Modelling micro- and macrophysical contributors to the dissipation of an Arctic mixed-phase cloud during the Arctic Summer Cloud Ocean Study (ASCOS)

The Arctic climate is changing; temperature changes in the Arctic are greater than at mid-latitudes, and changing atmospheric conditions influence Arctic mixed-phase clouds, which are important for the Arctic surface energy budget. These low-level clouds are frequently observed across the Arctic. They impact the turbulent and radiative heating of the open water, snow and sea-ice covered surfaces, and are influencing the boundary layer structure. Therefore the processes that affect mixedphase cloud life cycles are extremely important, yet relatively poorly understood. In this study, we present sensitivity studies 5 using semi-idealized large eddy simulations (LES) to identify processes contributing to the dissipation of Arctic mixed-phase clouds. We found that one potential main contributor to the dissipation of an observed Arctic mixed-phase cloud, during the Arctic Summer Cloud Ocean Study (ASCOS) field campaign, was a low cloud droplet number concentration (CDNC) of about 2 cm−3. Introducing a high ice crystal concentration of 10 l−1 also resulted in cloud dissipation, but such high ice crystal concentrations were deemed unlikely for the present case. Sensitivity studies simulating the advection of dry-air above the 10 boundary layer inversion, as well as a modest increase in ice crystal concentration of 1 l−1, did not lead to cloud dissipation. As a requirement for small droplet numbers, pristine aerosol conditions in the Arctic environment are therefore considered an important factor determining the lifetime of Arctic mixed-phase clouds.

used observations from the summertime high Arctic to describe a tenuous cloud regime where cloud formation is limited by CCN availability, and where a small increase in aerosols can result in a significant cloud warming effect at the surface. A subsequent modeling study by Birch et al. (2012) confirmed that accurately simulating cloud formation and dissipation under low CCN conditions improves the model representation of the surface energy budget and temperature. 5 In situ observations from field campaigns are a key part of improving model simulations of Arctic mixed-phase clouds and their impact on climate. This study exploits observations taken during the Arctic Summer Cloud Ocean Study (ASCOS) in summer 2008 (Tjernström et al., 2012). During this campaign, an Arctic mixed-phase stratiform cloud layer was observed that persisted for an extended period, and then suddenly dissipated.
We seek to find potential mechanisms leading to the dissipation of this cloud layer, using the COnsortium for Small-Scale 10 MOdeling (COSMO) model (Schättler et al., 2015), run in a Large Eddy Simulation (LES) mode with high vertical, horizontal, and temporal resolution, to explore the dissipation. The study is divided into the following sections: Section 2 outlines the ASCOS field campaign and the period of interest. An overview is presented of the model, its setup, and sensitivity experiments in Section 3. Section 4 describes the results of three sensitivity experiments. A discussion and conclusions are presented in Section 5. This study will focus on an episode towards the end of the ice drift, around 31 August 2008. A low-level stratiform cloud layer 20 had been quasi-persistent for about one week but dissipated rapidly in the evening of the 31 August 2008 (Sedlar et al., 2011;Mauritsen et al., 2011;Sotiropoulou et al., 2014). The period of this persistent cloud layer (23 August to 2 September 2008) ( Fig. 1) was dominated by a high pressure system, with passages of a few weak fronts (Tjernström et al., 2012). A detailed description of the meteorological conditions can be found in Tjernström et al. (2012). Observation from the vertically pointing Doppler millimeter cloud radar (MMCR) shows the cloud top at around 1 km during the morning hours, with a thinning and 25 lowering cloud top during the afternoon (Fig. 1). The laser ceilometer measured the cloud base at around 600 m to 700 m in the morning. During the day, the cloud base decreased towards the surface. With observations from the MMCR, a dual-channel microwave radiometer (MWR), a ceilometer, and radiosondes, the cloud type was classified as mixed-phase during the first half of the 31 August 2008. The cloud type classification follows the method by Shupe (2007). The retrieval of the liquid water path (LWP) from the MWR contains of an uncertainty of 25 g m-2 (Westwater et al., 2001), explaining the negative LWP 30 observations in Fig. 2a. During the 31 August 2008, the LWP increased from around 90 g m −2 to values over 300 g m −2 and varied considerably during the first half of the day. Finally, the LWP reached values around 50 g m −2 in the afternoon. The ice water path (IWP) is integrated from profiles of the ice water content (IWC), which are derived from MMCR reflectivity power-law relationships at vertical levels deemed to predominantly ice-phase by the cloud phase classifier (Shupe et al., 2005(Shupe et al., , 2006. The uncertainty in IWC retrieval, as large as a factor of two (Shupe et al., 2005(Shupe et al., , 2006 results from a combination of systematic and random errors. The IWP was in the range of 10 g m −2 in the morning and varied over a wide range until 12 UTC. After 12 UTC the IWP was around or even below 5 g m −2 (Fig. 2).
Mixed-phase stratiform clouds often tend to be decoupled from surface layer turbulence by a statically stable layer. During 5 ASCOS, low-level mixed-phased clouds were decoupled from the surface about 75 % of the time (Shupe et al., 2013;Sedlar and Shupe, 2014;Sotiropoulou et al., 2014). The cloud layer shown in Fig. 1 was decoupled during the 8 h period of interest (Shupe et al., 2013). A CCN counter fed from an inlet on Oden approximately 20 m above the surface measured a mean CCN concentration of about 25 cm −3 at a supersaturation of 0.2 % during the time period of the ice drift (12 August to 2 September 2008) (Martin et al., 2011). During the evening of 31 August 2008, CCN concentrations at the surface dropped below 1 cm −3 10 around the time that the cloud began to dissipate (Mauritsen et al., 2011;Leck and Svensson, 2015). It is important to understand that since the cloud layer was decoupled, at least initially, we do not know how representative these values are for the cloud layer.

Model description and setup
The COSMO model was used in a semi-idealized LES setup with periodic boundary conditions similar to Paukert and Hoose The limitation is that the temporal evolution of a cloud layer consuming CCN by aerosol processing and scavenging can not be captured. The model surface is set as sea ice, with an albedo of 70 %, consistent with observations. The surface temperature 30 was set to 271.35 K. Changes in surface properties can influence the air-sea interactions and are taken into account by a sea ice scheme (Mironov, 2008). Cloud ice processes were turned off during the initial 2 h of the simulation, in order to permit the liquid cloud layer to develop. The initial temperature and moisture profiles were taken from a radio sounding at 5:35 UTC on was divided into one control simulation and three sets of sensitivity experiments (Tab. 1). The initial profiles were the same in all simulations except for the moisture profiles of sensitivity experiment SensMoist (see below).

Control simulation
The control simulation has a CDNC of 30 cm −3 and a fixed ice crystal concentration of 0.2 l −1 ; these values are chosen based 5 on the mean values during ASCOS reported in Section 2. Observations at the surface did not record any IN concentrations because they were below the detection limit of the instrument, which ranges between 0.1 l −1 and 2 l −1 (Z. Kanji, personal communication). However, the fact that clouds during ASCOS precipitated predominantly ice crystals implies that IN must have been present (Shupe et al., 2013). It is possible that advection with or without entrainment of IN at cloud top rather than surface sources provided IN for the observed cloud. An earlier field campaign with Oden during September 1991 measured 10 a maximum ice-forming nuclei concentration of 0.25 l −1 at 88 • N (Bigg, 1996). Guided by these findings, the ice crystal concentration in the model was set to be 0.2 l −1 in the control simulation.

Sensitivity experiments
The first sensitivity experiment (SensMoist) includes 4 simulations where the moisture profile is changed either below the cloud base (sub-cloud layer) or above the cloud top in order to mimic the effect of dry-air advection (Tab. 1). Below cloud, the 15 moisture profile is linearly dried to resemble 99 % relative humidity (RH) at cloud base decreasing to 85 % RH at the surface ( Fig. 3, a), while keeping the temperature profile the same as in the control simulation. Above cloud top, a 450 m deep layer of the atmosphere is progressively dried in 3 different simulations corresponding to RH values of 36 %, 20 % and finally 10 % above and in contact with cloud top (Fig. 3, b-d).
In order to investigate the sensitivity of the modeled cloud to changes in ice crystal concentrations, the ice crystal concentration 20 was increased to values well above the expected low values in the Arctic in the second set of sensitivity experiments (SensIce).
Two simulations with ice crystal concentrations set to 1 l −1 and 10 l −1 were conducted.
The third sensitivity experiment (SensCDNC) considers the low CCN concentrations observed during the ASCOS field campaign. During the 31 August 2008, CCN concentrations at the surface dropped below 1 cm −3 (Mauritsen et al., 2011;Leck and Svensson, 2015). The CDNC was decreased to 2 cm −3 and 10 cm −3 , respectively, in two simulations to investigate the impact 25 of low CDNC on the mixed-phase cloud development.

Control simulation
The initial θ-profile shows a neutral to stable boundary layer (Fig. 4). A small inversion is seen in both θ and θ e profiles near 300 m; this is the decoupling inversion, separating turbulence driven by the cloud layer from surface-driven turbulence. After 30 this is also where the cloud top is located (Fig. 5). The sensible heat flux and the latent heat flux are weak, because the surface is covered with ice. The observed values of both of these fluxes are small, but positive (Sedlar et al., 2011). Thus, no strong influence of the surface on the cloud is expected (Fig. 6). The simulated cloud top corresponds well with the cloud top seen by the MMCR at around 1 km (Fig. 1, b). The cloud base, measured with a laser ceilometer, is between 600 m and 700 m at 5 the beginning of 31 August 2008. This altitude agrees well with the cloud base height of the simulated cloud layer, which is around 600 m (Fig. 5). The boundary layer deepens over the next 8 h, causing the main inversion and the cloud top to rise by around 90 m (Fig. 5). The θ profiles imply that the lower half of the boundary layer transitions towards less stable and hence the decoupling inversion around 300 m disappears after 4 h (Fig. 4). The boundary and cloud layers thus quickly become coupled in the simulations. This tendency to erode cloud decoupling is common in LES simulations (Savre et al., 2014).

10
The maximum IWC in the control simulation is two orders of magnitude smaller than the liquid water content (LWC), around 0.0015 g kg −1 (Fig. 5). After 3 h the ice particles grow and start to sediment. The ice crystals fall through the sub-cloud layer and reach the surface (Fig. 5, red). At the same time a secondary cloud layer briefly forms at the decoupling inversion, likely associated with a moistening of the sub-cloud layer through ice crystal sublimation. Rain (rain water content (RWC)) precipitates out of the liquid layer after around 7 h but does not reach the surface due to evaporation and conversion to ice (Fig. 5, green).

15
After 4 h, when ice formation is relatively constant, the control simulation develops a liquid cloudy layer that is persistent throughout the simulation with a thickness of approximately 200 m (Fig. 5, blue). The maximum LWC in the cloud is around Observations of the LWP show a more variable LWP in the morning than in the afternoon (Fig. 2). The simulated LWP is around 50 g m −2 and most of the time in the range of the observed LWP, which has an error of 25 g m −2 . Because the simu-20 lated cloud is not dissipating in the control simulation, the simulated LWP remains in that range and is not decreasing during the day. The IWP of the control simulation seems to be at the lower end of the observed IWP range and reaches only around 2 g m −2 after the ice processes are turned on.

Sensitivity experiment -SensMoist
The availability of moisture above and below the cloud is an important ingredient for the persistence of an Arctic mixed-phase 25 cloud. Fig. 7 shows the evolution of LWP in the SensMoist experiment (pink lines). Reducing the available moisture in the atmosphere below the cloud does not change the persistence of the cloud. Up to 8 h, the LWP of the simulation with reduced moisture below the cloud is slightly smaller (by approximately 8 g m −2 compared to the control simulation after 4 h) than in the control simulation and at the end of the simulation, the LWP is almost the same (Fig. 7, pink solid line). The mean profiles of the water vapor (QV) after 5 h of simulation show that the difference in moisture is small near the surface between the different 30 simulations indicating a strong mixing in the sub-cloud layer (Fig. 8). A strong difference in QV is only seen above the cloud top and between the control simulation and the two SensMoist simulations with reduced RH. This suggests that the supply of moisture from near the surface has only a limited influence on the cloud layer, resulting in a stable LWP around 50 g m −2 .
Imposing a region of dry-air above the cloud has a larger influence on the cloud evolution. Drier air above the cloud layer leads to a decrease in LWP in all three simulations (Fig. 7, dashed pink lines). The reduction is strongest when RH above the cloud was reduced to 10 %. The LWC is reduced by almost a factor of 2 compared to the control simulation (Fig. 9). When the source of moisture from above is decreased, the boundary layer and cloud layer become coupled between 2 h and 4 h, which is similar to the control simulation (Fig. 10).
The θ e profiles also show a clear weakening of the inversion after 2 h which is due to the thinner cloud layer and consequently 5 decreased turbulence (Fig. 10). Hence the boundary layer cannot grow with time as it does in the control simulation (Fig. 5,   9). Following the reduction in LWC, IWC is also reduced relative to the control simulation; the mass of the liquid droplets is decreased and therefore ice crystals grow less rapidly. This causes the ice crystals to remain suspended in the atmosphere longer due to their reduced size and fall speed (9, red). These results examining the sensitivity of cloud to the moisture profile changes agree with the behavior of the Arctic mixed-phase cloud as reported in Solomon et al. (2013).

Sensitivity experiment -SensIce
In the simulation with an increased ice crystal concentration to 1 l −1 , the cloud still persists over the simulation time, and IWC increases because of the large number of ice crystals (Fig. 11). The impact on the liquid layer is however marginal. The LWP is almost constant at around 50 g m −2 , very similar to the LWP evolution simulated when RH is reduced in the sub-cloud layer (Fig. 7). Further increasing the ice crystal number to 10 l −1 leads to glaciation and finally to dissipation of the cloud after 6 h 15 (Fig. 7, blue dashed line). In this simulation, the inversion near cloud top becomes weaker after 8 h and the weak stable layer near 300 m erodes more rapidly than in the control simulation (Fig. 12).

Sensitivity experiment -SensCDNC
When CDNC is reduced relative to the reference value in the control simulation, the LWP time series shows a decrease to around 40 g m −2 with the CDNC 10 cm −3 , and to below 10 g m −2 for CDNC set to 2 cm −3 (Fig. 7, black lines). The reduction 20 in CDNC also leads to a weakening of the inversion around 1 km, while the inversion near 300 m persists throughout the simulation duration, whereas it is eroded after roughly 4 h in the control simulation (Fig. 13). The weakening of the main inversion is likely due to less radiative cooling at the cloud top, because of the optically thinner, less opaque liquid layer due to the lower CDNC. This also decreases the cloud overturning circulation which in turn slightly strengthens the decoupling inversion. The cloud-top radiative cooling is reduced, and subsequently the cloud-driven circulation is unable to sufficiently 25 penetrate the static stable layer near 300 m. With an optically thinner cloud above, the sub-cloud layer can cool more efficiently, and this promotes the formation of secondary, thin liquid layer in the vicinity of the lower temperature inversion near 300 m (Fig. 14). Rain forms after 2 h from initialization, through collision and coalescence processes. Rain from the main cloud layer can moisten the sub-cloud layer due to evaporation until the cloud layer at 1 km almost dissipates. This simulation leads to a very thin cloud with LWC values reaching 0.03 g kg −1 and maximum values of IWC of 0.0015 g kg −1 close to the surface. Ice 30 crystals falling from the upper cloud layer pass through the lower liquid layer around 3 h simulation time, where they grow at the expense of cloud droplets, resulting in IWCs as large as the control simulation. This also causes the second, lower liquid cloud to become tenuous and briefly intermittent (Fig. 14).
Low aerosol concentrations are common in the high Arctic due to a lack of aerosol sources in this region in particular during summer (Bigg, 1996;Heintzenberg et al., 2006;Garrett et al., 2010;Heintzenberg and Leck, 2012). In persistent precipitating boundary layer clouds, the aerosol concentration can be further reduced through scavenging. Thus, changes in aerosol concentrations and consequently CDNCs may strongly influence the lifetime and development of an Arctic mixed-phase cloud. Our 5 current model study of an observed mixed-phase cloud during the ASCOS field campaign shows that a CDNC concentration of 10 cm −3 is sufficient to sustain the cloud while a CDNC of 2 cm −3 leads to cloud dissipation. The COSMO simulated cloud was also sensitive to changes in the moisture profile. Generally, LWC decreased when RH in the 30 atmospheric layer above the cloud top was decreased. This supports observational and modeling evidence suggesting that the source of water vapor above cloud top is important for the persistence of the liquid layer (Solomon et al., 2011;Sedlar et al., 2012;Morrison et al., 2012). However, in our simulations, introducing a dry layer above the inversion did not cause cloud dissipation. Reducing RH in the sub-cloud layer had only a modest impact on the mixed-phase cloud. Mixed-phase clouds in the Arctic are frequently decoupled from the surface (Sedlar and Shupe, 2014;Sotiropoulou et al., 2014) and therefore do not necessarily rely on a moisture source near the surface to persist.
Increasing the ice crystal concentration to 1 l −1 had a moderate influence on the simulated mixed-phase cloud, while an even higher ice crystal concentration of 10 l −1 led to glaciation and subsequent dissipation of the cloud. Rogers et al. (2001) found that for thin, low-level stratus clouds, the IN concentration at -15 to -20 • C was around 1 l −1 . Nevertheless, ice crystal concen-5 trations in the Arctic may vary over 3 orders of magnitude (Morrison et al., 2005) and a maximum ice crystal concentration of 0.25 l −1 has been observed in a similar season and geographic region as ASCOS (Bigg, 1996). Considering IN concentration of 0.25 l −1 or lower from a past field campaign in the high Arctic (Bigg, 1996) and taking the absence of IN measurements above the instrument detection limit during ASCOS into account, such a large increase in ice crystal number concentration, seems an unlikely mechanism responsible for the observed cloud dissipation during ASCOS. Hence, these results suggest that 10 reasonable increases in IN concentrations are not the primary mechanism leading to cloud dissipation for this observed case.
The sensitivity experiments tested here, altering CDNC, ice crystal number concentration, and changing moisture sources to the cloud layer, were designed to mimic changes in the large-scale circulation and advection of air masses with different thermodynamic profiles and aerosol properties. In reality, it is likely that changes in thermodynamical properties and aerosol will happen simultaneously, and that the combination of these processes will control the evolution of the mixed-phase cloud (Ka-15 lesse et al., 2016). Nevertheless, we have shown that, independently, dry-air advection above cloud top, ice crystal increase, and CDNC reduction all contribute to a reduction of the liquid condensate layer of a mixed-phased cloud. However, we find that the reduction of CDNC was likely the primary contributor to the dissipation of the observed mixed-phase cloud during this specific case. Morrison, H., Zuidema, P., Ackerman, A. S., Avramov, A., de Boer, G., Fan, J., Fridlind, A. M., Hashino, T., Harrington, J. Y., Luo, Y., Ovchinnikov, M., and Shipway, B.: Intercomparison of cloud model simulations of Arctic mixed-phase boundary layer clouds observed during SHEBA/FIRE-ACE, J. Adv. Model. Earth Syst., 3, M05 001-, 2011. Morrison, H., de Boer, G., Feingold, G., Harrington, J., Shupe, M. D., and Sulia, K.: Resilience of persistent Arctic mixed-phase clouds, Nat. Geosci., 5, 11-17, doi:10.1038/NGEO1332, 2012 interactions over the Arctic sea ice in late summer, Atmos. Chem. Phys., 13, 9379-9399, doi:10.5194/acp-13-9379-2013, 2013.      Figure 7. The domain-averaged LWP (including cloud droplets and rain) for the control simulation (red), the simulation of dry-air advection below the cloud (pink solid), the simulations of dry-air advection above the cloud top with a RH of 36 % (pink dashed), a RH of 20 % (pink dotted) and a RH of 10 % (pink dash-dotted), the simulations with an ice crystal concentration of 1 l −1 (blue solid) and 10 l −1 (blue dashed), the simulation with a CDNC of 10 cm −3 (black dashed) and a CDNC of 2 cm −3 (black solid).