We use the UK Met Office Large Eddy Model LEM, to investigate the influence of large-scale subsidence on mixed-phase marine Sc cloud microphysics. The set up is the same as that used by , whose study gives further details on the model itself. Momentum is conserved using the Piacsek–Williams PW; centred difference scheme, whilst the total variation diminishing (TVD) monotonicity-preserving scheme of , known as ULTIMATE, is used for scalar advection .

Cyclical boundary conditions and a damping layer (500 m below model lid) were imposed. Vertical profiles of potential temperature (Θ), water vapour mixing ratio (Qvap), and wind speed (U and V) were implemented to initialise the model (Fig. ): these profiles were extracted from previous LEM runs of Arctic mixed-phase Sc, specifically from 10 h in the ocean case detailed by , where a primary ice parameterisation derived from observations from the Aerosol-Cloud Coupling and Climate Interactions in the Arctic (ACCACIA) campaign was implemented. These fields give a stable BL experiencing strong (approximately 10–15 m s-1) N–S winds. A humidity inversion, coinciding with the BL temperature inversion, is present in the initial Qvap field (Fig. a): previous studies e.g. have shown that such inversions are often observed in the Arctic and may act as a source of moisture to BL clouds below. Surface sensible and latent heat fluxes were calculated by the model, using near-surface Θ and Qvap values, to represent an oceanic surface. The microphysics scheme was used, providing single-moment liquid (with a prescribed droplet number) and double-moment ice, snow, graupel, and rain.

In large-eddy simulation (LES) models, large-scale subsidence (Wsub) is often imposed as a tuning factor to maintain cloud-top height. In such models, Wsub is usually calculated from an imposed large-scale horizontal divergence. In practice, a constant divergence is assumed below the BL temperature inversion – with zero divergence above – producing a linear increase in Wsub with height below the inversion and a constant vertical wind above it . Here, we calculate Wsub using this method, increasing linearly with altitude up to 1500 m. At altitudes > 1500 m, Wsub=Wsub (1500 m) (representing zero divergence aloft).

In the literature, the imposed horizontal divergence in LES studies often ranges from 2.5 × 10-6 , through 3.75 × 10-6 , to 5 × 10-6 s-1 . In this study, three different levels of imposed divergence – 0, 2.5 × 10-6, and 5 × 10-6 s-1 – are used in four separate tests to investigate the role of large-scale subsidence in both stable and precipitation-favouring microphysical scenarios. The first three scenarios give an indication of how subsidence can affect the microphysics of Arctic mixed-phase clouds that remain at approximately the same latitude, whilst the fourth considers geographical movement. Details of the tests conducted are listed in Table . The control simulations apply no large-scale subsidence, apply a prescribed droplet number concentration (Ndrop) of 100 cm-3, and use the (hereafter, D10) parameterisation for primary ice nucleation. As in , an approximation of the D10 parameterisation is used, where we assume an aerosol particle number concentration of 2.20 cm-3 (for implementation in the parameterisation) throughout the domain.

Test 1 (Sect. ) considers the effect of imposing different levels of subsidence on the microphysical properties of a stable mixed-phase Sc layer. In Sect.  and , parameters relating to development of precipitation in the liquid or ice phase are varied to test the microphysical response under different levels of large-scale subsidence. For example, we expect to enhance rain formation by decreasing Ndrop (test 2, Sect. ) and increase snow formation by increasing Nice (test 3, Sect. ). However, decreasing Nice should sustain the liquid phase against the WBF mechanism, also likely affecting rain formation. Therefore, test 3 has the potential to affect both phases in the modelled clouds.

Test 4 investigates larger-scale BL interactions with a stable mixed-phase Sc layer. In CAOs, clouds move southwards off the sea ice and thus are subjected to a warming ocean surface. Model simulations in tests 1, 2, and 3 do not include any surface forcing: surface temperatures are allowed to vary through feedbacks with the BL above, yet they are not monotonically forced to become warmer. Such a forcing is applied in test 4 to investigate the combined influence of subsidence and a warming surface, simulating motion southwards. Near-surface temperatures are kept constant at 263.48 K until 5 h to allow adequate time for model spin-up, after which they are forced to warm linearly, in hourly increments, to 265.66 K at approximately 11 h 20 min. This warming profile was artificially constructed based on approximated ERA-Interim ECMWF reanalysis; 2 m temperature variations over the ocean in the Svalbard archipelago, close to the sea ice, during a cold-air outbreak 23 March 2013; seeand Fig. S1 in the Supplement for further details.

We employ a horizontal resolution of 120 m over a 16 km × 16 km domain centred on 71∘ N in the European Arctic to allow appropriate shortwave (SW) radiation calculations to be made by the model. Vertical resolution for the majority of model simulations was 20 m up to 1500 m, decreasing to 50 m between 1500 and 3000 m (domain lid) to reduce computational cost. A second domain structure was tested to check sensitivities to this set up: the high-resolution region was extended to 2300 m (again, reducing to 50 m above this height). Whilst our results are largely unaffected by the introduction of more vertical levels (not shown; see Fig. S2), this modified domain structure was applied in Sect.  (test 4) due to increasing cloud height.

Firstly, the influence of large-scale subsidence on the evolution of a stable mixed-phase marine Sc is examined. Prescribed droplet number concentrations and parameterised primary ice nucleation were not altered.

In all cases, the modelled clouds display the typical representation of a liquid-topped Arctic single-layer mixed-phase Sc, with heterogeneous ice number concentrations spread throughout the cloud below (not shown, Figs. S3–S5). Domain-averaged liquid- and ice-water paths (LWPs, IWPs) are shown in Fig. a and b, where the first 3 h of each simulation is excluded due to model spin-up. A stable Sc is modelled in the absence of Wsub (CNTRL, Fig. a). Increasing Wsub (LOSUB: low Wsub; HISUB: high Wsub) strengthens the temperature inversion, as shown in Table , thus reducing entrainment into the cloud from above the BL. Consequently, both the LWP and the IWP increase after approximately 5 h. These traces become more variable with time when subsidence is imposed, as is particularly visible in the IWP traces, suggesting increased dynamic activity in the modelled clouds. Longwave (LW) radiative cooling is instrumental in allowing this convection to develop (Fig. S6).

Planar X–Y views of the vertical velocity at 1000 m (W1000), LWP, and IWP fields at 11 h (Fig. c–k) further illustrate the effect subsidence has on the spatial structure of the clouds. With increasing Wsub, numerous regions of high LWP/IWP develop, with heightened heterogeneity across the domain. Domain-wide variability in W1000 also increases with Wsub. Broad updraught regions surrounded by narrow downdraught rings become apparent. Localised regions of high LWP and IWP can be associated with strong updraughts at 1000 m, and lower IWPs mirror the shape of the downdraught rings around the updraught regions. This locality becomes clearer with increasing Wsub (Fig. i, k).

Figure  shows a time series of turbulent kinetic energy (TKE) in panels a–c and vertical profiles of key properties in panels d–i. Increasing Wsub increases the snow and graupel mass tendencies below cloud (Fig. d). Strong snow sublimation is simulated at cloud top in all cases, with steady snow production in and below cloud. Regions of enhanced δQsg/δt coincide with strong rain evaporation (Fig. e); all of the rain produced evaporates below cloud, and no rain mass reaches the surface. Precipitation as snow does reach the surface; however, the spatial distribution becomes more heterogeneous with increased Wsub (not shown, Figs. S3–S5). Observational studies of Arctic mixed-phase marine Sc and North Atlantic CAOs have previously reported precipitation as snow below cloud with little rain measured, indicating that our idealised study is in broad agreement with measurements in this region.

A downward flux of heat and moisture into cloud top is modelled in all cases, caused by the temperature and humidity inversions (Fig. h, i): with increased levels of Wsub, w′Θ′ increases more so in the sub-cloud layer, whilst w′Qvap′ increases throughout the BL. Sub-cloud enhancement of w′Θ′ coincides with the top of regions of enhanced snow and graupel mass growth (Fig. d). Modelled ice–liquid potential temperatures Θil, following in the LO- and HISUB cases are colder than the CNTRL throughout the BL (Fig. f). All cases display a stable BL structure in the lower 1200 m of the BL and an unstable structure within cloud. A minor inversion is modelled at approximately 500 m in the CNTRL case which is co-located with a total water mixing ratio (Qtot) inversion and a moist surface layer (Fig. a).

TKE increases throughout the BL with increasing subsidence (Fig. b, c) and peaks at cloud top in all cases, likely influenced by the high evaporation and sublimations rates of rain and snow at the BL-capping temperature inversion. In all simulations, TKE typically increases with altitude through the BL. When subsidence is imposed, these TKE profiles tend towards a coupled, well-mixed BL through the top-down and bottom-up propagation of TKE. This coupling is particularly clear in the HISUB case (Fig. c); however, the cloud-top peak in TKE remains dominant throughout every case. Increasing Wsub produces a more coupled, dynamic BL due to a heightened LWP, efficient LW radiative cooling, and increased rain evaporation and snow growth below cloud.

The influence of large-scale subsidence on the formation of rain in a mixed-phase marine Sc is now considered. Prescribed droplet number concentrations were varied to a lower (Ndrop= 50 cm-3) and higher (Ndrop= 150 cm-3) threshold to affect rain formation: the modelled liquid mass is distributed amongst this concentration, such that a lower (higher) concentration will yield larger (smaller) cloud drops. Therefore, we expect the lower concentration of cloud droplets to allow for more efficient rain formation. conducted a similar sensitivity study when studying Sc-to-cumulus transitions with an LES model and found that decreasing droplet number concentrations, and enhancing precipitation, significantly affected the transition efficiency.

From Fig. a, the Ndrop50 scenarios produce a slightly greater LWP and IWP after 8 h than Ndrop100 or Ndrop150. Increasing Ndrop has little effect on the LWP or IWP; the results of the Ndrop100 and Ndrop150 cases are remarkably similar. Additionally, changing Ndrop has little effect on the depth of the cloud layer modelled in the CNTRL cases (shown by shading in Fig. c–k). In general, varying Wsub affects the modelled LWP, IWP, and dynamical fluxes more than the microphysical changes (varying Ndrop).

Figure  shows the mass production and sublimation/evaporation rates of snow/graupel and rain relative to the CNTRL in panels A and B respectively. Absolute domain-averaged number concentrations from each subsidence simulation are overlaid as contours. Varying Ndrop has only a minor effect on the time evolution of the snow and graupel number concentration (Nsg). Snow mass production rates (relative to the CNTRL) increase towards cloud base and below cloud with increasing Wsub, whilst snow sublimation rates at cloud top also increase. Non-zero snow concentrations reach the surface in all simulations (Fig. A).

Figure B shows a contrasting trend for rain production/evaporation. Decreasing Ndrop strongly affects Nrain as expected; for example, Nrain increases by approximately 9 L-1 between the HISUB_Ndrop100 and HISUB_Ndrop50 cases. For the LOSUB comparison, Nrain increases by approximately 6 L-1 in cloud. Increasing Wsub enhances the Nrain produced by decreasing Ndrop in the modelled cloud. δQsg/δt at cloud base, relative to the CNTRLs, does not change significantly when changing Ndrop, even with strengthened rain mass evaporation in this region; however, the below-cloud enhancement of δQsg/δt by increasing Wsub is apparent in each case.

Increasing Ndrop has a smaller effect on Nrain than decreasing it, as expected by the thermodynamic indirect effect; with more droplets available, droplet size decreases due to less competition for water vapour. Nrain decreases in Ndrop150 with respect to the Ndrop100 or Ndrop50 cases, and the in-cloud mass production and below-cloud evaporation rates are smaller. Despite this, increasing Wsub still marginally increases the mass production/evaporation rates with respect to the CNTRL_Ndrop150 case.

From these simulations, we suggest that the level of imposed large-scale subsidence can significantly affect the liquid phase in clean mixed-phase Sc, as Wsub positively forces the rain mass production/evaporation rates modelled in these precipitation-favouring microphysical scenarios.

The influence of Wsub on a mixed-phase marine Sc when changing ice number concentrations is now considered. Heterogeneous primary ice formation is represented using the D10 parameterisation with aerosol number concentrations calculated during the study by . Previous studies have shown that the lifetime of springtime single-layer mixed-phase clouds at high latitudes is strongly dependent on Nice. Here, a lower (Nice= D10 × 0.5) and higher (Nice= D10 × 2) threshold are implemented to change the number concentration of modelled ice, and snow, particles.

Figure  illustrates the domain-averaged LWP and IWP for test 3. The CNTRL cloud layer – as shown by the shaded area in Fig. c–k – becomes shallower with increasing Nice. When no subsidence is imposed (CNTRL, black/grey lines Fig. a), decreasing Nice increases the LWP as expected through the influence of the WBF mechanism, whereas increasing Nice has the opposite effect. However, in CNTRL_D10x2, both the LWP and IWP increase sharply after 9 h (Fig. a, b). This LWP peak occurs earlier with increasing Wsub (as shown by the blue and brown traces in Fig. a). This trend can also be seen in the IWP traces. The cause of this increase is not clear; however, it may be due to localised cloud convection caused by the high Nice, which has been previously modelled by .

Trios can be easily identified in Fig. a, where decreasing Nice affects the LWP more so than altering Wsub. Although the key factor influencing the LWP is Nice, Wsub acts to produce LWPs which are stable, or even increase, with time. In contrast, the CNTRL simulations typically produce a decreasing trend (with the exception of the D10 × 2 scenario).

Wsub affects the modelled fluxes (Fig. c–k) more so than altering Nice; however, the exception to this trend is the high Nice (D10 × 2) simulations, as subsidence does not stimulate this scenario as clearly as the other microphysical scenarios shown. Despite this, there are some notable differences in the flux profiles: for example, the extremes in the w′Θ′ profiles are more exaggerated in the LO- and HISUB cases than the CNTRL when a lower Nice is modelled (Fig. c–e). These comparisons suggest that Wsub can have a strong dynamical effect on liquid-dominated mixed-phase clouds, but its influence on those with more ice is limited.

From Fig. A, Nsg increases with increasing Nice and decreases only slightly between the LO- and HISUB cases. Additionally, snow mass sublimation rates at cloud top and production rates below cloud increase with increased Wsub. The increase in LWP and IWP in CNTRL_D10x2 at 9–10 h (Fig. 6a) affects our comparison, as increased snow mass is modelled at this time; therefore, the LO- and HISUB_D10x2 simulations produce less snow mass than the baseline.

Efficient rain mass production takes place with a lower Nice, as shown in Fig. B, due to the greater liquid mass being distributed over a fixed Ndrop. δQrain/δt increases with Wsub in the D10 × 0.5 case. In-cloud mass production and below-cloud rain evaporation rates increase in LO- and HISUB_D10x0.5 relative to CNTRL_D10x0.5, as do the snow growth rates below cloud (Fig. B). With less ice available, cloud-top radiative cooling becomes more efficient due to a heightened liquid fraction (Fig. a), increased rain formation (Fig. Bd), efficient snow growth (Fig. Ad), and vigorous turbulence (Fig. i). Consequently, cloud-top height increases in D10 × 0.5, whilst this ascent is suppressed in D10 × 2. This ascent adds complexity into the interpretation of Fig. Aa, Ad, Ba, and Bd as we are comparing clouds which are ascending at different rates. Strong cloud-top evaporation/sublimation of rain/snow is modelled above 1500 m with the ascending CNTRL cloud, whilst the LO- and HISUB cases have no activity at these altitudes; therefore the anomaly between the Wsub and CNTRL simulations appears positive at these heights.

LWP and below-cloud rain evaporation are enhanced in CNTRL_D10x0.5 with comparison to CNTRL_D10 and CNTRL_D10x2; however, w′2 is not strongly affected (Fig. i–k). Figure  shows δQsg/δt and δQrain/δt at 9 h to illustrate differences between the D10 × 0.5, D10, and D10 × 2 CNTRL cases. δQsg/δt is similar in the D10 and D10x0.5 simulations, whilst the LWP and rain evaporation/production processes are positively forced by decreasing Nice. In the turbulent subsidence cases, δQsg/δt does increase below cloud with increasing Wsub (Fig. A). This is the only key difference between decreasing Nice and increasing Wsub; therefore, increased latent heating through snow growth at cloud base – alongside heightened below-cloud rain evaporation and efficient cloud-top radiative cooling via a high LWP – is required to generate the heightened TKE (as illustrated here by w′2) in these scenarios. Convection is suitably induced in LO- and HISUB_D10x0.5 as the modelled snow growth rates are greater (Fig. S8). Whilst the same Nice is modelled in each of these scenarios, the subsidence cases produce a much colder BL than CNTRL_D10x0.5; therefore, the environmental conditions in LO- and HISUB_D10x0.5 facilitate snow growth below cloud, whilst the control produces comparatively inefficient growth conditions.

w′2 is greatest with the LO- and HISUB_D10x0.5 simulations (Fig. i) due to dynamical stimulation by the heightened rain mass evaporation and snow mass production at cloud base (Fig. B). The clouds are more dynamic with increasing Wsub, and it is the liquid-dominated (D10x0.5) clouds which are more vulnerable to this dynamic stimulation. Clouds with greater Nice suppress the liquid phase; therefore, Nice has a key role in mediating the strength of turbulent overturning generated in the mixed-phase clouds.

As described in Sect. , our previous tests consider scenarios that would elicit a microphysical response whilst keeping the surface boundary conditions approximately constant. Tests 1–3 are idealised and are not representative of the environmental forcings encountered when these clouds move southwards: observations show a sharp near-surface air temperature gradient in CAO flows transitioning southwards from the cold sea ice to the warm ocean. To address this, we further consider the combined dynamical impact of large-scale subsidence and a warming surface on both the BL and cloud microphysical structure. Whilst our domain size is not appropriate to resolve the explicit transition from closed- to open-cellular convection far downstream in a CAO, we will show how large-scale subsidence influences the microphysical stability of a stable mixed-phase marine Sc over a warming surface, upstream from this strong cellular convection.

More convection is modelled with time under the destabilising conditions of a warming surface (Fig. ). Domain-averaged LWPs and IWPs are similar in the subsidence cases, increasing almost monotonically with time (Fig. a, b). Slightly greater LWPs are modelled in HISUB_SURFWARM than in the LOSUB counterpart. Subsidence acts to produce greater LWPs and IWPs than the CNTRL up to approximately 10 h, at which point CNTRL_SURFWARM undergoes a significant convective transformation marked by a sharp increase in both LWP and IWP. Planar views of Fig. c–e show that, at this time, the CNTRL_ SURFWARM cloud contains numerous regions of very high LWP (> 200 g m-2) and IWP (> 50 g m-2) co-located with strong updraughts at 1000 m.

Cloud top and surface sources of TKE couple in all cases (Fig. a–c). The CNTRL case couples rapidly at approximately 10 h (Fig. a), coincident with the peak in LWP and IWP shown in Fig. a, b. Within approximately 1.5 h, the two TKE sources decouple again. Cloud top and surface sources of TKE dominate the LO- and HISUB_SURFWARM profiles separately from approximately 7 h onwards. LOSUB_SURFWARM displays a similar coupling at 10 h to CNTRL_SURFWARM, yet it remains coupled afterwards and undergoes a second TKE burst between 11 and 12 h. TKE evolution in HISUB_SURFWARM is more gradual than the CNTRL and LOSUB cases: the top-down and bottom-up propagation of TKE steadily increases with time to couple the separated cloud and surface sources.

Cloud-top height increases in CNTRL_SURFWARM (Fig. a), whilst this ascent is strongly suppressed in HISUB_SURFWARM (Fig. c) and marginally suppressed in LOSUB_SURFWARM (Fig. b). Negative w′Θ′ fluxes at cloud top again suggest entrainment of warm air into the cloud layer from above the BL in each case; however, this flux is stronger in CNTRL_SURFWARM than in the subsidence cases, indicating that greater entrainment rates are accompanying the cloud-top ascent.

Significantly larger values of w′Qvap′ and w′Θ′ are modelled below cloud in the CNTRL_SURFWARM simulation (0.052 g kg-1 m s-1 and 0.045 K m s-1, respectively) than in the subsidence cases, coinciding with the rapid BL coupling shown in Fig. a. Convective activity increases at this time, with w′2 increasing up to 0.90 m2 s-2 in cloud alongside a peak (cloud top) TKE of 2.8 m 2 s -2 (Table ). Additionally, rain mass production is enhanced in CNTRL_SURFWARM; however, below-cloud rain evaporation is still weaker than in the LO- and HISUB_SURFWARM simulations. Rain evaporative cooling below cloud in LO- and HISUB_SURFWARM acts to decouple the surface and in-cloud heat sources from each other (Fig. i). As a result, the w′Θ′ profiles swing through significant extremes below cloud: from 0.021, through -0.011, to 0.028 K m s-1 in the HISUB_SURFWARM case. Furthermore, the warming surface produces an unstable Θil profile at the surface in each simulation (Fig. f).

Z–X slices of several microphysical variables from CNTRL_SURFWARM are shown in Fig.  to illustrate the cloud structure at 10 h. In the bottom panel, below-cloud cumuli form which either couple to the Sc layer (white ellipses) or remain separate (red ellipses). Cumuli are identified by adjacent updraught/downdraught regions with 100 % relative humidity (or close to 100 %). These cumuli structures are clearly visible in the Qliq contour field (top panel, Fig. ). Cumuli can be seen from 8 h onwards and become more frequent with time. At this time, two spatially close cumuli form at approximately -7000 and -3500 m, marking the boundaries of a detraining layer of moisture above cloud top. Additionally, a similar, completely detached moist layer can be seen above cloud top coinciding with the 6000 m cumulus.

HISUB_SURFWARM has much larger updraught and downdraught regions than CNTRL_SURFWARM: from approximately 11 h onwards, these often extend to almost the full height of the BL (Fig. ). No distinct sub-cloud cumuli can be identified in HISUB_SURFWARM, whereas these are common in CNTRL_SURFWARM (Fig. ): the addition of subsidence acts to suppress their formation and allow a more homogeneous Sc layer to be maintained in a BL undergoing top-down and bottom-up coupling of TKE. The coupling process is more gradual in HISUB_SURFWARM than the CNTRL or LOSUB counterparts, suggesting that subsidence plays a role in whether or not this rapid TKE coupling and cloud-top ascent can take place.

Imposing large-scale subsidence in simulations of marine Arctic mixed-phase Sc increases the LWP and IWP of the modelled clouds through increased convective activity throughout the domain (Fig. ). Wsub does not affect the cloud depth (Figs. , ); only Nice notably affects the modelled cloud depth (Fig. ). Dynamical stimulation by subsidence – which would sustain a mixed-phase Sc for longer against the WBF mechanism – may therefore have been previously missed in observations and modelling studies. Increasing Wsub has a greater effect on the liquid phase than the ice phase (Figs. , , ); however, increasing subsidence causes the development of heterogeneity in the LWP and IWP fields, leading to instabilities in the modelled clouds. In particular, the radiative properties of the clouds would be affected by the heterogeneous spread in LWP, where regions of high LWP would be more reflective to incoming SW radiation and more efficiently cooled via longwave radiative cooling.

Localised regions of high IWP are typically co-located with updraughts in our simulations, likely due to the method of parameterising ice nucleation in our model. Namely, additional nucleation mechanisms (e.g. contact, immersion) are not represented to give us a predictable source of ice number concentrations similar to. These mechanisms would likely influence our results if they were explicitly resolved in our model; for example, we would expect contact nucleation in downdraughts, through interaction with interstitial aerosol particles.

Subsidence strongly influences the LWP; however, increasing Wsub marginally increases the domain-averaged IWP (Fig. b). Figure b shows that the peak IWP attained by CNTRL_D10x2 is also achieved in the HISUB_D10 case, suggesting that increasing Wsub can have a similar effect on the bulk ice properties of the cloud as increasing Nice. However, a much larger LWP is also modelled when subsidence is imposed, creating a microphysical structure that may be more robust against the WBF mechanism. This may allow mixed-phase conditions to be sustained for longer against a higher Nice – a problem that is often faced when modelling Arctic mixed-phase Sc .

In the chosen microphysical scenarios that may affect precipitation development in mixed-phase marine Sc, Wsub enhances rain evaporation at cloud top and base. Increased subsidence leads to larger rain mass production rates and a greater Nrain within cloud, and this effect is particularly clear when lowering either Ndrop (Ndrop50, Fig. ) or Nice (D10 × 0.5, Fig. ). In these cases, the increase in Nrain due to subsidence is less than can be attributed to the microphysical changes; for example, an increase of approximately 6 L-1 is modelled in the Ndrop50 scenario due to increasing Wsub, whilst an increase of approximately 9 L-1 is achieved by lowering Ndrop from 100 to 50 cm-3 (Fig. B).

From tests 2 and 3, Wsub amplifies the modelled turbulence in scenarios allowing for efficient rain formation (e.g. Ndrop50, Fig i). Wsub also acts to promote more turbulence (Fig. k) and rain formation (Fig. B) in a microphysical scenario that produces little rain in its absence (Ndrop150). Conversely, increasing Nice in test 3 does not have the same effect, and Wsub does little to promote turbulence in this scenario. Whilst the ice categories do little to stimulate convection, they are responsible for suppressing rain formation; for example, a higher Nice (and thus, Nsg) suppresses the strong rain production/evaporation processes modelled at a lower Nice (Fig. B). With weakened rain formation and evaporation, less vigorous overturning is modelled in D10 × 2 than D10 or D10 × 0.5. Whilst the liquid phase drives the development of dynamical overturning, the ice phase has a strong influence – through the WBF mechanism – on whether this convective activity can actually develop.

Similarly, total ice number concentrations (ice + snow + graupel, Nisg, Fig. b) are largely unaffected by a warming surface (with comparison to Fig. S3b); however, both Qliq and Nrain increase. Precipitation formation is enhanced in downdraught regions (Figs. , ). Weaker below-cloud rain evaporation occurs in CNTRL_SURFWARM (Fig. e), and the upward propagation of heat and moisture from the surface causes distinct cumuli to form below cloud and join with cloud base. These cumuli dynamically stimulate the cloud from below (Fig. ) and have a similar effect on the cloud as the introduction of subsidence in tests 1–3; for example, the warming surface allows a greater Nrain to form in cloud (Fig. ). In fact, Nrain in CNTRL_SURFWARM is much more comparable with the corresponding domain-averaged values of the LO- and HISUB_Ndrop50 simulations in test 2 (Fig. B) – the efficient-liquid-precipitation cases – than any of the previous control simulations.

These findings indicate that subsidence has the potential to positively force the liquid phase in Arctic mixed-phase marine Sc whilst having little effect on the ice phase. presented observations of cloud microphysics over the transition from sea ice to ocean and found that the ice phase changed little under the dynamical evolution of the BL, whilst the liquid water content increased four-fold. Therefore, mixed-phase clouds with low number concentrations of primary ice, such as those commonly observed in the springtime Arctic, may be vulnerable to dynamical changes induced by subsiding air from above or a warming surface from below.

In tests 1–3, convective activity increases with Wsub through increased BL TKE and below-cloud w′2. Solar heating acts to marginally offset the formation of defined closed-cellular structure; however, the cloud-driven convection is strongly dependent on cloud-top LW radiative cooling (see Fig. S6). Additionally, rain mass formation rates, number concentrations, and the domain-averaged LWP increase with increasing Wsub. This finding mirrors the conclusions of , where the authors found that increasing the resolved TKE and/or temperature positively influences the liquid phase in an ice-saturated environment, as these conditions contribute towards sustaining water saturation.

In the absence of surface warming, all modelled BLs display a stable Θil profile. This stability is likely influenced by the stable conditions used to initialise the model, and one must note that only a single set of initial conditions were used in this study. A moist layer is maintained close to the surface in the CNTRL simulation (Fig. a), below the sub-cloud mixed layer, whereas this moisture source is eroded in the subsidence cases. Additionally, the CNTRL case presents a minor BL Θil inversion, and a stronger Qtot inversion, at approximately 500 m. The combination of these inversions and the moist surface layer suggests that the CNTRL simulation is, in fact, more strongly decoupled from the surface than the subsidence cases at the time step shown (9 h, Fig. ). However, the subsidence cases display a similar strongly decoupled profile in TKE to the CNTRL at earlier times (e.g. 5 h, Fig. ). TKE increases with time in the BL when subsidence is imposed, and it promotes top-down mixing of TKE through the sub-cloud layer towards the surface by the end of the simulations, tending towards a coupled profile. However, cloud-top TKE still dominates the BL profiles in the LO- and HISUB cases (Fig. b, c), suggesting that mixing throughout the BL is still not homogeneous and the clouds remain approximately decoupled from the surface by the termination time of the simulations. This decoupling allows radiative cooling at cloud top and evaporative cooling/latent heating below cloud to drive convective activity in the cloud layers, irrespective of surface sources.

With a larger LWP, stronger cloud-top radiative cooling is expected, promoting a greater cloud-top height . Subsidence acts to restrict cloud-top ascent by reinforcing the BL temperature inversion (Table ), thus lowering the entrainment rate of air from above. Cloud LWP increases in the absence of notable entrainment, allowing for stronger cloud-top LW radiative cooling and subsequent precipitation development within cloud. BL temperatures are therefore cooler with imposed subsidence than without (Fig. i), due to the combined effect of reduced entrainment, strong cloud-top radiative cooling, and enhanced evaporative cooling below cloud.

A lack of subsidence combined with a warming surface acts to push cloud top significantly higher, and increase the LWP, through the formation of the below-cloud cumuli (namely in CNTRL and LOSUB_SURFWARM, Fig. d). Higher levels of Wsub act to stabilise the Sc layer and suppress cumuli formation from the warming surface (as is seen in the CNTRL_SURFWARM case). TKE production is positively influenced in the CNTRL and LOSUB_SURFWARM cases, with strongly separated cloud and surface sources, and peak values approximately 3 times greater than modelled in test 1 (Tables  and ). Cloud-top TKE splits in two in both CNTRL and LOSUB_SURFWARM (Fig. a, b); however, it is unlikely that this is a domain artefact as the vertical resolution is consistent through this altitude range. It is possible that the PW advection scheme is introducing spurious oscillations into the advected quantities, caused by the the sharp gradient at the cloud boundary due to the formation of these dynamic cumuli as discussed by. Peak TKE is only marginally stronger in HISUB_SURFWARM than in test 1 (Tables  and ), suggesting that higher Wsub offsets the efficient in-cloud TKE production which occurs when the system is additionally forced by a warming surface. By suppressing the formation of below-cloud cumuli, subsidence acts to produce a stable, yet dynamic, Sc layer, whilst strong convection and spatial heterogeneity are simulated with low or no subsidence. With more heterogeneity, there is an increased likelihood for instability in the cloud layer, which will likely influence the fate of the cloud downstream.

Whilst CAOs are discussed to motivate our study, we must stress that our chosen domain configuration is not optimal for the explicit study of Sc-to-cumulus transitions downstream in a CAO. Large high-resolution domains are required to accurately resolve the small-scale microphysical processes within these phenomena ; however, our domain size and resolution are restricted by computational expense. demonstrated that our spatial resolution may allow entrainment rates to be overpredicted by approximately 50 % . Whilst the authors concluded that the resolution imposed here can still provide a useful insight into BL evolution, accurately resolved turbulence requires higher spatial resolution. found that a higher-resolution set-up produced enhanced BL convection and a deeper BL depth. Furthermore, found that the simulated vertical mixing of vapour and Θ fields improved, and the modelled LWP increased, in their open-cellular convection simulations by increasing spatial resolution.

To test the influence of resolution on our findings, we increase the horizontal resolution to 60 m (Δx) and the vertical resolution to 10 m, whilst maintaining the domain height. This set-up therefore decreases the spatial extent of the domain by half in both X and Y. Vertical resolution was 10 m up to 2000 m, decreasing to 20 m above this height. By increasing our model resolution, we aim to provide a more accurate representation of the modelled entrainment rates. Due to computational expense, only two test cases are considered: the CNTRL and HISUB simulations from test 1 (Sect. ).

Little difference between the domain-averaged LWP and IWP can be identified between the CNTRL cases (black/grey, Fig. a, b). In the HISUB example, increasing the model resolution amplifies the irregularities in both the LWP and IWP traces. In particular, the IWP is significantly more variable with time than in the default set-up.

In general, increasing the resolution does not alter the trends identified previously – for example, the positive below-cloud moisture fluxes, higher below-cloud rain mass evaporation and snow mass growth rates, and increased TKE with increasing Wsub. In fact, it should be noted that the below-cloud rain mass evaporation rates are enhanced with comparison to the coarse-resolution HISUB case, suggesting that the evaporation rates shown in Sect. , , and  may be underestimated. The Qtot profiles illustrate clear decoupling in the CNTRLs, with a weaker inversion in the HISUB cases. Additionally, both Δx60 simulations produce a greater TKE peak towards the surface, in addition to the peak simulated at cloud top, due to the dominating influence of the sub-grid contribution to the TKE towards the surface (Fig. h).

Whilst we can test the influence of increased resolution on our findings, increasing our domain size would be too computationally expensive for our set-up. Larger domains are often used to allow mesoscale interactions between developing open convective cells to be resolved. suggest that a domain of 100 × 100 km, with 50–100 m spatial resolution, is required to truly encapsulate any mesoscale interactions between developing convective cells in CAOs. We cannot speculate what mesoscale interactions may occur between the different scenarios presented here; however, one must note that such interactions have been previously simulated to occur over the transition between closed and open convective cells in CAOs; thus these effects should be investigated in further work.