Arctic mixed-phase clouds sometimes dissipate due to insufficient aerosol: evidence from observations and idealized simulations

. Mixed-phase clouds are ubiquitous in the Arctic. These clouds can persist for days and dissipate in a matter of hours. It is sometimes unknown what causes this sudden dissipation, but aerosol-cloud interactions may be involved. Arctic aerosol concentrations can be low enough to affect cloud formation and structure, and it has been hypothesized that, in some instances, concentrations can drop below some critical value needed to maintain a cloud. We use observations from a Department of Energy ARM site on the north slope of Alaska at Oliktok Point (OLI), the ASCOS 5 field campaign in the high Arctic Ocean, and the ICECAPS-ACE project at the NSF Summit Station in Greenland (SMT) to identify one case per site where Arctic boundary-layer clouds dissipated coincidentally with a decrease in surface aerosol concentrations. These cases are used to initialize idealized large eddy simulations (LES) in which aerosol concentrations are held constant until, at a specified time, all aerosols are removed instantaneously – effectively creating an extreme case of aerosol-limited dissipation which represents the fastest a cloud could possibly dissipate via this process. These LES simulations 10 are compared against the observed data to determine whether cases could, potentially, be dissipating due to insufficient aerosol. The OLI case’s observed liquid water path (LWP) dissipated faster than its simulation, indicating that other processes are likely the primary driver of the dissipation. The ASCOS and SMT observed LWP dissipated at similar rates to their respective simulations, suggesting that aerosol-limited dissipation may be occurring in these instances. We also find that the microphysical response to this extreme aerosol forcing depends greatly on the specific case being 15 simulated. Cases with drizzling liquid layers are simulated to dissipate by accelerating precipitation when aerosol is removed while the case with a non-drizzling liquid layer dissipates quickly, possibly glaciating via the Wegener-Bergeron-Findeisen (WBF) process. The non-drizzling case is also more sensitive to INP concentrations than the drizzling cases. Overall, the simulations suggest that aerosol-limited cloud dissipation in the Arctic is plausible and that there are at least two microphysical pathways by which aerosol-limited dissipation can occur. conclude that the observed dissipation was not driven entirely by a lack of aerosol particles. Prior to this aerosol removal time, a temperature nudging scheme is used to maintain a 205 stable cloud. At each time step, each grid point is linearly nudged back to the initial temperature profile with a time scale τ = 1 h. Nudging values are computed based on the current domain-average temperature profile, so all grid points at a given height z are nudged the same amount. The result in all simulations is a cloud that is quasi-steady in thickness and water content. After the removal of aerosol from the model, the temperature nudging scheme is turned off and the thermodynamics of the system are allowed to evolve naturally - this is done so that the post-aerosol environment is able to evolve naturally. Large-scale 210 subsidence is applied throughout the simulation by imposing a horizontal divergence of 2 × 10 − 6 s − 1 at every model level, with a boundary condition of w sub = 0 at the surface.


Introduction
The Arctic has been shown to be extremely sensitive to a warming climate, with data showing the Arctic warming anywhere from 1.5 -4.5x the global mean warming rate (Holland and Bitz, 2003;Serreze and Barry, 2011;Cohen et al., 2014;Previdi et al., 2021). Clouds, in general, directly affect the surface energy budget and can act as net-warming or net-cooling influences, depending on their specific physical characteristics. Of particular note in the Arctic environment are low-level, boundary layer 25 stratocumulus clouds which cover large fractions of the Arctic throughout the year (Shupe, 2011). They have been found to be a net-warming influence on the surface, except for a short period in the summer when they act as a net-cooling influence (Intrieri et al., 2002;Shupe and Intrieri, 2004;Sedlar et al., 2011). These clouds tend to be mixed-phase, meaning they simultaneously contain liquid and ice water. Shupe et al. (2006) found that mixed-phase clouds accounted for 59% of the clouds identified during a year-long campaign on an icepack in the Beaufort Sea, with the remaining 41% consisting of mostly ice-only clouds. In this study, we investigate whether or not aerosol-limited dissipation occurs on a case-by-case basis. While likely infrequent, this method of cloud dissipation is worth examining in more detail because of how sensitive the Arctic environment is 90 to low-level cloud cover, and the highly uncertain changes in Arctic aerosol concentration (both natural and anthropogenic) in a warming climate (e.g. Schmale et al., 2021). We examine three observed cases of potential aerosol-limited dissipation across three different environments (northern Alaskan coast, high Arctic pack ice, and the Greenland ice sheet) and use large eddy simulations (LES) to simulate a "worst-case scenario" of aerosol-limited dissipation: immediately removing all aerosols from a simulated cloudy environment and comparing changes in cloud properties to observations, which should indicate whether or 95 not these cases should continue to be investigated as examples of this phenomenon.  boundary layer mixed-phase clouds. Figure 1 shows the three case locations on a map. Details of each case are summarized in were observed to decrease from > 50 cm −3 to < 20 cm −3 in a span of 4 hours. Many such periods exist, and the results were 110 examined manually to select cases where aerosol-limited dissipation may have been a factor in transitioning from a cloudy to cloud-free environment.
One such case (Fig. 2) occurred on the 12th of May, 2017. At 09:00 UTC, the CPC measured a transition in aerosol concentration from ∼ 100 cm −3 to <10 cm −3 in the span of about one hour (Fig. 2c). Aerosol data from OLI was particularly noisy, with a clear trend of concentrations ∼ 100 cm −3 but with intermittent spikes upwards of 1000 − 10000 cm −3 (not shown).

115
To smooth out the data and best show what we consider to be a representative aerosol concentration timeseries, we filtered out values > 1000 cm −3 and downsampled the result from one-second to one-minute averages. Data from the Ka-band ARM prior, dissipated coincidentally with a decrease in surface aerosol concentration from >100 cm −3 to <10 cm −3 (Sedlar et al., 2011;Mauritsen et al., 2011;Sotiropoulou et al., 2014). aerosol concentrations were collected using a differential mobility particle sizer (DMPS; Birmili et al. 1999;Tjernström et al. 2014) measuring size distributions of particles between 3 nm -10 µm. Helicopter flights measuring aerosols during this time found that concentrations were below 10 cm −3 (for aerosols 145 < 14 nm) for the entirety of the boundary layer (Stevens et al., 2018) during a flight in the dissipation period at 20:13 UTC.
After cloud dissipation, winds which were previously calm were observed to more consistently blow from the northeast (not shown). At the same time, surface temperature drops over 6 • C, though it's unclear whether there was a change in airmass or if temperature dropped as cloud is no longer present as a warming influence on the surface. Surface pressure analysis (Fig. 3d) shows the extension of northern high pressure directly over the location of Oden at this time, suggesting a possible change in 150 airmass.
Like the OLI case, RH values are generally high throughout the boundary layer. Unlike OLI, there is a dry layer at 400 m.
A change in θ and RH profiles at 400 m indicate weak decoupling at this level. This case has previously been investigated as existing in a potentially tenuous regime (Mauritsen et al., 2011;Loewe et al., 2017;Stevens et al., 2018;Tong, 2019).  OLI and ASCOS, below-cloud RH decreases towards the surface, reaching ∼50% directly above the surface inversion.

Model Description
The Colorado State  et al., 1992;Jiang et al., 2001;Jiang and Feingold, 2006). Radiation parameterization is provided by the Harrington scheme (Harrington, 1997), and turbulence is parameterized by a Deardorff level 2.5 scheme, which parameterizes eddy viscosity as a function of turbulent kinetic energy (TKE).
RAMS uses a double-moment bulk microphysics scheme (Walko et al., 1995;Meyers et al., 1997;Saleeby and Cotton, 175 2004) that predicts the mass and number concentration of eight hydrometeor categories: cloud droplets, drizzle, rain, pristine ice, aggregates, snow, hail, and graupel. Each of these hydrometeor categories is represented by a generalized gamma distribution. The scheme simulates nucleation (cloud and ice), vapor deposition, evaporation, collision-coalescence, melting, freezing, secondary ice production, and sedimentation. Cloud droplets are activated from aerosol particles using lookup tables (Saleeby and Cotton, 2004) built based on Köhler theory and cloud droplet growth equations formulated in Pruppacher and Klett (1997).

180
Water vapor is depleted from the atmosphere upon activation by assuming that newly activated droplets have a diameter of 2 µm.
Ice crystals are heterogeneously nucleated by the parameterization in DeMott et al. (2010), with the number of ice nuclei (L −1 ) given by: Where n in is the ice nuclei number concentration, T k is the air temperature in Kelvin, and a, b, c, d are constants. The variable n in the original DeMott parameterization is the number concentration of aerosol particles with diameters larger than 0.5 µm,  Table 1).

190
The observations were used to generate an initial sounding and to specify aerosol concentration for each simulation. We use a simplified aerosol treatment in which number concentrations are fixed to a single value throughout the domain. A list of experiments and initial aerosol/ice nuclei concentrations is found in there was little change in the liquid water for n = 1, 5, or 10 L −1 . There were moderate differences in ice water content, and as there are ice water path (IWP) retrievals for both the OLI and ASCOS cases, we picked a value of n that yielded simulated IWP values closest to observations. For the SMT case, no ice measurements were available. Both simulated ice and liquid were sensitive to choice of n, so a value of 0.1 L −1 was used; this value is consistent with currently unpublished INP data from Summit Station (available upon request) and resulted in simulated liquid water path that was closest to observations. 220 3 Results Figure 5 shows domain-averaged liquid water (color shading) and ice (dashed contours at 0.01 and 0.001 g kg −1 ) show typical Arctic mixed-phase clouds in which a layer of supercooled liquid water is situated at cloud top with ice precipitating below.
In OLI and ASCOS, the liquid layer is well-above the ice layer (∼200 m from cloud top to the 0.001 g kg −1 ice contour), whereas in SMT the ice extends nearly to cloud top.
225 Figure 6 shows the domain-mean liquid water path (LWP) for the OLI, ASCOS, and SMT simulations and the corresponding observed LWP. Observed LWP data were taken from microwave radiometers at OLI (Gaustad, 2014), ASCOS (Westwater et al., 2001), and SMT (Cadeddu, 2010) This figure shows that, in all cases, the simulated LWP decreases to near-zero within hours of the aerosol removal time (09z in OLI, 06z in ASCOS and SMT). Both the OLI and ASCOS simulations show a slow LWP response to aerosol removal, with LWP approaching 0 g kg −1 in about 4-5 hours. The SMT simulation, on the other hand, has 230 a very pronounced LWP response to aerosol removal, with LWP approaching zero within 2 hours. With instantaneous aerosol removal, the simulations should theoretically represent the fastest possible dissipation of a cloud due to insufficient aerosol.
Where this simulated LWP response is slower than observations -such as OLI -it is likely that a lack of aerosol is not in fact the primary driver of dissipation. Where the simulated LWP response is more similar to observations (ASCOS and SMT), it is more likely that these are indeed cases of aerosol-limited dissipation. c) SMT Observed Modeled 08z 09z 10z 11z 12z 13z 14z 15z 16z 17z 05z 06z 07z 08z 09z 10z 11z 12z 13z 14z 05z 06z 07z 08z 09z 10z 11z 12z 13z 14z Each case will now be discussed in detail; since the time of aerosol removal was determined rather subjectively, and because the aim of this paper is not to compare directly with observations (but instead to compare timescales), all further discussion will be discussed in the context of hours before/after aerosol removal, instead of UTC, to better compare cases with one another.

OLI
It is evident from Figure 6a that the OLI cloud dissipation was not due to a lack of available aerosol. While the observed LWP 240 decreased from 100 g kg −1 to <10 g kg −1 in ∼1 hour, modeled LWP took 4-5x this time. While the OLI case may not be a real-world example of aerosol-limited dissipation, examining its simulated response to aerosol removal when compared to the different cases still yields valuable insights to this phenomenon.
Domain-average 2D and column-integrated liquid and ice budgets, radiative heating, and vertical momentum flux for OLI are shown in Figure 7. After a 1-hour spin-up period (not shown), the cloud settles to quasi-equilibrium with approximately 245 constant liquid precipitation reaching the surface and consistently positive integrated cloud droplet growth by condensation, which occurs primarily at cloud base, where supersaturation is largest, and at cloud top. In the cloud interior there is slight net liquid evaporation and net ice depositional growth due to an active WBF process. The growth of ice and liquid are balanced by persistent precipitation of both liquid and ice hydrometeors throughout the pre-aerosol removal time period. Riming makes up only a small part of the liquid and ice budgets. Radiative cooling (Fig. 7c) is strongest at cloud top as expected, which drives the  Liquid budget (a) shows condensational growth of cloud and rain, and removal by precipitation (precip) and riming. Ice budget (b) shows growth of all ice species by condensation (cond), riming, and removal from precipitation (precip).
After the removal of aerosol, a large increase in liquid precipitation and a smaller relative increase in ice precipitation occur.
Removing aerosol inhibits the nucleation of new cloud droplets, meaning that any supersaturation must be condensed onto existing droplets rather than being used to create new droplets. This results in a rapid increase in droplet sizes (not shown) 255 and an enhanced collision-coalescence process, leading to increased liquid precipitation. The precipitation is initially strongest near cloud base and contributes to a rise in cloud base. Since new droplets are unable to be nucleated (and available liquid to condense upon is being precipitated), supersaturation levels increase (not shown). Approximately three hours after aerosol removal, cloud condensation falls off sharply. Figure 7(d-e) show that, after aerosol removal, there is an increase in ice growth which maximizes after liquid is mostly removed (Fig. 6a). However, at this point the cloud top radiative cooling has ceased, 260 circulations weaken, and the ice begins to slowly decay as well. Figure 5a shows that, after aerosol removal, the OLI simulation dissipates with a rising cloud base, and a lesser rising of the cloud top. However, radar observations (Fig. 2b) show a cloud that dissipates with a cloud top that is lowering. After aerosol removal (and temperature nudging is turned off) in the OLI simulation, the entire boundary layer cools and stabilizes (not shown). As a result of this stabilization, turbulence generated by cloud top cooling is not able to extend as far down as before, 265 resulting in a rising cloud bottom. It is not clear what is causing the cloud top to lower in the observed case, but this difference in the cloud shape during dissipation -combined with the much faster observed LWP response compared to simulationsindicates that the observed dissipation is likely due to larger-scale factors such as the possible weak frontal passage described in section 2.1.1. We also speculate that the liquid water profile added to the model initialization results in cloud-top LWC