions for the aerosols can be taken from global HAM-lite simulations, the Copernicus Atmosphere Monitoring Service (CAMS; ), the MERRA-2 reanalysis from NASA , or even from idealized profiles.

Emissions are either calculated online or prescribed according to emission inventories and associated scenarios. In the absence of explicit aerosol microphysics, emitted species are assigned directly to the prescribed modes, omitting intermediate transformation steps. These include secondary inorganic and organic aerosol formation via gas-to-particle conversion, as well as aging processes such as coagulation and condensation-driven particle growth. Their effects are nevertheless partly represented through the prescribed modal composition and size.

Anthropogenic emissions are based on commonly used inventories e.g., typically reported as monthly means, while wildfire emissions are taken from satellite products e.g., available as daily means. As newly implemented in this study, the sensible heat flux during wildfire events is incorporated using the satellite-derived fire radiative power (FRP), enabling grid cells with active fires to produce more realistic buoyancy-driven plumes of wildfire smoke. The specific emission inventories used in this study, as well as the FRP-based heat flux scaling, are described in detail in Section 2.5.1 “Model Configuration”. Sea salt emissions are calculated interactively following the parameterization by , which depends on surface wind speed and sea surface temperature. Similarly, dust emissions are calculated interactively using the scheme of , with modifications described by and . The emissions are added to the aerosol modes of HAM-lite without any intermediate steps. To feed emissions into the modes, the mass fluxes from the emission datasets are converted into number fluxes. For mixed modes with multiple emission sources, the aerosol composition can be precomputed from the predefined emission fluxes as outlined in .

Aerosols are removed from the atmosphere by sedimentation as well as dry and wet deposition. The processes are briefly summarized here. More details can be found in the paper of . Gravitational sedimentation is computed throughout the vertical column using the wet particle radius. The settling velocity is based on Stokes’ law as described by , with non-continuum corrections following . For numerical stability, the settling velocity is limited by the ratio of model layer thickness to time step. Sedimentation from the lowest model layer to the surface, which is driven by surface-dependent turbulent exchange, is treated in the dry deposition scheme.

The dry deposition flux is calculated as the product of number mixing ratio, air density, and dry deposition velocity. This velocity depends on aerodynamic and surface resistances, which vary with land surface types and are integrated over the fractional areas of different surface categories. The formulation follows the scheme of . The aerodynamic resistance is computed using the turbulence parameterization from ICON, which provides the friction velocity and similarity profile . The laminar sub-layer resistance incorporates empirical corrections that distinguish between vegetated and non-vegetated surfaces, with vegetation characteristics such as vegetation fraction and leaf area index provided by the land surface model JSBACH . Collection efficiencies due to Brownian diffusion and impaction depend on the Stokes and Schmidt numbers, with the latter incorporating a diffusion coefficient corrected for non-continuum effects following . The Stokes number is formulated differently for vegetated and non-vegetated surfaces, using characteristic collector sizes specific to surface types such as coniferous and deciduous forests, grasslands, and water bodies. Similar to sedimentation, a numerical limiter is applied to the dry deposition velocity to ensure stability, based on the surface layer thickness and model time step. The resulting dry deposition flux is subtracted from the surface emission flux such that a net flux is passed to the turbulence scheme.

Wet deposition is implemented as a column-integrated process acting on cloudy model levels. A grid cell is considered cloudy when the cloud water and cloud ice mixing ratio exceed a threshold of 10⁻⁶ kgkg-1 consistent with the criteria used in the cloud microphysics scheme . The wet deposition tendency combines three factors: (1) the number of aerosol particles activated at cloud base, which depends on the cloud base height and the local activation fraction; (2) the ratio of precipitating hydrometeors (rain, graupel, snow) to the total condensed water content (including cloud water and ice), representing the efficiency of precipitation formation; and (3) the ratio of surface precipitation to the column-integrated precipitating water content, representing the fraction of precipitation that reaches the surface. Together, these terms quantify the scavenging efficiency and removal of aerosols through in-cloud and below-cloud wet deposition processes.

The used ICON cloud microphysics is represented by the one-moment scheme of , which computes the masses of water vapor and five hydrometeor types, including cloud water, cloud ice, rain, snow, and graupel. The number concentrations of cloud droplets and ice particles are not prognostic but are diagnosed based on aerosol and meteorological conditions. As described by , the aerosol module HAM-lite interacts with cloud microphysics via droplet activation and wet deposition processes. Cloud droplet activation is computed using the parameterization of , which essentially depends on the updraft velocity and wet particle radius. The activated particles of all modes are added together to estimate the cloud droplet number concentration at cloud base, which is then assumed to remain vertically constant throughout the entire cloud column. Since a minimum cloud droplet number concentration of 30 cm-3 is imposed, similar to the threshold used by . The diagnosed number concentration is used both in the autoconversion rate of cloud droplets within the microphysics scheme and in the calculation of cloud radiative properties within the radiation scheme .

The radiative transfer in ICON is parameterized using the scheme by , which computes radiative fluxes across 14 shortwave and 16 longwave spectral bands. The aerosol tracers from HAM-lite are integrated into the radiative transfer calculations. Aerosol optical properties are derived at each vertical level using Mie theory following . These properties are determined from look-up tables as functions of standard deviation, size parameter, and complex refractive index. The refractive index is computed as a volume-weighted average over all aerosol species, including water. The extinction coefficient, single scattering albedo, and asymmetry factor are computed per aerosol mode and combined to yield bulk optical properties. The extinction coefficient is computed for both longwave and shortwave radiation, whereas the single scattering albedo and the asymmetry factor are evaluated only for the shortwave bands . As dry size and composition in HAM-lite are prescribed, aerosol radiative properties are pre-calculated during the initialization stage of the model and tabulated as a function of relative humidity.

To demonstrate the flexibility and applicability of the new regional version of ICON-HAM-lite, we perform three distinct case studies, each targeting a specific geographic region particularly relevant to one or more aerosol modes of HAM-lite. The regions are chosen to cover key aerosol regimes: (1) Central Europe for anthropogenic sulfate and carbonaceous aerosols, (2) Atlantic Arctic for sea salt aerosol, and (3) Southeast Australia for desert dust and wildfire aerosol. With these three studies, we aim to show the ability of ICON-HAM-lite to represent region-specific aerosol processes in distinct environments across the globe.

The LAM simulations with ICON-HAM-lite were carried out and analyzed on the Levante high-performance computing system at the . A total of 64 compute nodes were utilized, each equipped with 128 CPU cores and 256 GB working memory. The system requirements vary depending on the domain size. Therefore, to provide a basis for performance estimation, a computational unit (cu) is defined as 1 million grid cells × 1000 time steps. With a model time step of 10 s, this corresponds to 8.6 cu per simulated day for a domain with 1 million grid cells, and thus to 86 cu for a ten-day period. The values reported were averaged from 10 daily runs for each of the three case studies (see Table ). For 1 cu, ICON-HAM-lite at R2B10 requires approximately 4 node hours. To put these performance metrics into context, the R2B10 LAM configuration in this study spans a range from about 2 cu for the German domain with roughly 0.2 million grid cells to about 14 cu for the Australian domain with roughly 1.6 million grid cells, with all cu values referring to the computational cost for one simulated day. This amounts to 8 node hours per simulated day for the German case and 56 node hours for the Australian case. In contrast, a global R2B10 simulation with roughly 84 million grid cells would require about 725 cu, corresponding to approximately 3000 node hours per simulated day. This illustrates the significant reduction in computational cost that can be achieved with the LAM version, provided the scientific focus can be limited to a regional domain.

The ability of the model to reproduce particulate pollution over Central Europe, and Germany in particular, is evaluated by comparing the model results with ground-based observations from operational air quality monitoring networks. These observational datasets are provided by the European Environment Agency (EEA), which offers access through its download service (https://www.eea.europa.eu/en/datahub, last access: 15 June 2025) to time series of air composition at more than 5600 sites across Europe. The primary parameter analyzed in this study is fine particulate matter (PM2.5), i.e., particles with an aerodynamic diameter of 2.5 µm or less. It is a key indicator for the assessment of air quality with regard to human health. The dataset contains hourly PM2.5 concentration values from monitoring stations ranging from rural to urban sites, enabling spatiotemporal comparison with model simulations. This allows for a robust evaluation of the model's ability to represent observed air pollution levels and their time evolution under real-world conditions, including the four aerosol modes dust, sea salt, carbonaceous and sulfuric aerosol that together constitute PM2.5 in HAM-lite. For comparison with the measurements, aerosol mass concentrations are calculated offline from the simulated number concentrations using the volume-mean radius and prescribed particle density of each log-normally distributed aerosol mode. Total aerosol mass is obtained by summing contributions from the four model aerosol modes, and a PM2.5 size cut-off is applied.

Modeled sea salt aerosol is evaluated by comparison with mass concentrations derived from in-situ chloride ion (Cl-) measurements collected by the European Monitoring and Evaluation Programme (EMEP) network. The data, which comprises various inorganic compounds in the gas and particle phase (e.g., ammonium, calcium, chloride, magnesium, sodium, and sulphate), can be obtained from the EBAS database (https://ebas.nilu.no, last access: 10 April 2025), maintained by the Norwegian Institute for Air Research (NILU). Sea salt mass concentrations are estimated from Cl- using a conversion factor of 1.65, corresponding to its weight fraction in sea salt . Measurements from three high-latitude and coastal EMEP stations were used: Kårvatn (62.783° N, 8.883° E; 210.0 ma.s.l.) and Tustervatn (65.833° N, 13.917° E; 439.0 ma.s.l.) in Norway; as well as Irafoss in Iceland (64.083° N, 21.017° W; 66.0 ma.s.l.). These stations provide a reliable observational basis for evaluating the model over the North Atlantic region.

In addition to these stationary and permanent observations, in-situ aerosol measurements from the MOSAiC (Multidisciplinary drifting Observatory for the Study of Arctic Climate) drift experiment are used to evaluate modeled sea spray aerosol in the central Arctic. Conducted from October 2019 to July 2020 aboard the German research vessel (RV) Polarstern, MOSAiC offers a unique opportunity for a process-level evaluation of Arctic aerosol throughout the full Arctic year . Among other measurements, a comprehensive set of in-situ observations was collected that includes the bulk size-resolved chemical composition and mass concentration of non-refractory sub-micron aerosol particles (e.g., sulfate, nitrate, ammonium, chloride, and organics) (see for further details on measurement procedures and data processing).

Lidar observations from the Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP; ) instrument aboard NASA’s Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite are used to retrieve vertical profiles of the extinction coefficient at 532 nm. In this study, the level-2 version 4 aerosol profile product (05kmAPro), averaged over 5 km horizontal segments along the near-nadir ground track, is used to evaluate the atmospheric layering and plume heights of the modeled dust and bushfire aerosol in the Australian case. Although CALIOP often fails to detect thin aerosol layers in the upper troposphere and lower stratosphere e.g.,, the significant aerosol enhancements resulting from the Australian wildfires are clearly captured. These extinction profiles, despite sampling and detection biases in individual profiles, provide valuable constraints for model evaluation. The modeled extinction coefficients are calculated offline from the number mixing ratios using a mass-specific extinction coefficient of 0.5 and 1.0 m2g-1 for desert dust and carbonaceous aerosol, respectively , with particle properties summarized in Table .

PM2.5 is a core target parameter for both health risk assessments and air quality regulations, but also plays a complex and significant role in the Earth’s climate. Dominant sources in Germany are primary emissions from road traffic, industry, domestic heating, agriculture, and construction. In addition, long-range transport and secondary aerosol formation can contribute substantially to PM2.5 concentrations, particularly during springtime pollution episodes. Consequently, air quality in Germany can be strongly influenced by emissions from neighboring countries and meteorological conditions, which can trap pollutants or enhance long-range transport impacts e.g.,. In 2019, all German air quality monitoring stations complied with the EU annual mean limit for PM2.5 of 25 µgm-3, continuing the trend of steadily decreasing concentrations since the threshold was implemented in 2015. However, more than half of the nearly 200 stations (57 %) still exceeded the WHO’s stricter annual recommendation of 10 µgm-3, and almost all stations (98 %) recorded more than three daily exceedances of 25 µgm-3, thus failing the WHO’s 24 h guideline .

Figure  shows the spatial distribution of modeled mean PM2.5 concentrations over Central Europe for the period from 3 to 10 March 2019, as simulated by the ICON-HAM-lite model. In HAM-lite, PM2.5 represents the total mass concentration of four aerosol modes: dust, sea salt, carbonaceous, and sulfuric. For the comparison to measurements, the modal number concentrations are integrated over their size distributions and truncated at a 2.5 µm diameter cutoff. Observations from the European Environmental Agency (EEA) air quality monitoring network for Germany are superimposed on the map as circles. The modeled PM2.5 concentrations exhibit pronounced regional gradients. Low concentrations below 4.0 µgm-3 dominate rural and mountainous areas such as central, southern, and northeastern Germany, the Alps, and eastern France. High concentrations up to 25.0 µgm-3 occur in densely populated and industrialized regions – especially northwestern Germany, i.e. Ruhr region, Rhine–Main region, the Bohemian basin in Czech Republic, southwestern Poland, and the Po Valley in northern Italy. These hot-spots align with sources of high anthropogenic emissions and regions with limited atmospheric ventilation. The model captures the major regional features and concentration gradients reasonably well. However, it tends to underestimate PM2.5 levels in central remote areas and in an eastern range from the Thuringian Forest to the Bavarian Forest, as well as in the Black Forest region in the southwest, likely due to the absence of processes forming secondary inorganic and organic aerosols. Secondary inorganic aerosol components such as ammonium and nitrate, which can contribute substantially to PM2.5 during pollution episodes and long-range transport events, are not explicitly represented in the present HAM-lite configuration. Conversely, peak concentrations are frequently overestimated in urban and industrial regions, possibly due to simplified assumptions in the time profiles of the prescribed anthropogenic emissions.

To further examine the spatial context, time series of PM2.5 concentrations are shown for two urban/industrial sites (Frankfurt am Main and Merseburg near Leipzig) and two rural background sites (Buckow in Märkische Schweiz and Burghausen in Bavaria) in Fig. . These plots compare hourly modeled values with observed concentrations over the same period. In the observations, rural sites show average concentrations around 6.5 µgm-3 (Buckow) and 4.4 µgm-3 (Burghausen), with peak concentrations reaching up to 21.2 µgm-3 (Fig. b an d). While mean PM2.5 levels at urban and rural sites partly overlap, urban locations show enhanced temporal variability and more pronounced short-term pollution episodes. The average urban concentrations lie at 5.5 µgm-3 (Frankfurt) and 4.1 µgm-3 (Merseburg), and the peak concentrations are up to 18.0 and 13.3 µgm-3, respectively (Fig. a and c). The model reproduces the timing of concentration peaks relatively well in urban areas but consistently overestimates their magnitude. In rural areas, modeled values show better agreement in magnitude but tend to underestimate short-term peaks, suggesting missing processes related to secondary aerosol formation or long-range transport.

This is consistent with quantitative evaluation metrics shown in Fig. , which indicate moderate agreement between simulations and observations, with correlation coefficients of 0.52–0.73 and normalized RMS errors of 0.33–0.65 across the four stations. A broader statistical evaluation across all German stations is summarized in Table . Rural background stations show the lowest normalized RMS errors (0.67), while industrial stations exhibit the largest errors (1.06). Correlation coefficients remain moderate and relatively similar across all station categories (R = 0.37–0.44). The larger errors at industrial sites are consistent with the model tendency to overestimate short-term pollution peaks in regions with strong anthropogenic emissions, whereas the lower variability at rural background stations likely reflects missing secondary aerosol formation and long-range transport contributions.

Together, the spatial and temporal comparisons demonstrate the model’s skill in capturing regional-scale patterns and diurnal variability of PM2.5, while also indicating systematic biases related to the representation of emission sources and the simplification of aerosol microphysics and chemistry.

Sea spray aerosol, which originates from open ocean water and leads and consists mainly of sea salt, is a significant but highly variable source of atmospheric particles in the Atlantic Arctic, especially during the melt season and summer months . When reaching a certain size, the sea spray particles can act as nuclei in cloud formation processes that influence the surface energy budget by scattering and absorbing incoming and outgoing radiation. The MOSAiC 2019–2020 expedition covered extended periods in the Atlantic sector of the Arctic Ocean. The German research vessel (RV) Polarstern drifted through the Fram Strait/Svalbard region during late summer and early autumn 2020 and provided observations relevant to the production of sea spray in open waters and marginal ice zones .

Figure  shows the spatial distribution of surface concentrations of sea salt aerosol averaged from 4 to 12 April 2020 over the Atlantic Arctic region. The map covers the area between Greenland and Svalbard, a region strongly influenced by marine boundary layer dynamics. Observational data from European Monitoring and Evaluation Programme (EMEP) stations and the MOSAiC drift campaign aboard the RV Polarstern are overlaid as circular markers for comparison. The modeled sea salt concentrations range from below 0.03 µgm-3 in the inner Arctic to more than 30 µgm-3 over the Greenland Sea and coastal North Atlantic. Elevated concentrations are most prominent in regions of active sea spray generation, particularly in the vicinity of Iceland and along the Norwegian Sea coast. The simulation captures fine-scale emission structures, such as enhanced concentrations along the east coast of Greenland linked to Piteraq wind events, and along the sea ice edge west of Svalbard associated with cold-air outbreaks.

The comparison in the map is supported by the time series in Fig. , which show daily mean sea salt concentrations at the three EMEP stations and on board RV Polarstern. These time series provide additional context for the point measurements overlaid in Fig. , showing the day-to-day variability relative to the temporal mean in the map. At Tustervatn (Fig. a), observed concentrations vary between about 0.5 and 4 µgm-3, with a pronounced peak around 4 µgm-3 on 7 April. The model captures both the timing and magnitude of this event reasonably well, although the peak concentration is somewhat overestimated. Kårvatn (Fig. b) exhibits observed concentrations between about 0.07 and 2.0 µgm-3 with a peak just above 2 µgm-3. The model reproduces the temporal evolution of the sea salt episode very well, including the timing and magnitude of the main peak, while slightly overestimating concentrations during the later part of the period. At Irafoss (Fig. c), observed concentrations span roughly 0.3–12.0 µgm-3, with a pronounced maximum around 7 April. The model captures the occurrence of this event but underestimates the peak concentration and does not fully reproduce the subsequent decline. Along the RV Polarstern drift (Fig. d), observed concentrations remain relatively low, below 0.1 µgm-3. The model reproduces the low-concentration background conditions and some of the observed variability, although concentrations are generally underestimated during the second half of the period.

Quantitative evaluation metrics shown in Fig.  indicate strong temporal agreement at the Scandinavian stations, with correlation coefficients of 0.86 and 0.95 at Tustervatn and Kårvatn, respectively, while agreement is more moderate at Irafoss and along the RV Polarstern drift (R = 0.51 and 0.55). Normalized RMS errors remain below unity at all stations (0.69–0.94), indicating generally good agreement between simulated and observed concentrations. The remaining discrepancies are primarily related to the magnitude of individual peak events and the representation of low background concentrations.

The comparison with the available measurements shows that ICON–HAM-lite successfully reproduces the key spatial patterns and temporal variability of sea salt aerosol concentrations in this climatically important region, although biases in concentration magnitude remain.

The austral summer of 2019–2020, known as Australia’s Black Summer, saw unprecedented wildfires in southeastern Australia, burning nearly twice the area of any previous record . The activity of wildfires peaked around 29 December to 4 January, during which time deep pyro-convective towers lofted smoke aerosols up to 14–16 km into the lower stratosphere . According to , a remarkable amount of 0.3–1.1 Tg of smoke was injected in these 4 d, which was then rapidly advected eastward across the southern mid- and high latitudes. At the same time, intense dust emissions occurred in the interior of the continent, triggered by frontal passages and moist convection (see Fig. 6 in for a synoptic overview), with mineral dust likewise lofted and transported eastwards. The coexistence of dust and bushfire aerosol makes this case particularly interesting for studying vertical transport processes and aerosol–radiation interactions.

Figure  shows column burdens of desert dust and carbonaceous aerosols over southeastern Australia from 28 to 31 December 2019. The maps illustrate the intense plumes of desert dust and bushfire smoke over southeastern Australia and adjacent oceanic regions. The simulations include the effect of fire radiative heat release on surface temperature, which induces strong buoyant updrafts and enables pyro-convective lofting of smoke emissions from the surface into higher atmospheric layers, in some cases up to the lower stratosphere. Fire radiative power (FRP) from NASA's MODIS observations, as provided by the GFASv1.2 dataset, is used to represent the spatial and temporal variability of fire activity and the corresponding intensifications of the sensible heat flux on the ground.

The smoke is lofted rapidly and dispersed over long distances, with prominent eastward advection over the Tasman Sea and into the South Pacific. Figure a–d, shows the wide spread of carbonaceous aerosol, which is consistent with satellite observations during this period (cf. ). In regions with intense bushfire activity, carbonaceous aerosol burden can locally exceed 15.0 gm-2, while values within the broader fire plume reach up to 0.3 gm-2. Desert dust, mainly originating from central Australia, is also advected southeastward, interacting with the smoke plumes and contributing to the complex aerosol mixture. Maximum dust burdens of locally 7 gm-2 are reached in the source areas and up to 0.3 mgm-2 along the transport path (Fig. e–h).

To illustrate the vertical distribution and long-range transport of aerosols in more detail, Fig. a shows a zonal cross-section of aerosol concentrations averaged over a 10°-wide latitude band centered at 35° S on 31 December 2019. The figure distinguishes between desert dust (colored shading) and carbonaceous aerosol (grayscale shading), originating primarily from the intense bushfire activity in southeastern Australia. Figure b shows the corresponding zonal mean vertical profiles.

Carbonaceous aerosols are clearly dominant, with near-surface concentrations reaching up to about 110.0 µgm-3 in proximity to the wildfire source regions. Elevated smoke layers are also evident above 10 km and extend toward the modeled tropopause at approximately 14 km (indicated by the dashed blue line). These high-altitude plumes are consistent with pyro-convective events that loft smoke and associated aerosol particles into the upper troposphere and lower stratosphere. This vertical transport mechanism plays a key role in extending the atmospheric lifetime of carbonaceous aerosols and enhancing their radiative impact, as particles at these altitudes are subject to reduced removal processes .

In contrast, desert dust shows a more localized and lower-altitude distribution. Dust concentrations reach more than 40 µgm-3 but are largely confined to the lower troposphere, typically below 6 km. This reflects the difference in emission sources and injection mechanisms between the two aerosol types: while dust is lifted by mechanical processes such as turbulent winds and frontal activity, wildfire smoke is buoyantly injected through intense thermal updrafts known as pyro-convection. Nevertheless, a fraction of dust is also lofted to the transport heights of the smoke in this case.

The zonal mean profiles in Fig. b highlight these differences. The carbonaceous aerosol profile shows a broad, bimodal shape with peaks near the surface and at upper tropospheric levels, whereas the dust profile peaks at an altitude of about 2.5 km. These vertical gradients have implications for both short- and longwave radiative forcing in this region and throughout the Southern Hemisphere . While further analysis is beyond the scope of this study, the ICON-HAM-lite LAM model is ideally suited to investigate these effects in future work. In particular, the impacts of the pyro-convective parameterization through the FRP-scaled surface sensible heat flux, as well as the role of complex aerosol–radiative interactions in shaping the observed aerosol layering, will be analyzed in detail in a follow-up study.

To evaluate the aerosol layering, Fig.  shows extinction coefficient profiles at 532 nm from ICON-HAM-lite together with observations from the CALIOP instrument aboard NASA's CALIPSO satellite. Two CALIPSO overpasses from 31 December 2019 at 03:21 and 14:04 UTC that intersect the model domain are selected as shown in the inset map of Fig. . For each track, the profile with the maximum extinction above 5 km altitude is identified to focus on the evaluation of aerosols in the upper troposphere and lower stratosphere (UTLS). This profile is then averaged with adjacent profiles within a 5 km radius along the track to reduce noise and better represent the regional vertical structure.

The profile from the northern CALIPSO track, i.e. track 1 in the inset of Fig. , shows elevated aerosol extinction between approximately 10–12.5 km altitude, with a peak near 11 km reaching values of up to  2 km-1. This suggests lofted smoke layers near the upper troposphere, likely due to pyro-convection, as discussed above. The model reproduces the plume height well but shows somewhat higher extinction values and slightly broader vertical structure. In the southern track, i.e. track 2 in the inset of Fig. , CALIOP shows a pronounced extinction maximum of about 12 km-1 at 2.5 km altitude, along with additional extinction signals above 8 km. This indicates the presence of both boundary-layer smoke and lofted aerosols, including desert dust. Again, the model captures the vertical extent reasonably well but likely underestimates the magnitude of the extinction coefficient. While the modeled profiles show reasonable agreement with observed vertical layering, CALIOP profiles are notably more fragmented. This likely reflects known limitations in CALIOP's ability to detect thin aerosol layers in the upper troposphere and lower stratosphere and to distinguish between aerosol and cloud features e.g.,. Nevertheless, the agreement in plume height provides confidence in the model representation of smoke injection and transport processes.

Overall, these results underscore again the importance of accurately representing aerosol vertical distribution in atmospheric models, particularly for extreme biomass burning events with injection heights, which can reach stratospheric levels . The ICON-HAM-lite LAM version offers a considerable advantage in this context, as it enables explicit calculation of pyro-convective injection instead of relying on uncertain parameterizations. Furthermore, it allows systematic experiments with different sources of uncertainty, which would be prohibitively expensive in a global model configuration.