Clouds play a crucial role in the hydrological cycle and the radiative balance of the Earth–atmosphere system. However, clouds still comprise high uncertainties and their behavior under climate change scenarios is not yet well-understood. Hence, the magnitude of the cloud radiative forcing in the upcoming years remains unclear . Mixed-phase clouds (MPCs) contain both phases, i.e., ice and water, and are important for the radiative balance and climate sensitivity . MPCs occur in regions of deep convection, where the cloud top reaches temperatures low enough for ice formation , in mountainous terrain , or in cold regions of the planet, i.e., in high latitudes . In the Arctic, MPCs occur approximately 40 % of the time and are often observed as persistent low clouds . Their radiative forcing at the surface is still ambiguous and determined in part by the distinct seasonal cycle at high latitudes. In summer, the reflection of incoming shortwave (SW) radiation dominates, while during the rest of the year absorption and emission of longwave (LW) radiation prevails, causing a warming at the surface . In recent decades the Arctic has been warming at a faster rate than the rest of the globe . As changes in the Arctic can impact midlatitude weather conditions, the climate state of the Arctic is important not only regionally but also hemisphere-wide . Due to their strong radiative impact, MPCs can alter the Arctic climate system e.g.,, potentially accelerating or slowing the current high-latitude warming.

Arctic MPC fraction and phase partitioning are governed by a multitude of processes operating in conjunction across a wide range of spatial scales, such as the large-scale dynamical forcing, surface processes, and ambient aerosol concentration. The large-scale dynamical forcing determines air mass and hence water vapor advection, which is found to be crucial for the persistence of Arctic MPCs . With ongoing sea ice loss and the possibility of an ice-free Arctic by mid-century , the impact of surface conditions on Arctic MPCs has gained increasing attention in the past decade e.g.,. A more exposed open ocean surface has potential implications for cloud dynamics . , using the 40-year European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA-40) product, demonstrated that sea ice loss increased boundary layer height and led to more midlevel clouds. In addition, found increased stratocumulus or cumulus cloud formation over the ocean in contrast to thin stratus clouds over sea ice in observations from the Arctic Clouds in Summer Experiment (ACSE) campaign. These observed changes in cloud height were also observed during the Aerosol-Cloud Coupling And Climate Interactions in the Arctic (ACCACIA) campaign . In addition, the authors reported fewer and larger cloud droplets as well as increased precipitation rates over the open ocean compared to over sea ice. In large-eddy simulation (LES) experiments for the same case, could reproduce these observations and simulated cumuli tower development over a warming open ocean surface, in agreement with previous results of more convective cloud systems over a destabilized surface.

In addition to increased surface heat fluxes, aerosol emissions may increase in the Arctic , which could impact cloud microphysics. Since the Arctic is a pristine environment and aerosol concentrations are generally lower than in the lower and midlatitudes , any aerosol perturbations could substantially impact MPC formation and persistence. With decreasing sea ice, trans-Arctic shipping is also projected to increase, exerting local aerosol perturbations . An increased availability of cloud condensation nuclei (CCN) resulting from both sea salt and dimethyl sulfide emissions from the ocean and predicted ship emissions may lead to increased cloud formation and a net surface cooling during summer, as projected by global climate and Earth system models . Locally, aerosols released in ship tracks alone can change cloud liquid and ice water content (LWC and IWC, respectively) as found in the studies of and . Equivalently, a reduction in the ambient CCN and hence cloud droplet number concentration (Ndrop) could induce cloud dissipation . However, disentangling the competing effects of environmental conditions and aerosol disturbances appears challenging . In the past, argued for a buffered aerosol response in certain cloud regimes. For midlatitude convective clouds, showed that cloud fraction is not impacted by aerosol perturbations, but that aerosols may affect the organization of cloud pockets with fewer but larger cloud cells under levels of increased pollution. In simulations of trade wind shallow cumuli by an initial aerosol response is seen, with an increased number of cumulus structures and decreased precipitation. Yet the system efficiently returns to an organized cloud structure in a quasi-stationary state after some hours, which is insensitive to the background aerosol concentration. Turbulent mixing, entrainment, and detrainment of aerosols out of polluted regions could also potentially impact the aerosol concentration and the long-term aerosol response, as has been simulated by for ship tracks in the Monterey Bay. Conversely, found that entrainment of aerosols from the free troposphere into the boundary layer represents an important source of aerosol particles for Arctic MPCs as the authors showed in observations and LES experiments of the Arctic Summer Cloud Ocean Study (ASCOS) field campaign.

In this study, we investigate how the response to increased aerosol concentrations may differ for different cloud regimes of Arctic MPCs. For this purpose we perform high-resolution idealized LES experiments to resolve the multitude of boundary layer processes that impact the cloud state. We contrast our results for different surface conditions (open ocean surface versus sea ice) and apply different perturbations across a ±2 K temperature range. To validate our simulations we use observations obtained during the recent ACCACIA campaign in the European Arctic.

LES experiments are performed with the Consortium for Small-scale Modeling (COSMO) model in its configuration for idealized simulations (COSMO LES) . The COSMO LES has been proven to simulate MPCs in the Arctic with reasonable accuracy . Here, we simulate a single-layer stratocumulus case during the ACCACIA campaign on 23 March 2013. All simulations are initialized with the dropsonde profile number 5 released during the campaign . The obtained profiles are smoothed to exclude small-scale variability from the measurements as model input. In addition, the water vapor mixing ratio (qv) was increased by 20 % to account for the dry bias in dropsonde data . Note that in contrast to we initialize the open ocean as well as the sea ice simulations with the same atmospheric profile, to narrow down dynamic changes in the cloud-topped boundary layer to changed surface conditions alone (i.e., turbulent surface fluxes) and exclude any impact from varying large-scale conditions or boundary layer stability.

The domain covers a 19.2 km × 19.2 km large area centered around the location of the release of dropsonde number 5 (75∘ N, 24.5∘ E). The horizontal resolution is 120 m, the vertical resolution is variable and specified with 20 to 25 m within the entire boundary layer and coarser resolution above cloud top up to the model top at 23 km. The temporal resolution is 2 s and the model has been run for 20 h, including a 1.5 h spin-up period. Radiation is treated interactively according to the radiation scheme and includes a diurnal cycle. The cloud microphysical tendencies are parameterized following the two-moment scheme. The scheme considers five hydrometeor types (cloud droplets, rain drops, cloud ice, snow, and graupel) represented as gamma distributions with prescribed shape parameters and prognosed bulk mass and number concentrations. As in we use a prognostic treatment of ice-nucleating particles (INPs) while we keep the background CCN fixed, with cloud droplet activation calculated according to Köhler theory . The fixed background CCN ensure that sufficient CCN are available throughout the whole simulation for droplet activation. CCN are assumed to be pure ammonium bisulfate particles. Prognostic INPs are implemented as in . The scheme parameterizes immersion freezing following the temperature dependence and captures the depletion and replenishment of INPs. Following the COSMO setup for the model intercomparison performed by , ice crystals and snow flakes are assumed to be dendrites. As secondary ice processes are observationally poorly constrained, only the HP mechanism is included in our model, which is inefficient at cold temperatures (-15 to -20 ∘C).

We initialize the simulations with one background mode of potential CCN (0.2 µm mean diameter and 1.5 standard deviation), represented by a lognormal size distribution. For direct comparison to observations and the model study, the CCN concentrations were chosen to match the observed Ndrop over the ocean and the fixed Ndrop in , and were set to 100 cm-3. As we do not expect every CCN to activate, we initialized with a CCN concentration larger than the mean Ndrop measured over the ocean. The initialized CCN concentration is still within the spread of the measured Ndrop range though. INPs were initialized with a concentration of 3.3 L-1, which is at the high end of predicted ice crystal number concentrations (Nice) by different parameterizations in (assuming one INP per ice crystal). Due to the interactive INPs in our simulations, we used a relatively high initial INP concentration to prevent an underestimation of Nice. For simplicity we assumed a constant aerosol profile with height. As for the background thermodynamic conditions, we kept the background aerosol concentrations the same in the open ocean and sea ice cases.

We performed control simulations over sea ice and open ocean and evaluated these against available observations. For the sea ice case, the COSMO sea ice model was switched on. To exclude influences from variable turbulent fluxes, the sensible and latent heat fluxes were set to 25 and 23 Wm-2 over ocean and to 1 and 0.8 Wm-2 over sea ice. These prescribed fluxes are at the lower end of the observed range . However, larger fluxes were found to increase the strength and size of the convective cells in sensitivity simulations not shown here. Therefore, we would need larger domain sizes to simulate cases with larger surface fluxes. This was not possible due to the high computational demand of each simulation. Surface roughness length was assumed to be higher over the ocean with 0.0002 m in contrast to 0.0001 m over sea ice. Divergence was prescribed as zero at the surface and was relaxed linearly to 4×10-6 s-1 at the inversion height and kept constant above. To compensate for the subsidence heating, we included negative horizontal advective temperature tendencies, while all other tendencies were set to zero to prevent any influence of boundary layer moistening or drying by large-scale advection.

The local atmospheric conditions over open ocean as observed during the ACCACIA campaign (hereafter named observations) are characterized by a single temperature inversion at 1.3 km, capping a single-layer MPC between approximately 0.3 and 1.2 km . Our simulated case similarly features a strong inversion (Δθ=6 K) at a height of 1.4 km, capping a single cloud layer below (Fig. ). The boundary layer in both control simulations (named ocean_control and ice_control over open ocean and sea ice, respectively) is stably stratified, as seen in the positive gradient in the ice–liquid potential temperature (θil) and the negative gradient in the total water content (qt) in Fig. . Over the ocean surface an unstable surface layer forms due to the nonzero surface fluxes. The remainder of the boundary layer is stably stratified, which prevents the formation of a well-mixed boundary layer. As a result of stronger surface fluxes, the boundary layer retains more water vapor over the ocean compared to sea ice (Fig. a, b).

Our model successfully simulates a liquid-topped MPC with ice sedimenting out of the liquid layer in both control simulations, in agreement with observations. The observed cloud properties obtained from , our simulated values of the unperturbed simulations, and the LES results from are summarized for comparison in Table 2. From we only included the simulation using the ice parameterization that was fitted to the observations (termed ACC), which best reproduced the observed case .

The simulated mean Nice of 0.27 L-1 in ocean_control (Table 2) is slightly lower compared to observations, but within the observed range. Ndrop agrees well in our model simulations compared to observations, but the maximum Ndrop in Fig. a is simulated at a higher altitude (1.4 km instead of 1.0 km) due to the upward shift of the simulated stratiform cloud deck. The cloud droplet radius (Rdrop) is smaller than observed, due to an underestimation of the liquid water mixing ratio (LWMR) in the ocean_control simulation by a factor of 2. This underestimation of the liquid phase is a general issue in high-resolution simulations of mixed-phase clouds. In particular, the potential impact of the autoconversion rate on cloud evolution in a similar context has recently been discussed in .

The ice_control simulation can only be compared to observations in qualitative terms, as the initialization relies on the open ocean dropsonde profile (see Sect. 2). In the observations, the boundary layer over the sea ice was less well-mixed and colder and drier compared to the open ocean . As a result, the observed LWMR is smaller over sea ice than over the ocean, which is reproduced in our simulations. Our simulated Nice is also considerably lower over sea ice than over ocean. In contrast to observations, Rdrop is only 0.7 µm smaller in ice_control than in ocean_control, instead of 5 µm smaller in the observations. Additionally, Ndrop is smaller instead of larger in ice_control (Table 2 and Fig. b). We relate these differences in cloud properties between our simulated and the observed MPCs to the difference in the observed and simulated thermodynamic profiles: the drier boundary layer observed over sea ice suppresses cloud droplet growth. Moreover, the warmer and more turbulent boundary layer over the open ocean favors collision–coalescence of cloud droplets, leading to larger and fewer Ndrop over the open ocean. By choosing the same initial conditions for our open ocean and sea ice simulations, these processes are not equally represented.

The simulated effect of surface fluxes is illustrated in Fig. , showing a snapshot of the updraft velocities and LWP over ocean and sea ice after 3 h of simulation time. The different surface conditions lead to two different cloud regimes: over ocean, where surface fluxes are increased, the updrafts are higher, leading to cumulus towers detraining into the stratus deck and to a domain-wide shallow stratocumulus cloud structure. Within the shallow cumuli the LWP increases up to 300 gm-2, 4 times higher than in the surrounding stratus layer. In contrast, over sea ice the updrafts are low and a spatially homogeneous stratus forms. The LWP of the stratus cloud remains below 80 gm-2.

These dynamic differences feed back onto the vertical cloud structure (Fig. ). Supported by the stronger updrafts over the open ocean, the cloud base and top of the stratiform cloud deck are lifted by 200 and 100 m, respectively, compared to the cloud over sea ice (dashed lines in Fig. ). These high updrafts over the open ocean sustain an increased rate of cloud droplet activation. Despite this increased rate of cloud droplet activation, the mean effective radius remains unchanged between the two cloud regimes due to the increased rate of condensate forming in the updraft. The higher updrafts also facilitate rain formation over the open ocean, where droplets can grow at a faster rate than in the surrounding stratus cloud. As a result, total precipitation is increased over the open ocean (on average 1.12 mmd-1 as opposed to 0.59 mmd-1 above sea ice, Fig. S1a, b in the Supplement). Over sea ice, relatively low updraft speeds prohibit a strong upward moisture flux into the cloud layer due to the large thermodynamic stratification in the sub-cloud layer. This results in a drier boundary layer at cloud height and an optically thinner cloud (Table 3).

In addition to Ndrop, Nice is also increased in ocean_control compared to ice_control. As suggested by , the higher liquid water content in the air column increases the cloud LW emissivity. Thus, the higher LWP over the open ocean increases LW cloud top cooling, which initiates immersion freezing at cloud top (Fig. a, b). Through cooling in the updrafts and more available moisture, ice crystals can grow more efficiently by vapor deposition over the ocean (Fig. a, b). Overall, these processes lead to a higher IWP over open ocean than over sea ice. Note that over the ocean sedimenting ice in the form of snow contributes to 20 % of total rain and snow at the surface, while over sea ice this is reduced to 2 %.

These differences in cloud structure and properties (i.e., changes in cloud base and top, liquid and ice content, and precipitation efficiency) between the two cloud regimes agree with observations and previous LES results .

Due to the distinctly different cloud dynamics in both regimes, the effect of the aerosol perturbations on the clouds also differs. In the following we present results from several sensitivity simulations, where we investigated the cloud response to CCN and INP perturbations across different temperature ranges for the two cloud regimes.

To summarize the cloud micro- and macrophysical responses to both, INP and CCN perturbations, we calculated the mean cloud properties in Table 3 (and Table S1 for all CCN and INP perturbation simulations not listed in Table 3). Additionally, a schematic of our findings is shown in Fig. 11. The first panels in each row conclude our results from Sect. 4, indicating the existence of two different cloud regimes, a stratocumulus regime over open ocean and a homogeneous stratus regime over sea ice. These distinct regimes mainly result from differences in updraft speed, leading to different efficiencies in vertical moisture transport, subsequent cloud droplet growth, precipitation, and ice formation. Our results agree with previous findings obtained from satellites and measurement campaigns as well as the ACCACIA observations and modeling results. As has been observed by and simulated by , we also simulate a MPC over the ocean with a higher cloud top, larger droplets, increased LWP and IWP, and increased precipitation rates. The development of cumuli over the ocean as a response to increased surface fluxes additionally supports findings by .

As in and in agreement with previous ACCACIA studies our results indicate a higher cloud base over the open ocean and geometrically thicker clouds than over sea ice supporting findings by. Similarly to we also note structural differences over both surfaces with a stratocumulus cloud regime over the ocean versus a stratus cloud over sea ice. However, while relate changes in cloud properties mainly to changes in atmospheric stability over the open ocean and sea ice, our case studies are initialized with the same atmospheric stability profile; hence we suggest that the differences in surface latent and sensible heat fluxes may play a stronger role than previously suggested. In terms of radiative effects, the cloud over the open ocean and sea ice have different impacts on the net surface radiative balance. Note that the prescribed surface emissivity for ocean and sea ice is unchanged in both simulations. However, due to the 3 K warmer ocean, the LW surface emission is slightly increased over the open ocean and was quantified as 2.4±1.1 Wm-2 (spatiotemporal average over the first cloud-free hour). Additionally, we find cloud base height to be the dominating factor determining the net surface LW radiative balance for clouds sufficiently optically thick in the LW spectrum (LW and SW radiation fluxes are defined to be positive downwards throughout our study). As the cloud over sea ice has a lower cloud base, the cloud re-emits LW radiation at warmer temperatures, which reduces the net surface LW cooling (Table 3). The net surface SW radiation is directly coupled to cloud optical depth and by around 4 Wm-2 lower over the ocean, where the optically thicker cloud reflects incoming solar radiation more efficiently. Hence, we can extrapolate that during months with sufficient incoming solar radiation, clouds over the ocean might have a net zero to cooling effect compared to clouds over sea ice as also found by.

In a next step, we applied aerosol perturbations to the two contrasting cloud regimes. As our model setup allows for a prognostic treatment of aerosol–cloud interactions, we are able to quantify the cloud response to spatiotemporally resolved aerosol perturbations, which is a novel aspect compared to previous ACCACIA modeling studies . Both studies and used a prescribed Ndrop concentration and parameterized Nice concentrations (not considering interactive INPs) in their model setup, which have been adjusted in sensitivity simulations by . In their study, the authors found smaller droplets in a simulation with increased Ndrop, but found little effect on LWP or IWP. In contrast, we see a strong initial sensitivity of Arctic MPCs to CCN perturbations. Over ocean and sea ice, the LWP is already substantially increased with a perturbation of 200 and 100 CCN cm-3, respectively. With increasing CCN perturbations, Ndrop (Rdrop) increases (decreases), accompanied by an increase in LWP in agreement with. As a result of the larger LWP, LW cooling increases in the perturbed simulations throughout the cloud and the cloud deepens, such that more ice crystals nucleate through increased immersion freezing. Additionally, ice crystals grow by enhanced deposition rates in the perturbed simulations. This increased IWP in simulations solely perturbed by CCN was noted before by as well as . As a result of higher IWP and LWP, the cloud becomes optically thicker, reflects more SW radiation, and reduces the LW emission from Earth's surface (Table 3). This has a net cooling effect over daytime and during months with incoming solar radiation, but we expect the warming effect to dominate during polar winter.

Changes in the LW radiative properties are overall only moderate between the control and 1000CCN simulations, ranging from 6 % to 13 % over sea ice and ocean, respectively. Most likely the change in cloud structure between the two regimes determines the smaller response in net surface LW radiation to CCN perturbations over sea ice than over the open ocean. The temporary transition from a stratocumulus to a stratus cloud over the ocean for a perturbation of 1000 CCN cm-3 (Fig. ) increases the cloud re-emittance throughout the domain. On the contrary, the additional thickness of the stratus cloud over sea ice has a smaller effect, as the cloud structure is not considerably changed. Interestingly, the change in cloud base as simulated between ocean_control and ice_control has a stronger LW radiative effect on the Earth's surface (4.3 Wm-2) than CCN perturbations of 1000 cm-3 (3.4 Wm-2 over the ocean and 1.3 Wm-2 over sea ice). Conversely, the increased optical thickness of the perturbed clouds increases the reflectivity of the cloud and reduces the net surface SW radiation by 33 %–45 % over sea ice and ocean, respectively. This effect is larger than changes in net surface LW radiation, but is only important during daytime and spring to fall.

There is a strong regime dependence of the MPC response to CCN perturbations, which is novel in the context of aerosol–cloud interactions. Over sea ice, the cloud evolution remains substantially changed throughout the simulation period for any CCN perturbation >200 CCN cm-3. Over the open ocean, ice formation and growth as well as an increase in precipitation buffer the LWP response and lead to a relaxation of the LWP to its unperturbed state after 18 h simulation time. The cloud microphysical properties such as Ndrop and Rdrop remain perturbed. This results in a sustained Twomey brightening of the cloud even 18 h following the CCN perturbation. Combined with a lowering of the cloud base, the outgoing surface LW radiation is reduced by 2.5 Wm-2 and the incoming SW radiation by 3.5 Wm-2 during the last two simulated hours in ocean_1000CCN. The sustained net cooling is considerably smaller compared to cooling rates simulated during the whole period (Table 3), but indicates a remaining perturbation of the cloud radiative properties. Additional observations such as the ACCACIA campaign, but in polluted environments, could help to constrain such regime-dependent aerosol–cloud interactions. Also, further model studies including prognostic aerosols could expand our findings to a wider range of meteorological conditions (which we touched upon with our temperature change sensitivity tests).

The initial relative impact of increasing INP concentrations is larger compared to CCN concentrations. With more potential INPs, more particles are available for ice crystal formation by immersion freezing, which increases Nice and IWP. The increase in IWP is accompanied by a decrease in LWP through the removal of liquid water by deposition via the WBF process. This is consistent with previous studies investigating the effect of increasing INP or Nice on Arctic MPCs . The lower LWP in the simulations perturbed by INPs leads to an optically thinner cloud in the 10INP simulations, which increases net LW cooling at the Earth's surface, but has little effect on the net surface SW radiation (Table 3). This is an opposing effect to CCN perturbations, which generally have a moderate LW warming and a strong SW cooling effect on the underlying surface. Interestingly, the IWP increase for a perturbation of 10 INP L-1 is smaller than the IWP increase in the 1000CCN simulation over the open ocean (Table 3). We relate this difference to more efficient ice crystal growth by deposition in 1000CCN than in 10INP, supported by higher deposition rates (not shown) in experiments perturbed by CCN (Table 3).

The stratus cloud over sea ice initially shows a stronger response to INP perturbations than the stratocumulus cloud over open ocean. This different sensitivity to INP changes between surfaces is consistent with findings from . Similarly, found Arctic stratus over sea ice to be specifically vulnerable to INP perturbations. Note that with time, the IWP increase (LWP decrease) is more pronounced over the ocean, which can be related to stronger updrafts and cooling as well as the continuous cloud deepening over open ocean.

The analysis of MPCs within a changing Arctic environment has been the subject of a number of recent studies . Here, we addressed the cloud properties of MPCs in two differing regimes (i.e., sea ice and open ocean) in a series of high-resolution LES experiments. The robustness of the response to an aerosol perturbation was evaluated by applying our perturbation scenarios in warmer and colder environmental conditions. Our key findings are summarized as follows. 1.

The surface properties have a considerable impact on MPC properties. Our simulations support previous results obtained for the ACCACIA campaign : over the open ocean, strong turbulent surface fluxes increase the updraft velocities, which in turn favor the development of cumuli towers feeding moisture into the stratus layer. This increased vertical moisture flux leads to an increase in the cloud LWP and IWP, larger cloud droplets and ice crystals, and a higher cloud base and cloud top. Over sea ice, surface fluxes and in turn updraft velocities are low, which confines the cloud to a homogeneous stratus cloud. As the boundary layer is generally less moist, cloud droplet and ice crystal formation and growth are limited compared to the cloud over open ocean.

Extrapolating our findings to a future ice-free Arctic, increased ship traffic, and higher levels of pollution at the high latitudes, we find that changed surface conditions are likely to highly affect MPC dynamics, properties, and hence the radiative budget of the surface. The effect of pollution will be most effective in stratiform clouds over sea ice, where INP perturbations on the order of 10 L-1 lead to a strong cloud thinning and thus a change of the radiative balance on the order of a 4 Wm-2 cooling at the surface. Similarly, CCN perturbations may also cool the underlying surface through increased reflection of incoming SW radiation, but might have a warming effect in the absence of solar radiation. Considering that ship exhaust plumes may consist of both, CCN and INPs , the combined aerosol effect on Arctic MPCs may offset Arctic warming during the summer months, but we are doubtful it completely counteracts Arctic warming during the full year as also suggested by and .

Nevertheless, we note that our study has come caveats. We used the open ocean initial dropsonde profile to initialize both our cases (open ocean and sea ice), which is in contrast to . Over vast sea-ice-covered surfaces the boundary layer profile might highly differ from the boundary layer over open ocean and thus the clouds may evolve differently. However, we wanted to narrow possible differences over open ocean and sea ice down to surface fluxes, which become important over freshly melted sea ice or polynyas . In addition, due to runtime limitations it was not possible to simulate these high-resolution simulations for a longer time period. Thus, we unfortunately cannot draw any conclusions concerning cloud stability and persistence beyond 20 h.