Primary aerosols are aerosols directly emitted from various sources. In UKESM1.1, these include mineral dust, seasalt, black carbon, organic matter, and sulfate. Mineral dust is simulated using a scheme with six size bins ranging from 0.06 to 60 µm in diameter. The dust within each bin is treated independently and assigned a density of 2.65 kgm-3 .

Seasalt emissions are calculated using the bin-resolved parameterisation of . The emitted mass and number are distributed between the soluble accumulation and coarse modes, depending on whether the centre diameter of the source bin is below or above the diameter threshold separating these modes (approximately 500 nm). A density of 2.165 kgm-3 is assumed for seasalt in UKESM1.1.

Primary carbonaceous aerosol emissions include black carbon and organic matter originating from both anthropogenic sources (biofuel and fossil fuel combustion) and biomass burning processes. Aerosols emitted from biomass/biofuel sources are assigned a geometric mean diameter of 150 nm, while those from fossil fuel sources are assigned 60 nm; both emission types assume standard deviation of 1.59 . Notably, although the assigned diameter for biomass/biofuel aerosol emissions (150 nm) exceed the typical upper size limit of the Aitken mode (100 nm), these aerosols are nevertheless emitted into the model's insoluble Aitken mode. Anthropogenic emissions are released into the lowest model layer, whereas biomass burning emissions are distributed vertically between the surface and approximately 6 km above ground level.

Primary marine organic aerosols represent another primary aerosol source which was recently implemented in UKESM . These organic aerosols are thought to be emitted as components of organic-enriched sea spray , and their emissions show a high correlation with marine biological activity, often indicated by chlorophyll concentrations . Consequently, UKESM1.1 adopts the parameterisation of , which relates emissions to both wind speed and biological activity represented by surface chlorophyll concentration. The calculated organic mass emission flux is partitioned, with 25 % attributed to the soluble Aitken mode and 75 % to the insoluble Aitken mode. An emission diameter of 160 nm is assumed for both fractions, based on experimental and observational constraints .

Consistent with , the model assumes that 2.5 % of anthropogenic SO2 emissions by mass are directly emitted as primary sulfate aerosols. This primary sulfate mass is distributed with an initial size distribution specified by : 50 % is allocated to the accumulation mode (assuming d‾ = 150 nm) and 50 % to the coarse mode (assuming d‾ = 1500 nm).

Sulfate aerosols form through the oxidation of SO2, either via gas-phase reactions that produce H2SO4 and trigger NPF and growth, or through multiphase oxidation by hydrogen peroxide (H2O2) and O3 dissolved in cloud liquid water, which contributes to aerosol mass only. The gas-phase oxidation of SO2 produces gaseous H2SO4, which drives NPF and growth in the model. In the aqueous phase, SO2 dissolves into cloud droplets, where it undergoes oxidation by dissolved H2O2 and O3 to form sulfate. The model does not include an explicit representation of multi-phase chemical species; instead, the sulfate formed through these pathways is treated as a direct mass flux contributing to the accumulation and coarse aerosol modes. Finally, because certain removal processes (e.g. precipitation removal) associated with aqueous sulfate formation are not represented in the model, a 25 % reduction factor is applied to the calculated aqueous sulfate formation mass flux .

In UKCA, SO2 originates from both the oxidation of DMS and direct anthropogenic emissions. Anthropogenic emissions of SO2 are taken from the Community Emissions Data System (CEDS) inventory . SO2 emitted from both the energy and industrial sectors is released into the model's surface layer (up to 35 m), consistent with the treatment of other trace gas emissions in UKCA. Additionally, natural SO2 emission from continuously degassing volcanoes is prescribed using the climatology developed by . A major revision of the SO2 dry deposition scheme, as described by , was implemented in UKESM1.1. This update, along with several additional bug fixes and model improvements detailed in , contributed to reduced surface SO2 concentrations in UKESM1.1 compared to the earlier UKESM1 version evaluated by .

Marine DMS emissions in the model are calculated using an interactive scheme. The ocean biogeochemistry component, incorporating the Model of Ecosystem Dynamics, nutrient Utilisation, Sequestration and Acidification (MEDUSA) module, simulates the DMS concentration in the surface ocean. This follows the formulation modified from to ensure energy balance within the coupled system . Additionally, terrestrial DMS emissions are also included, based on an earlier climatology .

The representation of DMS chemistry within UKCA is simplified, a common necessity in large-scale global models. The primary oxidation pathway for DMS represented in the model is the reaction with OH radicals, which is the dominant atmospheric sink for DMS. Additionally, DMS is also oxidised by nitrate (NO3) radicals and atomic oxygen (O(3P)), although the latter pathway is generally negligible in the troposphere. The default chemistry scheme employed in UKCA is the Strat-Trop scheme , but the specific reactions of DMS oxidation pathways differ between UKESM1 and UKESM1.1 (Table ). The reaction of DMS with OH proceeds via two channels: addition and abstraction. In both UKESM versions, the abstraction channel of OH oxidation, along with DMS oxidation by NO3, leads directly to the formation of SO2. While suggested that the UKESM1.1 included the reaction DMS + O(3P) → SO2, it is not included in the version of UKESM1.1 used in this study. The addition channel in UKESM1.1 differs from that in UKESM1; the latter neglects the formation of dimethyl sulfoxide (DMSO). However, DMSO is recognised as an important intermediate species subject to atmospheric transport and deposition and it is a precursor of MSA. Consequently, DMS oxidation by OH directly produces MSA in UKESM1.0, partially accounting for the neglected DMSO pathway. In UKESM1.1, the addition channel was modified to produce a mixture of SO2 and DMSO, with molar yields of 0.6 and 0.4, respectively. Subsequent DMSO oxidation by OH in UKESM1.1 further produces SO2 and MSA, with yields of 60 % and 40 %, respectively (Table ).

Several key airborne aerosol production processes are simulated by default in UKCA. First, new particles are formed via the binary H2SO4–H2O nucleation scheme, producing aerosols in the smallest (nucleation) mode. Second, aerosols grow via the condensation of H2SO4 and organic oxidation products onto existing aerosol surfaces. In the meantime, coagulation between aerosol particles reduces the number concentration in smaller modes (e.g. the nucleation mode) while contributing to aerosol growth in larger modes (e.g. the Aitken and accumulation modes). When aerosols within a specific mode grow beyond the upper size threshold for that mode, they are transferred to the next largest mode via mode merging .

The default binary H2SO4–H2O aerosol nucleation mechanism follows the theoretical parameterisation described by , which is most effective at producing new particles in the free troposphere. Mechanisms for NPF specifically within the planetary boundary layer are not explicitly included in the default model configuration. However, the model includes an option to use a boundary layer nucleation parameterisation based on cluster activation theory, wherein the nucleation rate exhibits a power-law dependence on the H2SO4 concentration . Additionally, a multi-component nucleation scheme described by , which considers the roles of both H2SO4 and organic vapours, is also available as an option within the model. It should be noted that these optional boundary layer nucleation schemes do not quantitatively represent the state-of-the-science atmospheric nucleation mechanisms and are therefore not utilised in this study .

Condensable organic oxidation products, which contribute to aerosol growth, are represented in the model as originating from the oxidation of monoterpenes, assuming a lumped mass yield of 13 % . However, to account for SOA production from other sources (such as isoprene oxidation) and to compensate for the absence of explicit anthropogenic and marine volatile organic compound (VOC) emissions in this model configuration, this lumped monoterpene oxidation yield is scaled by a factor of two.

Besides nucleation, vapour condensation, and coagulation, other aerosol microphysical processes simulated in the model impact the aerosol lifecycle, including dry deposition, wet scavenging, and the ageing of insoluble aerosols (representing processes that increase aerosol solubility). Readers are referred to for the detailed formulation of these microphysical processes.

The implemented H2SO4–NH3 nucleation scheme is based on the parameterisation developed by , which was derived from measurements performed in the CERN CLOUD (Cosmics Leaving OUtdoor Droplets) experiments. A key feature of the CLOUD experiments, compared with previous laboratory studies, is the extremely high standard of cleanliness. This allows the nucleation capability of individual vapours, and mixtures thereof, to be studied independently with minimal contamination. Taking advantage of this capability, experiments were performed to separately evaluate the influence of H2SO4, NH3, H2O, and atmospheric ions on the aerosol nucleation rate. Readers are referred to the original paper for detailed formulations and parameter values .

Experiments reported in identified four distinct aerosol nucleation regimes, differentiated by the concentrations of NH3 and the presence of atmospheric ions. The first regime is termed binary neutral nucleation of H2SO4–H2O; this represents the experimental version of the theoretical nucleation scheme used by default in UKESM1.1 . This regime's rate depends only on the concentrations of H2SO4 and H2O. The second regime is termed binary ion-induced nucleation of H2SO4–H2O, describing nucleation in the presence of atmospheric ions. Consequently, the nucleation rate of this regime exhibits an additional dependence on the concentration of atmospheric ions. The third regime is ternary neutral nucleation of H2SO4–NH3(–H2O), which depends on the concentrations of H2SO4, NH3, and H2O. As H2O participation is fundamental to atmospheric nucleation, it is often implicitly assumed and omitted from the nomenclature. Finally, the fourth regime is ternary ion-induced nucleation of H2SO4–NH3, which incorporates the enhancing effect of atmospheric ions on the ternary system.

The impact of H2O on aerosol nucleation is represented as a relative humidity dependent multiplier (KRH) applied to the nucleation rate calculated based on H2SO4, NH3, and atmospheric ion concentrations. The formula reads KRH=1+c1(RH-0.38)+c2(RH-0.38)3(T-208)2, where c1 = 1.5 ± 1.3, c2 = 0.045 ± 0.003, RH is the relative humidity expressed as a fraction, and T is the temperature in Kelvin. This formulation uses 38 % RH (RH = 0.38) as the reference condition, reflecting the humidity level at which most of the underlying experiments were performed .

As atmospheric ion pair production rates are not explicitly simulated in UKESM1.1, they are prescribed using a climatology output from the simulations described by . NH3 emissions in UKESM1.1 are taken from the CEDS dataset . However, simulated atmospheric NH3 concentrations within the model are not well constrained. For example, the uptake of gaseous ammonia by acidic aerosols (leading to ammonium formation), which is widely acknowledged as an important NH3 sink, is not represented in the default model configuration. This leads to an overestimation of gas-phase NH3 concentrations in the model. Therefore, the recently developed ammonium nitrate scheme is incorporated in most simulations in this study, aiming to improve the representation of the atmospheric NH3 budget.

The new H2SO4–NH3 nucleation scheme is implemented in all simulations listed in Table , except for the control simulation using the default binary scheme, labelled SA–H2O(default). The SA-NH3(benchmark) simulation represents a key experiment in this study; it incorporates the ammonium nitrate scheme using parameter settings corresponding to the fast nitrate formation scheme (specifically, a nitric acid uptake coefficient γ = 0.193) as described by . Another sensitivity simulation, SA-NH3-slow, explores slower nitrate formation by utilising γ = 0.001. Additionally, the SA-NH3-noNit simulation is performed using the H2SO4–NH3 nucleation scheme but without activating the ammonium nitrate module, serving as a reference to evaluate the impact of the nitrate scheme itself on aerosol formation.

In the default UKESM1.1 configuration , MSA is treated as an inert tracer. This treatment causes MSA to accumulate indefinitely in the model, leading to unrealistically high concentrations. To address this limitation, a scheme of MSA condensation onto pre-existing aerosols, as well as its wet and dry deposition, are implemented in the model in this study.

The wet deposition of MSA is treated analogously to other soluble gas-phase species in the model, following the formulation described by . Consistent with recommendations by and , an effective Henry's law constant of 10⁹ Matm-1 is adopted for MSA. The dry deposition rate of MSA is assumed to be the same as that calculated for gaseous H2SO4, for the similarity in their molecular sizes and, consequently, their diffusivities in air.

Finally, the MSA condensation scheme implemented assumes irreversible condensation onto pre-existing aerosol particles, at the same rate as H2SO4. This assumption is based on the comparable molecular weights (MSA = 96 gmol-1; H2SO4 = 98 gmol-1) and bulk densities (MSA = 1.5 gcm-3; H2SO4 = 1.8 gcm-3) of the two species. This treatment is supported by recent Arctic observations which suggest that MSA can condense effectively onto pre-existing aerosols, potentially at a rate comparable to that of H2SO4 . The MSA aerosol mass is currently merged with the existing sulfate aerosol mass in the model for simplicity. The MSA condensation scheme is added to the SA-NH3 (benchmark), hereafter referred to as SA-NH3-MSA.

However, it must be emphasised that this MSA condensation scheme is preliminary and has not yet been fully validated against comprehensive experimental data. For instance, a potential humidity dependence of MSA partitioning to the aerosol phase has been implicated in aircraft measurements . Other modelling studies have adopted volatility-dependent parameterisations for MSA condensation, assuming it to be a temperature- and humidity-dependent process . Unfortunately, there is currently limited experimental data available to rigorously validate MSA condensation rates predicted by either type of scheme; future experimental work is crucial for developing and constraining MSA condensation parameterisations in models. Thus, the MSA condensation scheme in this study aims to provide an initial proof-of-concept implementation, allowing for the evaluation of MSA's potential impact within the current chemistry framework.

Although the default UKESM1.1 configuration employs an interactive marine DMS emission scheme coupled to the MEDUSA module, this study also evaluates the widely used DMS climatology developed by . This comparison is motivated by the significant uncertainties associated with modelled DMS emissions. For example, demonstrated that simulated DMS emissions in the Southern Ocean are highly sensitive to the choice of emission parameterisation. Specifically, the seawater DMS concentrations predicted by the interactive MEDUSA scheme generally exhibit limited spatial variability, although they show low values surrounding the Antarctic continent. In contrast, the climatology predicts distinct hotspots of high seawater DMS concentration near Antarctica. While rigorously verifying the true spatial pattern of global DMS emissions is beyond the scope of this study, the distinct differences between the interactive scheme and the climatology are substantial. This provides an opportunity to evaluate the sensitivity of simulated atmospheric chemistry and aerosol properties in response to different DMS source mechanisms. The DMS climatology is implemented in the simulation labelled SA-NH3-Lana, which also utilises the H2SO4–NH3 nucleation scheme and the ammonium nitrate scheme.

The coupling of the Common Representative Intermediates mechanism (CRI) with the existing UKCA stratospheric chemistry scheme (Strat) was initially carried out by and subsequently updated by to create the CRI-Strat2 (CS2) chemistry scheme. The development of CS2 aimed to improve the model's representation of the oxidation of non-methane volatile organic compounds and provide traceability to the CRI2.2 scheme . For instance, compared to the default Strat-Trop scheme, the CS2 scheme includes a more detailed representation of the oxidation pathways for DMS and other key biogenic VOCs, such as isoprene and monoterpenes. It is worth noting that the original CS2 scheme is not only a new chemistry mechanism but also uses different emission inventories – distinct from those used in simulations with the Strat-Trop chemistry scheme. Details of the emission inventories employed in the CS2 scheme are provided in and . For example, most anthropogenic and biogenic emissions are based on datasets spanning 2000–2010 or on a 2010 time-slice. In this study, we updated selected anthropogenic emissions – including SO2, nitrogen monoxide, organic carbon, black carbon, ammonia, methane, and carbon monoxide – to use CMIP6 SSP3-7.0 emissions to be consistent with the ST simulations. Emissions of volatile organic compounds (VOCs), most of which are unique in CS2, are left unchanged and therefore remain consistent with the original implementations described by and . This modification removes the influence of inconsistent anthropogenic emissions, such that the comparison between CS2 and ST presented in this study primarily reflects differences in the underlying chemistry schemes, as well as the impact of the additional VOCs included in CS2.

The key reaction pathways of the DMS oxidation within the CS2 version used in this study are tabulated in Table . Similarly to other sensitivity simulations, the CS2 scheme is coupled with the H2SO4–NH3 nucleation scheme and the ammonium nitrate scheme, which is labelled SA-NH3-CS2. It is important to note, however, that despite providing an improved representation compared to Strat-Trop, the key DMS oxidation pathways included in CS2 (primarily following those described by ) do not reflect the latest scientific understanding . Numerous recent studies, emerging from theoretical, laboratory experiments, and field observations, have provided a more comprehensive picture of DMS oxidation mechanisms . However, incorporating these more up-to-date reaction pathways into the model is beyond the scope of this study. The primary purpose of this simulation is therefore to evaluate how employing a different, more complex tropospheric chemistry scheme (CS2 vs. Strat-Trop) influences the simulated atmospheric sulfur cycle, the resultant aerosol number size distributions and composition.

The default UKESM1.1 configuration incorporates a simplified representation of SOA formation, based solely on the oxidation of monoterpenes. SOA formation from isoprene oxidation is not explicitly represented; instead, its contribution is implicitly considered by scaling the monoterpene oxidation SOA yield by a factor of two. However, recently implemented a new scheme in the UKCA model to explicitly represent isoprene SOA formation. This new scheme treats SOA formed from isoprene oxidation as a distinct aerosol component, produced with a fixed mass yield of 3 % from the reaction of isoprene with the major atmospheric oxidants (O3, NO3, and OH). Other aspects of the treatment of this isoprene SOA component are assumed to be identical to those of the SOA derived from monoterpene oxidation. The choice of the 3 % yield is based on the work of , which in turn relies on the experimental findings of . Concurrently with the introduction of this explicit isoprene SOA source, the scaling factor previously applied to the monoterpene SOA yield is removed. Therefore, the mass yield for SOA formation from monoterpene oxidation reverts to its base value of 13 % in simulations using this scheme (Table ). The simulation incorporating this explicit isoprene SOA scheme is referred to as SA-NH3-IPSOA. After formation, the condensable oxidation products of monoterpenes and isoprene are assumed to condense irreversibly onto all particle modes, including nucleation, Aitken, accumulation, and coarse.

It should be acknowledged that representing SOA formation using fixed yields is a significant simplification, as actual yields are known to be strongly influenced by factors such as temperature, oxidant concentrations, aerosol acidity and nitrogen oxide (NOx) levels. Furthermore, the formation pathways differ: monoterpene oxidation can produce extremely low-volatility compounds that contribute effectively to SOA mass through irreversible condensation , whereas isoprene SOA formation is thought to be dominated by the reactive uptake of gas-phase oxidation products, such as isoprene epoxydiols (IEPOX) . This reactive uptake process is significantly affected by factors like aerosol acidity; for instance, yields as high as 28.6 % have been observed under low-NOx conditions onto acidified sulfate seed aerosol . Additionally, many previous laboratory experiments on isoprene SOA yields were conducted at room temperatures. Given that lower temperatures reduce the volatility of organic vapours, the effective yield of isoprene SOA formation is expected to be higher under colder atmospheric conditions. This expectation is supported by the recent work of , which revealed that isoprene oxidation products can initiate both aerosol nucleation and subsequent growth at low temperatures, implying a significantly higher potential SOA yield under such conditions. Therefore, to explore the sensitivity to this uncertainty, we also perform a simulation where the mass yield for the explicit isoprene SOA formation scheme is increased tenfold (to 30 %). This simulation is referred to as SA-NH3–IPSOA×10.

The measured O3 and OH mixing ratios both exhibit a distinct inter-hemispheric asymmetry, with consistently higher values observed in the Northern Hemisphere (NH) compared to the Southern Hemisphere (SH). This asymmetry is well-understood to be driven by higher anthropogenic emissions of ozone precursors, such as nitrogen oxides (NOx) and hydrocarbons, in the NH . Regarding latitudinal distribution, observed O3 mixing ratios are generally lowest in the tropics and increase towards higher latitudes, a pattern consistent with recent compilations of oceanic and polar O3 data . In contrast, observed OH concentrations show approximately the opposite latitudinal trend to O3, with the highest values found in the tropics. This inverse latitudinal distribution is primarily attributed to the stronger solar radiation in the tropics which leads to more rapid photochemical destruction of O3, which in turn drives higher OH production rates.

Comparing the benchmark SA-NH3 simulation with the ATom measurements reveals that O3 is generally overestimated by the model in the tropics throughout the vertical profile, transitioning to a slight underestimation in the polar regions. Interestingly, despite the asymmetry in the absolute observed concentrations, the pattern of modelled O3 discrepancies relative to observations appears broadly symmetrical between the hemispheres. Quantitatively, the median overestimation of O3 in the tropics (25° S to 25° N, full altitude range unless otherwise specified) is 29 % (i.e. model-to-ATom ratio of 1.29), while the median underestimation in the polar regions (60–90° N/S) is 15 % (i.e. model-to-ATom ratio of 0.85). However, the model discrepancies for OH differ significantly from those for O3 and exhibit a strong inter-hemispheric asymmetry. North of approximately 50° S latitude, modelled OH concentrations are generally overestimated by around 8 %, whereas south of 50° S, modelled OH is substantially underestimated, by 54 %.

The reason for this systematic underestimation of OH concentrations south of 50° S by the model is currently unclear and warrants further investigation in future studies. However, some potential contributing factors can be explored using the data presented. The primary photochemical production pathway for OH involves the photolysis of O3 followed by the reaction of the resulting O(1D) atom with water vapour (H2O). Our evaluation of relative humidity (Fig. ) indicates a general model overestimation south of 50° S. Given that modelled O3 is only slightly underestimated in this region (by 19 %), inaccuracies in the modelled concentrations of the primary precursors (O3 and H2O) seem unlikely to be the main drivers of the substantial OH underestimation. However, uncertainties in the modelled photolysis rate of O3 (J(O1D)) represent one factor to be examined, as this directly impacts the OH production rate. Additionally, the primary loss processes for OH involve reactions with methane (CH4) and carbon monoxide (CO); therefore, the accuracy of their simulated concentrations is also important. Other factors potentially contributing to the discrepancy include inaccuracies in modelled NOx concentrations (which influence OH recycling) or other unexpected OH loss processes in the model.

As key precursors for atmospheric H2SO4, DMS and SO2 are also evaluated against ATom observations in this study (Fig. ). The measured DMS mixing ratios during ATom are generally highest in the MBL, with concentrations decreasing with increasing altitude. Within the MBL, measured DMS mixing ratios are relatively higher in the tropics compared to mid-latitude and polar regions, and the distribution appears essentially symmetrical between the hemispheres. Median DMS mixing ratios measured by ATom were around 5.6 pptv below 2 km altitude and 0.6 pptv above 2 km. In contrast to DMS, observed SO2 mixing ratios in ATom4 exhibit a more homogeneous distribution throughout the marine atmosphere, with an overall median value of 11.9 pptv.

Comparing modelled DMS with observations reveals that simulations utilising the interactive DMS emission scheme consistently overestimate observed mixing ratios by roughly a factor of 19.32 south of 40° S across all altitudes (SA-NH3 simulation). The SA-NH3-Lana simulation, which employs a climatological DMS emission scheme, shows marginally better agreement in this high southern latitude region, although it still overestimates observed DMS by a factor of 11.24. Conversely, considering the global MBL, the median modelled overestimation factor for DMS is lower in the interactive scheme simulation (SA-NH3, factor of 2.58) compared to the climatology simulation (SA-NH3-Lana, factor of 3.28) in Table . Therefore, judged by median performance, the interactive DMS emission scheme appears to perform better within the MBL globally but significantly worsens the model performance at high southern latitudes compared to the climatology. Similarly, modelled SO2 mixing ratios south of 40° S (full altitude range) are also overestimated relative to ATom4 data, by 33 % in the SA-NH3 simulation and by a lesser amount, around 20 %, in the SA-NH3-Lana simulation.

While switching the DMS emission scheme primarily impacts modelled DMS and SO2 concentrations in a latitude-dependent manner (particularly at high southern latitudes), changing the core tropospheric chemistry scheme from Strat-Trop to CS2 exerts a global influence on simulated SO2 mixing ratios. Simulated SO2 mixing ratios are consistently lower in the SA-NH3-CS2 simulation compared to simulations employing the Strat-Trop scheme (Fig.  and Tables , , ). Consequently, the SO2 simulation in SA-NH3-CS2 shows reasonable agreement with observations (slight overestimation) between approximately 40° S and 60° N, but tends towards underestimation outside this latitude range. This suggests that the modified DMS oxidation chemistry within the CS2 scheme significantly alters the simulated SO2 budget compared to Strat-Trop (Table ). However, it must be pointed out that this apparent improvement in the SO2 simulation with CS2 is not necessarily indicative of a better overall representation of sulfur chemistry, given that DMS remains significantly overestimated by the model in most regions. Therefore, the seemingly better SO2 agreement might arise from compensating errors within the model's sulfur cycle representation, an issue warranting further investigation.

As previously reported by , atmospheric NH3 mixing ratios estimated during ATom are generally low, often below 1 pptv (Fig. ). The highest NH3 mixing ratios were observed in the tropical MBL (25° S to 25° N; altitude < 2 km) with a median value of 6.6 pptv. Outside this tropical MBL region, observed NH3 mixing ratios frequently dropped below 1 pptv, at which NH3 would typically have a negligible impact on aerosol nucleation processes involving H2SO4.

In contrast to observations, modelled NH3 mixing ratios in the default SA-H2O simulation are globally overestimated by several orders of magnitude compared to observations (Fig. ). For example, the median NH3 mixing ratio in the tropical MBL is simulated to be 215.4 pptv in this configuration, 32.64 times higher than the observed median. In the SA-NH3-noNit simulation (which includes H2SO4–NH3 aerosol nucleation but not the ammonium nitrate scheme), the median tropical MBL NH3 mixing ratio remains high at 212.0 pptv. However, after implementing the ammonium nitrate scheme developed by (which enables the uptake of NH3 by acidic aerosols), the modelled median NH3 mixing ratios in the tropical MBL are significantly reduced, to 28.6 and 25.1 pptv in the SA-NH3-slow and SA-NH3 benchmark simulations, respectively. These values are considerably closer to the observed median. This result strongly suggests that the uptake of NH3 by acidic aerosols to form ammonium is a significant sink for gaseous NH3 in the remote marine atmosphere, and that simulated NH3 concentrations are highly sensitive to the representation of this process.

It should be noted, however, that even in the SA-NH3 benchmark simulation, global atmospheric NH3 is still overestimated by a factor of 12.12 when compared to ATom data in the MBL (Table ). Using the slow nitrate uptake coefficient, the SA-NH3-slow scheme does not improve the ammonia simulation, with the model-to-ATom NH3 ratio remaining at 11.78. A similar trend of atmospheric NH3 overestimation is consistently observed in other chemical transport models as well . This persistent, large overestimation of gaseous NH3 inevitably drives an even larger overestimation in the formation rate of new aerosol particles via the H2SO4–NH3 nucleation pathway, as will be discussed later. Therefore, further constraining modelled NH3 mixing ratios, likely through improved emission inventories or more detailed representation of its uptake and loss processes, remains a critical area for model development to improve aerosol simulations of NH3 in the remote marine atmosphere .

The ATom measurements reveal a ubiquitous presence of nucleation mode aerosols throughout the marine atmosphere . The concentrations of nucleation mode aerosols are largely symmetrical between the hemispheres. Another notable feature of the observed nucleation mode aerosol distribution is that its vertical variation is considerably larger in the tropics than in mid-latitudes and polar regions (Fig. ). For example, nucleation mode aerosol concentrations measured at approximately 12 km altitude in the tropics are around two orders of magnitude higher than those at the marine surface level, whereas this vertical gradient in mid-latitudes and polar regions is typically less than one order of magnitude. This phenomenon has previously been associated with strong NPF events occurring in the tropical UT, as proposed by and . Similar to the nucleation mode, Aitken mode aerosols also exhibit a comparable vertical trend in the tropics, with enhanced concentrations observed in the UT relative to the lower troposphere.

Comparing the default simulation (SA-H2O) with ATom observations (Figs.  and ), the bias in nucleation mode aerosol concentrations shows a clear dipole pattern: a significant overestimation globally above approximately 4 km, and a strong underestimation below this altitude. For example, the default simulation overestimates nucleation mode aerosols by roughly one order of magnitude at 12 km altitude (Fig. ), while it underestimates them with a ratio of 0.09 in the MBL (Table ). The overestimation of the nucleation mode in the UT by the default simulation is attributed to several factors. The primary factor is that the theoretical binary H2SO4–H2O nucleation mechanism from itself overestimates the neutral H2SO4–H2O nucleation rate by up to three orders of magnitude in the UT, as highlighted by . Other secondary factors include the model's underestimation of temperature, overestimation of SO2 and RH (discussed in previous sections), all of which contribute to an increased nucleation rate. Taking the tropical UT (8–12 km, 25° S to 25° N) as an example, the model-to-ATom ratios of SO2 and RH are 1.46 and 1.12, respectively, in the default simulation. The only evaluated factor that might partly counter this tendency to overestimation is the condensation sink, which the model overestimates by 11 % in the tropical UT; however, this effect is outweighed by the other contributing factors.

Implementing the H2SO4–NH3 nucleation scheme without the ammonium nitrate scheme (SA-NH3-noNit simulation) does not improve model skill in simulating aerosol number size distribution in the tropical UT; the overestimation of nucleation mode aerosols remains high, similar to that in the default SA-H2O simulation (Fig. ). On the other hand, the SA-NH3-noNit simulation shows nucleation mode aerosol concentrations more than 7 times higher in the MBL compared to the default simulation, due to the aerosol nucleation enhancement from H2SO4–NH3 nucleation mechanism (Table ). However, despite this enhancement, the simulated nucleation mode aerosol concentrations are still lower than the ATom observations in the MBL. Furthermore, it must be noted that this apparent improvement in predicting MBL nucleation mode aerosols in the SA-NH3-noNit simulation is concurrent with a enormous overprediction of NH3 by the model, which is higher than ATom observations by a factor of 329.52 (Table ). After implementing the ammonium nitrate scheme, with either a slow (SA-NH3-slow) or fast (SA-NH3 benchmark) nitric acid uptake coefficient, the overestimation of NH3 is reduced to factors of 11.78 and 12.12, respectively. However, the underestimation of nucleation mode aerosols worsens, with ratios of 0.19 and 0.15 for these simulations, respectively. This represents a marginal enhancement compared with the default H2SO4–H2O nucleation mechanism over pristine marine environments.

In general, the biases in modelled Aitken mode aerosol concentrations show similar patterns to those for the nucleation mode. In the MBL, Aitken mode aerosols are underestimated by a factor of approximately two for most model simulations except the SA-NH3-noNit and SA-NH3-CS2 simulations (Table ). The median Aitken mode aerosol concentration is well simulated in the SA-NH3-noNit simulation. However, as discussed previously, this improved agreement is associated with the significant overestimation of NH3 in this particular model configuration. Despite this major caveat, the SA-NH3-noNit simulation is the only simulation that substantially improves the model skill in predicting both nucleation mode and Aitken mode aerosol concentrations in the MBL, with model-to-ATom ratios of 0.68 and 1.03, respectively. The aerosol probability density function in the MBL for this simulation also shows much closer agreement with the ATom observations than other simulations (Fig. ). This result – where better aerosol predictions require unrealistically high precursor concentrations – indicates that the H2SO4–NH3 mechanism cannot adequately explain observed MBL airborne aerosol formation. Additional or alternative aerosol formation processes are necessary to explain the high nucleation mode aerosol concentrations observed in the MBL.

The observed global distributions of accumulation mode and coarse mode aerosols differ significantly from those of the nucleation and Aitken modes. Generally, regions with high Aitken mode aerosol concentrations are typically associated with low concentrations of accumulation mode aerosols, suggesting a somewhat inverse relationship in their spatial distributions (Figs.  and ). This relationship is partly regulated by the amount of condensable vapours available for growth, through which Aitken mode aerosols grow into the accumulation mode. Coarse mode aerosols exhibit significantly higher concentrations in the lower troposphere (below approximately 4 km; Fig. ). This is attributed to their larger size, which limits their atmospheric mobility and makes them more susceptible to deposition processes; thus they are concentrated nearer to their emission sources. Overall, both accumulation and coarse mode aerosol concentrations are significantly higher in the mid- to high-latitudes of the NH compared to the SH due to greater anthropogenic activities in the NH.

Accumulation mode aerosols are notably the only mode consistently overestimated by the presented model simulations in the MBL, with a model-to-ATom ratio of 1.25 in the SA-NH3 benchmark simulation (Table ). There are several potential reasons for this overestimation: (1) excessive aerosol nucleation and subsequent growth processes in the MBL, (2) excessive primary emissions of accumulation mode aerosols, or (3) inefficient aerosol removal processes. The first reason is unlikely, as both nucleation mode and Aitken mode aerosols are significantly underestimated in the MBL, as discussed previously. The second reason is likely, as the model appears to overestimate primary emissions contributing to accumulation mode aerosols, such as seasalt, in the MBL (supported by aerosol composition data discussed in the next section; see Table ). The third reason cannot be ruled out, but it cannot be assessed with the data available in this study.

While accumulation mode aerosols are overestimated in the MBL, they are underestimated in the UT, where particle growth is strongly constrained by a limited supply of condensable vapours (Fig. ). Since H2SO4 is likely already overestimated in the tropical UT by the model (due to overestimation of SO2 and OH), the condensable vapours driving this additional growth are likely to be oxidised organic compounds, such as isoprene-derived SOA, transported from rainforests to the marine UT, as suggested by and . This is supported by the SA-NH3-IPSOA×10 simulation which leads to a significant increase in modelled accumulation mode aerosol concentrations in this region, resulting in a model-to-ATom ratio of 0.65. This represents an increase of 0.19 in the ratio compared to the benchmark simulation (which has a ratio of 0.46). Simultaneously, the overestimation of Aitken mode aerosols in the tropical UT is also reduced by these additional condensable vapours, with the SA-NH3-IPSOA×10 simulation yielding a model-to-ATom ratio of 1.35. This is a decrease of 0.19 in the ratio from the benchmark simulation (which has a ratio of 1.54).

Coarse mode aerosols are generally underestimated by the model in the MBL, with a median model-to-ATom ratio of 0.71 in the SA-NH3 benchmark simulation (Table ). In regions other than the MBL, coarse mode aerosols are generally overestimated by the model (Fig. ). The most severe overestimation occurs south of 40° S above the MBL, where the model overestimation can exceed a factor of 10, although the absolute number concentrations in this region are very low (less than 0.1 cm-3).

The combined surface area of accumulation and coarse mode aerosols, and to a lesser extent Aitken mode aerosols, determines the available surface for vapour condensation, and thus defines the condensation sink (CS). The CS is calculated here following the methodology of , based on the dry aerosol size distribution. The CS typically exhibits a negative correlation with the aerosol nucleation rate, as a higher CS indicates a larger aerosol surface area for vapour condensation, thereby reducing the vapour concentration available for aerosol nucleation. In a large fraction of atmosphere sampled by ATom, the CS is overestimated by the model (Fig. ). In the MBL, the benchmark simulation overestimates CS by 27 % (Table ). Considering that the primary production pathway for H2SO4 in the MBL is the reaction of SO2 with OH, while its major loss process is condensation onto existing aerosols (governed by CS), a simplified steady-state approximation suggests [H2SO4]∝([SO2][OH])/CS. Given that the model overestimation of CS in the MBL is roughly counterbalanced by the overestimation of OH (by 31 %) in the benchmark simulation (Table ), the model-to-ATom ratio for the term [OH]/CS is close to unity. Consequently, the bias in simulated H2SO4 in the MBL is likely dominated by the bias in SO2, suggesting that H2SO4 is probably overestimated by a factor of 2 (Table ).

Except in very specific regions (e.g. the UT south of 40° S and MBL), sulfate aerosol mass is consistently overestimated by the model, with the magnitude of this overestimation typically being a few tens of percent across much of the sampled atmosphere (Figs.  and ). For example, between 2–8 km, the default (SA-H2O) and benchmark (SA-NH3) simulations overestimate sulfate by 42 % and 34 %, respectively (Table ). The only model simulations that significantly change sulfate aerosol mass concentrations, between 2–8 km, are SA-NH3-Lana, SA-NH3-CS2 and SA-NH3-MSA. By modifying the DMS emission scheme, the SA-NH3-Lana simulation increases the sulfate overestimation by 11 %, resulting in a model-to-observation ratio of 1.45. This increase is likely driven by higher DMS emissions from the adopted climatological dataset, relative to the default interactive scheme, as discussed earlier. On the other hand, the SA-NH3-CS2 simulation reduces the sulfate overestimation from 34 % (in the benchmark) to 5 %, while the SA-NH3-MSA simulation increases the overestimation to 45 % (Table ). The reduction in sulfate overestimation in the SA-NH3-CS2 simulation is attributed to the lower SO2 concentrations resulting from the CS2 chemistry scheme (see earlier discussions). The additional 11 % increase in bias in the SA-NH3-MSA simulation is due to the inclusion of condensable MSA vapours, which are treated as part of the sulfate aerosol component in the model for mass accounting, as MSA is not independently treated in the model.

It is important to mention that the most significant changes in sulfate aerosol mass concentrations relative to ATom, due to these scheme modifications, appear to concentrate in relatively pristine environments, such as the tropical UT and the SH (Fig. ). This supports that these changes are primarily driven by changes to the atmospheric chemistry of DMS (in SA-NH3-CS2) and the inclusion of its oxidation products like MSA as condensable species (in SA-NH3-MSA). Anthropogenic primary sulfate aerosol emissions and direct emissions of SO2, which are more dominant in the NH, are not significantly affected by these specific scheme changes; therefore, sulfate aerosol mass concentrations in the NH are less impacted. This highlights the need to improve the representation of DMS sources and its atmospheric chemistry in the model for simulating aerosols in the pristine regions of the atmosphere.

One primary reason for this general global overestimation of sulfate aerosol is the model's overestimation of SO2 concentrations, a key precursor for sulfate aerosol formation. Sulfate aerosol in the model is formed either from primary sources from anthropogenic emissions, or secondary processes such as from the gas-phase condensation of H2SO4, or, more significantly, through aqueous-phase oxidation of SO2 in cloud droplets by H2O2 and O3 . Consequently, the accuracy of simulated sulfate aerosol concentrations is also sensitive to the modelled concentrations of other species involved in these pathways, such as gas-phase OH and aqueous-phase oxidants like H2O2 and O3. Furthermore, the use of a fixed cloud pH value of 5.0 in the model may also affect the efficiency of SO2 uptake . Since the model's treatment of these multi-phase processes is relatively simplified, further mechanistic implementation and evaluation of these processes are needed to improve the model's representation of sulfate aerosol.

Another potential contributor to the overestimation of sulfate is the underestimation of its removal processes, leading to an overestimated aerosol lifetime. For instance, reported a sulfate (SO42-) lifetime of 5.57 d in UKESM1.0, which is around 2 d longer than the estimate of 3.7 d by . This value also lies at the upper end of the range reported by AeroCom-I models, which spans from 3 to 5.4 d, suggesting that sulfate removal may be too slow in UKESM.

The model-ATom discrepancy in organic aerosol mass exhibits a distinct dipole pattern: underestimation in the upper troposphere (8–12 km) and the tropical lower to mid free troposphere (2–8 km), and moderate overestimation across the rest of the sampled atmosphere (Figs.  and ). In the UT, this underestimation is most likely attributed to a lack of significant organic aerosol sources in the model at these altitudes. A clear missing source of organic aerosol in the model's UT is SOA derived from the oxidation products of isoprene, which are not explicitly included as a significant SOA source in the default model configuration. Recent ambient observations and laboratory studies have shown that isoprene oxidation products can contribute significantly to new particle formation and growth in the tropical UT .

Although a comprehensive isoprene new particle formation mechanism cannot be implemented in the model at this stage due to a lack of readily available parameterisations, this study attempts to partially account for this missing contribution of isoprene oxidation products to aerosol growth using the SA-NH3-IPSOA and SA-NH3-IPSOA×10 simulations. The implementation of the explicit IPSOA scheme significantly increases modelled organic aerosol mass concentrations in the tropical UT. The median measured organic aerosol mass concentration during the ATom campaigns in this region is 0.083 µgm-3. The model-to-ATom ratio in the benchmark SA-NH3 simulation for this region is 0.32, indicating an underestimation by around 3 times. With a 3 % SOA mass yield from isoprene oxidation, the SA-NH3-IPSOA simulation improves the model-to-ATom ratio for organic aerosol mass concentration in the tropical UT to 0.41. A 30 % SOA mass yield (SA-NH3-IPSOA×10 simulation) further improves this comparison, increasing the model-to-ATom ratio to 0.86 (Fig. ).

In the MBL, the model simulations generally overestimate organic aerosol mass concentrations, for example by a factor of 1.42 in the benchmark SA-NH3 simulation. The SA-NH3-IPSOA simulation results in a model-to-ATom ratio of 1.35 for MBL organic aerosol. This slight reduction in the MBL organic aerosol overestimation, despite adding an isoprene SOA source, is likely due to the concurrent reduction in the prescribed monoterpene SOA yield (see Table  and Methods). The SA-NH3-IPSOA×10 simulation, on the other hand, significantly worsens the organic aerosol mass overestimation in the MBL, increasing the model-to-ATom ratio to 1.72. This increased overestimation likely results from an isoprene SOA yield (30 %) that is unrealistically high for the warmer MBL conditions, where higher temperatures generally lead to lower SOA yields. Therefore, accurately quantifying the SOA yield from isoprene oxidation, both in the cold UT and warm MBL, is crucial for improving the model's representation of organic aerosols.

Another notable feature is the generally stronger underestimation of submicron organic aerosol mass in the NH compared to the SH. This disparity is likely due to the model's omission of secondary organic aerosol contributions from anthropogenic precursors, a limitation that warrants further investigation.

The ammonium and nitrate aerosol scheme, recently developed by , is not included in the default UKESM1.1 configuration. Therefore, the default (SA-H2O) and SA-NH3-noNit simulations do not include these two aerosol components, while the ammonium nitrate scheme is included in all other simulations discussed in this study. ATom observations indicate that the mass concentrations of both ammonium and (inorganic) nitrate are significantly lower than those of sulfate and organic matter globally. This is particularly true for nitrate mass, which exhibits low concentrations globally (Fig. ).

For the region where the nitrate component is above the detection limit of AMS, model simulations generally show an overestimation of nitrate aerosol mass concentrations compared to ATom observations below 8 km (Tables  and ). For example, the benchmark SA-NH3 simulation reports a model-to-ATom ratio of 11.27 in the MBL, indicating a significant overestimation by the model (Table ). Using the slow nitric acid uptake coefficient (SA-NH3-slow simulation) reduces this overestimation in the MBL, with the ratio reduced to 2.93. This discrepancy might suggest that the nitric acid uptake coefficient is still too fast, and using an even slower value could improve the simulation of nitrate aerosol mass concentrations in this NH. Alternatively, the overprediction may be from overestimated anthropogenic NOx emissions, which would lead to excessive atmospheric nitric acid and, consequently, particulate nitrate. A third possible explanation is that the model underestimates the deposition rates of NOx and its oxidation products. Therefore, the exact cause of the nitrate aerosol overestimation in the NH remains uncertain and requires further investigation.

In contrast to nitrate, ammonium aerosol mass concentrations are ubiquitously overestimated by the model throughout the sampled atmosphere when the ammonium nitrate scheme is active (Figs.  and ). This overestimation of ammonium likely translates into an overestimation of aerosol pH (i.e. an underestimation of aerosol acidity) globally, consistent with findings for other chemical transport models reported by . In the MBL, the benchmark SA-NH3 simulation overestimates ammonium aerosol mass concentrations by a factor of 3.20 compared to ATom observations (Table ). Furthermore, this positive model bias for ammonium is even larger above the MBL. For example, the bias increases to a factor of 7.73 between 2–8 km altitude. It is worth noting that both ammonium aerosol and gas-phase NH3 concentrations are significantly overestimated by the model in the MBL (Table ). This pervasive overestimation of ammonium is unlikely to be solely explained by flaws in the implemented ammonium nitrate scheme itself. Improvements to either or both the NH3 emission inventories and the representation of its atmospheric removal processes are needed to enhance the simulation of both gaseous NH3 and particulate ammonium in the model.

Seasalt aerosols are a major component of primary marine CCN and play a crucial role in cloud microphysics and climate. The precise relative contributions of seasalt aerosols (primary) and new particle formation processes to the total CCN population remain uncertain. These two major aerosol sources have distinct feedback mechanisms within the climate system and are projected to exhibit different responses to future environmental changes. It is therefore essential to accurately represent both processes in models to enable robust assessments of future changes in CCN, cloud microphysics, and ultimately, climate.

As expected, the measured seasalt concentrates in the MBL . However, the simulations without the ammonium nitrate scheme (SA-H2O and SA-NH3-noNit) overpredict seasalt aerosol mass concentrations in submicron aerosols within the MBL, with the default simulation predicting concentrations 2.09 times higher than observed (Table ). This overestimation persists consistently across latitudes, from the tropics to high-latitude regions (Fig. ). Additionally, simulations that include the ammonium nitrate scheme show a reduction in seasalt aerosol mass. This decrease is possibly linked to chemical processing, specifically the displacement of chloride by nitrate through heterogeneous uptake of HNO3 on seasalt particles, leading to the formation of hydrogen chloride (HCl) . However, given the very low nitrate concentrations in the marine boundary layer, it remains uncertain whether sufficient nitrate is available to drive appreciable chloride displacement. Further investigation is needed to better understand the implications of the ammonium nitrate scheme on seasalt conservation in marine environments.

Constraining seasalt emissions, particularly with geometric diameter smaller than 1 µm, is crucial for accurately understanding the CCN budget in the MBL, as Aitken and accumulation mode aerosols are the major contributors to CCN concentrations. The default model's overestimation of seasalt aerosols in the MBL will likely lead to an exaggerated contribution from primary seasalt to the CCN budget, potentially masking or underestimating the CCN contribution from airborne aerosol formation processes. The default model's overprediction of seasalt aerosols in the marine atmosphere is also supported by a recent study by , which found that the default seasalt emission scheme in UKESM overpredicts the dependence of seasalt emissions on wind speed when validated against aerosol optical depth observations.