The Arctic is a key component of the global climate system. Over the past 5 decades, the mean surface temperature of the Arctic has increased 3–4 times faster than the global average . Regional changes in the Arctic can have global impacts, such as climate feedbacks from albedo changes due to loss of sea ice, land ice, and snow ; changes in carbon sources due to increased wildfires and melting permafrost ; and global sea level rise from melting of the Greenland ice sheet .

Clouds are a major control on the surface energy balance in the Arctic. Due to the low solar insolation and high albedo of sea ice in the high Arctic (i.e. pack ice regions north of 80∘), the shortwave cooling effect of clouds is less important than at lower latitudes. Instead, longwave effects dominate such that the net radiative effect of low-level Arctic clouds is surface warming for all but a few weeks in the middle of summer . Low clouds are common and can persist in the mixed phase for several days . Aerosol particles act as cloud condensation nuclei (CCN) and ice-nucleating particles (INPs) and therefore influence the behaviour of clouds. Observational Arctic case studies have shown that perturbations in aerosol concentrations can change the radiative effect of clouds . Modelling studies reproduce this behaviour, while also highlighting the difficulty in creating models that can accurately simulate the complex behaviour of Arctic clouds .

The aerosol budget of the high Arctic is a balance of different processes, including primary emissions, new particle formation, condensational growth from vapours, long-range transport of anthropogenic emissions, and wet scavenging . Transport of aerosol from lower latitudes is more efficient during the winter and spring because it is thermodynamically easier in winter for air to enter the Arctic region from the south than it is in summer . Cold air masses sitting over ice-covered portions of the Arctic Ocean create sharp north–south temperature gradients, described as the polar dome, that act as a barrier for air moving towards the Arctic lower troposphere from the south. In the winter, several factors make it easier for air from the south to penetrate the polar dome, including a more southward extent of the dome, and cooling of air during transport due to proximity to snow and ice surfaces over land. Also, the removal of aerosol by frozen precipitation in winter is less effective than removal by drizzle in summer . These seasonal cycles of aerosol transport and removal efficiency result in a transition from Arctic haze in the spring to more pristine conditions in the summer, with fewer accumulation mode particles; less anthropogenic influence; and more, smaller nucleation and Aitken mode particles. Such a transition is evident in measurements, e.g. from Svalbard , from pan-Arctic observatories , or from the recent year-long MOSAiC campaign in the pack ice region .

Unlike the winter, when long-range transport is the dominant source of Arctic aerosols, summertime particles are thought to be strongly controlled by new particle formation (NPF) and growth from precursor vapours, which takes place in many Arctic locations. However, the large variety of vapours that can play a role, as well as the strong seasonal variation in key processes, have made it difficult to understand the main drivers of Arctic NPF. Observations of Arctic NPF are discussed in and will be briefly summarised here. Sulfuric acid (H2SO4) and methanesulfonic acid (MSA) have been shown to drive NPF and growth in regions close to or influenced by open water such as Svalbard and the Canadian Arctic Archipelago . The marine biogenic precursor dimethylsulfide (DMS) is the main source of H2SO4 and MSA to these regions. Open water is also a source of organic vapours which are observed to contribute to condensational growth of small particles . Ammonia has biogenic sources in the Arctic, for example seabird colonies, and has been observed to contribute to NPF events in Svalbard and Greenland . Iodine is known to be emitted from kelp in coastal areas and iodine-containing compounds have been observed to drive NPF events near the coast of Greenland . Of relevance to this study, recently observed NPF driven by iodic acid (HIO3) in the pack ice region of the Arctic Ocean during the sea ice freeze-up, suggesting an iodine source from snow, ice, or ocean water. Recent laboratory results elucidate the chemical pathway for the creation of iodic acid from iodine .

Modelling studies have demonstrated the importance of NPF for the budget of Arctic aerosol. and used the global aerosol model GLOMAP to show the importance of NPF on a global scale. Both studies indicate that a high fraction (greater than 80 %) of particles and CCN in the Arctic are derived from new particle formation. While the model configuration in only included parameterisations based on H2SO4, the model used by included parameterisations for neutral and ion-induced binary (H2SO4–water) and ternary (H2SO4–ammonia–water) NPF, NPF from organic molecules and H2SO4, and pure organic NPF driven by highly oxygenated molecules (HOMs). The authors found that a significant fraction (greater than 40 %) of Arctic CCN originate from secondary organic aerosol, including HOMs. and , both using GLOMAP to investigate Arctic aerosol, found that boundary layer NPF driven by H2SO4 was required to explain measured size distributions or CCN concentrations in the high Arctic during summer. used an aerosol dynamics model to study NPF events that were observed during the summers of 1996, 2001, and 2008. H2SO4 and organic vapours were used in the model to drive NPF events. Simulations of NPF driven by H2SO4 followed by growth from condensation of organic vapours were able to reproduce the particle size distributions observed during NPF events. The inclusion of organic vapours as a driver of NPF led to an overprediction in the concentration of particles, though the authors note significant uncertainty in the sources and concentrations of organic vapours in the high Arctic. used a chemical transport and aerosol microphysics model to show that condensation of secondary organic vapour played a key role in particle formation events in the Canadian Arctic during the summer. The vapours were assumed to have a marine, biogenic source.

Boundary layer NPF as a source of Arctic aerosol implies a potentially high sensitivity of the local aerosol budget to the changing climate. Any future increase in the extent of open water and the marginal ice zone in summer will affect the emission of aerosol and precursor gases to the atmosphere. have found a correlation between frequency of NPF events and the extent of open water near Svalbard and Greenland, respectively. Such an increase in occurrence of NPF with sea ice loss could be expected to increase CCN concentrations in the Arctic under future warmer conditions, though this is far from certain since changing sea ice extent is not the only controlling factor. found that reduction of sea ice in the year 2050 leads to increased emissions of DMS in a global aerosol–climate model. The increased DMS emissions, along with increased sea spray aerosol and meteorological changes, cause higher cloud drop number concentrations over the Arctic Ocean. This is in line with results from another global model study, , which used an atmospheric climate model to investigate the response of sea spray aerosol to sea ice loss. They found a strong increase in sea salt emissions and thus higher cloud drop concentrations, but the effect on clouds and energy budget was uncertain due to poor model representations of aerosol–cloud interactions. In contrast to these studies, the results from suggest that interactions between aerosol particles of different sizes could lead to a suppression of NPF in an ice-free Arctic summer. Their results showed an increase in the sink of condensable vapours due to stronger emission of sea spray aerosol from the open water, resulting in a decrease in the concentration of smaller particles from NPF. In addition, the growth of sea spray particles from the condensation of vapours shifted the size distribution to larger sizes, leading to enhanced scavenging by precipitation and a decrease in drop concentrations.

The conflicting results from , , and highlight the difficulty in modelling the Arctic aerosol budget and how it might change in future. Uncertainties in model parameterisations stem from the knowledge gaps in processes controlling the aerosol budget and cause differences in climate projections from different model set-ups. For example, the sea spray parameterisations used in the and studies include empirical representations of the effect of temperature on the sea spray aerosol size distribution, while the parameterisation used in the study does not. This could be a cause of the discrepancy in their predictions of sea spray aerosol response to sea ice loss, though this has not been studied. The studies also differ in their treatment of primary marine organic emissions, with only the study including such a source. Some field studies from the high Arctic pack ice region have indicated the importance of primary marine organics in the region , as well as raising questions about possible recycling mechanisms of particles after they are emitted i.e. through ageing or the particle break-up theory,. The open questions surrounding primary marine emissions complicate the modelling of Arctic NPF due to the sink of condensable vapours from larger particles. Models with size-resolved aerosol microphysics and chemistry are better equipped to study these questions than bulk, single-moment models considering total mass only.

Aitken mode particles can act as CCN in the Arctic, which increases the influence of NPF over the Arctic aerosol budget. used measurements of cloud residual particles (i.e. particles obtained by drying cloud droplets or ice crystals) from the high Arctic to show that Aitken mode particles were acting as CCN during a period of frequent boundary layer new particle formation. Observations of aerosol size distributions in and out of cloud in sub-Arctic Finland showed that on average 30 % of Aitken particles (defined as 25–95 nm) were activated to form droplets . In Svalbard, measurements of cloud residuals and cloud drop number concentrations have been used to show the activation of particles smaller than 50 nm diameter during periods of high supersaturation, such as when updraft speeds are high or when accumulation mode concentrations are low. Results from parcel models and large-eddy simulation (LES) models are in agreement with the observations that Aitken particles can act as CCN , though find that two sets of aerosol conditions under which Aitken activation is favourable in their model have real-world occurrence probabilities of 5 % and 17 %, raising questions about how widespread the phenomenon could be. Activation of Aitken particles means that particles formed by NPF and subsequent growth may only need to grow up 20–50 nm diameter to act as CCN in Arctic clouds, making it more plausible that such particles could survive long enough in the atmosphere to be important for cloud formation.

A further complicating factor that affects our understanding of summertime Arctic aerosol processes is the decoupling of the surface, where most aerosol measurements have been made in the high Arctic, and the cloud layer, where the aerosols act as CCN. Such decoupling is caused by the thermodynamic structure of the high Arctic summer boundary layer. Regimes of turbulence at the surface and in the cloud mixed layer have previously been shown to occur in configurations where the two regimes do not interact, inhibiting transport of moisture or particles vertically between the different layers . Decoupling of the surface in this way has implications for aerosol–cloud interactions because it implies that the aerosol sources, concentrations, and size distributions at the surface may only be relevant for the cloud layer during sporadic events of mixing. This is very different from dynamics typical of lower latitudes, where strong convection can promote mixing of heat, moisture, and aerosols from the surface up to higher altitudes. Models have struggled to capture decoupling in the Arctic, showing a tendency to become coupled too often .

Recent observations from the high Arctic provide a new opportunity to explore the questions surrounding the source of summertime aerosol. The Arctic Ocean 2018 (AO2018) expedition took place in August and September 2018 between Svalbard and the North Pole . The observed properties and behaviour of the aerosol challenge our current understanding of aerosol sources and sinks. Firstly, mass spectrometry measurements clearly show that iodic acid is the main driver of NPF events, which occur primarily during sea ice freeze-up . This process was not included in previous large-scale modelling studies. Secondly, the time series of particle concentrations and size distributions, spanning several weeks over the end of the sea ice melt period and transition to freeze-up, provide an opportunity to understand how sources and sinks are related to the melting and freezing cycle and changes in photochemistry. In particular, the observations show a distinct transition in aerosol behaviour in late summer, when the iodic acid NPF events begin to take place. Here we aim to interpret this change in behaviour in terms of changes in the dominant aerosol sources.

In this study we used measurements from AO2018 to evaluate the Arctic aerosol budget in the global climate model UKESM1. We compared the accuracy of simulations with NPF in the boundary layer to that of simulations with NPF in the free troposphere. We also introduced an iodic acid NPF scheme to the model to investigate iodic acid as an Arctic aerosol source compared to other components of the Arctic aerosol budget. The observational datasets are introduced in Sect. . The model is described in Sect. , including the different NPF schemes and our approach to the inclusion of iodic acid. Results are presented in Sect.  and discussed in Sect. .

We use data from three campaigns in different years: Arctic Ocean 2018 (AO2018), the Arctic Summer Cloud Ocean Study (ASCOS) in 2008, and the Atmospheric Tomography Mission (ATom) in 2016. We only use model output from a simulation of the year 2018 because the Arctic Ocean 2018 dataset is the main focus of this study. The interannual variability in aerosol concentration may limit how representative observations from 2008 (ASCOS) or 2016 (ATom 1) can be to assess model output from 2018. However, as we will show in later sections, the difference in particle concentration from different simulations in this study can be several orders of magnitude and therefore is likely to exceed the range of concentrations that would be measured in different years. Simulated values of N50 (number concentration of particles larger than 50 nm diameter) over a 30-year period suggest that the interannual variability of N50 in the Arctic does not account for such large differences in our simulated particle concentrations .

We used the UK Earth System Model version 1 UKESM1, in its atmosphere-only configuration, which uses output from a fully coupled run of UKESM1 to prescribe sea surface temperatures (SSTs) and some biogenic emissions, such as DMS, from the ocean. The dynamical core of the model is the UK Met Office Unified Model (UM) in global atmosphere configuration version 7.1 GA7.1, with a horizontal resolution of 1.875∘ longitude by 1.25∘ latitude and 85 vertical levels. The vertical resolution is approximately 50 m at the surface, 150 m at 1 km altitude, and on the order of 5 km at the highest model level. We ran the model in its “nudged” configuration, which means horizontal winds and potential temperature are relaxed to ERA-Interim values on a 6-hourly timescale above approximately 1 km. The model simulates gas and aerosol chemistry using the UK chemistry and aerosols model UKCA,. Aerosol microphysics is simulated by the Global Model of Aerosol Processes GLOMAP-Mode, using five log-normal aerosol size modes (four soluble and one insoluble). Mode sizes and geometric widths are given in Table . The model uses 2-moment aerosol microphysics, meaning that the number and mass in each mode are prognostic variables. There are four aerosol species: sulfate, organic carbon, black carbon, and sea salt. Processes handled by GLOMAP include primary emissions, coagulation within and between modes, condensational growth and ageing, new particle formation, dry deposition, wet deposition within and below clouds, and aqueous sulfate production in cloud droplets.

Emissions of SO2, black carbon (BC), and organic carbon (OC) are for the year 2014 for all emissions sectors except biomass burning, which are from a climatology of the years 1995–2004. We use a climatology for biomass burning to minimise any potential bias from boreal forest fire emissions not being from the same year as the simulation time. The emissions datasets used for aerosols and precursor vapours are (SO2, anthropogenic OC, and BC), (biomass burning OC and BC), and (monoterpenes). Sea salt emissions are calculated using the parameterisation. Primary marine organic emissions are then calculated using the sea spray flux, 10 m wind speed, and chlorophyll-a concentration, using the parameterisation. Note that because we use an atmosphere-only configuration in this study, the chlorophyll-a concentration is taken from an ancillary file, produced using model output from the fully coupled model. Marine emissions are scaled by gridbox open water fraction for sea ice regions.

We organise the results as follows. First we examine the behaviour of our empirical iodic acid scheme by comparing modelled surface iodic acid concentrations to observations in Sect. . Then in Sect.  we consider the effect of BL NPF in the model on the surface aerosol concentration during AO2018. Since we show that BL NPF is not taking place locally during the melt period of AO2018, we consider FT NPF as a source of surface particles in Sect. , using surface observations from AO2018 and aerosol profiles from ASCOS in 2008 and ATom in 2016. Finally, we consider a combination of BL and FT sources in Sect. .

When comparing model output to observations of surface aerosol concentration, we use the overlap index defined in to quantify the similarity between the distribution in aerosol concentration from observations and different model simulations. For two probability density functions A(x) and B(x), the overlap index η(A,B) is defined as 5η(A,B)=∫min⁡[A(x),B(x)]dx, which can be thought of as integrating the area where the distributions overlap. If the two distributions overlap completely, the entire area under the distribution will be integrated so the index will be 1, whereas if they do not overlap at all, no area will be integrated and the index will be 0. Overlap indices closer to 1 therefore indicate simulations with good model–observation agreement in terms of the magnitude and variability of the two time series even if individual peaks and troughs do not match temporally. In our case, we apply the overlap index to discrete distributions of binned aerosol concentration, so the integral becomes a sum.

We have used field observations and a global aerosol–climate model with an empirical iodic acid nucleation scheme to investigate sources of aerosol in the high Arctic summer. Our results point to a regime transition occurring in late summer from a free-tropospheric source of secondary particles to in situ new particle formation in the boundary layer, driven by iodic acid. The onset of iodic acid new particle formation (triggered by sea ice freeze-up) coincides with a decline in the free-tropospheric source rate brought on by declining rates of photochemical production of precursor vapours in the free troposphere. The net effect of the transition from free troposphere to boundary layer nucleation is a fairly constant nucleation and Aitken mode at the surface.

There are several key conclusions we can draw from the simulations we have presented here. They are listed below.

The default settings of the UKESM1 model cannot capture Arctic aerosol concentrations. The nucleation mode particle concentration at the surface is at least an order of magnitude too low in CONTROL compared to AO2018, ASCOS, and ATom observations, sometimes underestimating observed concentrations by as much as 5 orders of magnitude. The surface Aitken mode concentration is also underestimated by up to 3 orders of magnitude. The model performs better at simulating the surface accumulation mode concentration. The accumulation mode concentration in CONTROL is generally within the range of the observations but does not capture the lowest concentrations (less than 1 cm-3) seen in the time series of AO2018 observations. These periods of low accumulation mode concentration are important in controlling the behaviour and radiative effects of low-level Arctic clouds .

Our results have implications for the future Arctic aerosol budget. Iodic acid new particle formation in the Arctic boundary layer is strongly coupled to the surface and therefore sensitive to changes in the sea ice. However, our simulations show that entrainment of secondary particles from the free troposphere is also an important source of surface aerosol at nucleation and Aitken mode sizes. Free-tropospheric new particle formation is unlikely to have such a strong sensitivity to local sea ice changes, since the precursor vapours and background aerosol in the free troposphere are likely to have been transported from other regions. Further work will be required to understand how the balance of boundary layer versus free-tropospheric nucleation will evolve in the changing Arctic, since predictions of cloud behaviour and surface energy balance in the future Arctic depend on knowledge of the Arctic aerosol budget. Previous studies have shown that models struggle to simulate the thermodynamic structure of the boundary layer in the high Arctic, for example by failing to produce the decoupled conditions that are common . Such biases could inhibit accurate predictions of free-tropospheric aerosol sources, since the entrainment of particles in the boundary layer will rely on accurate representation of the turbulent mixing created by the structure of the cloud layer.

This study shows that it is important to assess the influence of boundary layer versus free-tropospheric sources compared to other uncertain Arctic aerosol processes and their different representations in models, for example sea spray parameterisations and primary marine sources. We did not consider primary sources for the Aitken mode during AO2018, such as biogenic marine particles. Although our model includes primary marine aerosol emissions, these are mostly limited to the accumulation mode. Thus, our model does not include direct marine emissions of smaller particles. Previous field studies in the Arctic pack ice region have found evidence that organics in Aitken and accumulation mode aerosols and cloud water were related to polymer gels found in the surface microlayer of sea water, suggesting a primary marine aerosol source , although our earlier modelling results suggest that accounting for the observed Aitken mode concentrations would require an unrealistically high surface source . Also proposed in the literature is an atmospheric processing pathway where larger primary particles break-up to form more, smaller particles. We cannot use our model results to exclude such a source from the aerosol size distributions measured during AO2018. The balance of primary and secondary aerosol sources in this region merits further work and would likely require improvements to existing model parameterisations (i.e. new particle formation and growth rates, primary marine emission fluxes) and the development of new parameterisations to test in the model (i.e. particle emissions from open leads independent of wind speed and size-resolved break-up rates for primary marine particles).

A greater understanding of aerosol conditions in the Arctic free troposphere is needed. It is not possible to distinguish from the model output presented here whether particles coming from the free troposphere have been created there or have first been transported from lower latitudes, where they were created in the boundary layer. Our results indicate the importance of obtaining measurements of aerosol size distributions and precursor vapours from above the boundary layer in the high Arctic, which have previously been sparse. Moreover, the H2SO4 and organically mediated new particle formation schemes we use produce very similar surface particle concentrations in the region of study, such that we cannot use these datasets to evaluate the accuracy of one over the other. This highlights an open question in modelling of Arctic aerosol. It is crucial to understand which precursor species are important for Arctic aerosol in order to understand how changes to different parts of the climate system will affect the formation and behaviour of Arctic clouds. The production of secondary organic vapour in UKESM1 is crude, accounting only for the oxidation of monoterpenes, and we have not included any effects of ammonia, which has been shown in laboratory and Arctic field studies to contribute to new particle formation. The inclusion of ammonia in the model could lead to more new particle formation in the boundary layer since ammonia can stabilise H2SO4 clusters. However, before these potential sources of model bias can be improved, more observational data about what drives new particle formation throughout the Arctic, both in the free troposphere and at the surface, are needed.

The accumulation mode in the model is relatively unaffected by the different new particle formation schemes we used and is better simulated than the smaller modes. This shows that accumulation mode aerosol and the contribution of the particles to CCN is not affected by new particle formation in these simulations. However, comparison with AO2018 observations shows that the model cannot capture periods of up to a day of very low accumulation mode concentration (approximately 1 cm-3 and below). Given that the accumulation mode affects new particle formation via the sink of condensable vapours, it will be important to consider this model bias in future work. Moreover, observations indicate the importance of fog as a sink of the iodic acid, so more work is needed to assess the role of fog in controlling the frequency of iodic acid new particle formation events. A higher resolution model with better cloud parameterisations would be better suited to such studies than the coarse, global model we have used here.