Coupled mode of cloud, atmospheric circulation, and sea ice controlled by wave-3 pattern in Antarctic winter

This study examines coupled relationships among clouds, atmospheric circulation, and sea ice in Antarctic winter. We find that the wave-3 pattern dominates the leading covariability mode among cloud, atmospheric circulation, and sea ice. Both horizontal transport and vertical motion contribute to cloud formation, resulting in maximum cloud anomalies spatially between maximum meridional wind and pressure anomalies in the coupled system. The radiative effect of the clouds related to the wave-3 pattern can generate sea ice anomalies up to 12 cm thick in one month in the Amundsen Sea. It also strengthens the sea ice anomalies that are directly induced by low-level atmospheric circulation anomalies. In addition, the radiative forcing of the leading cloud mode in the lower troposphere is suppressed by the dynamic and thermodynamic effects of the circulation anomalies. These discoveries provide a better understanding of Antarctica’s interactive processes, and also offer physical evidence for climate model validations.


Introduction
Antarctic sea ice extent (SIE) dropped to a record low in 2017 after a gradual increase over more than three decades (Parkinson 2019). There were also significantly larger fluctuations after 2011 than in earlier decades (Meehl et al 2019). The SIE's counter-intuitive increase in the warming climate followed by its sudden reduction in recent years has been the subject of many studies. Moreover, CMIP5 models significantly overestimate the natural variability of Antarctic SIE after the long-term trends are removed (Zunz et al 2013), which is due to a lack of in-depth knowledge of this highly interactive climate system in the Antarctic. Sea ice is a highly active component in the polar climate system, contributing to the complexity of simulating the Earth's system. The variation in SIE is associated with many processes in the ocean and atmosphere (Comiso 2001, Kwok et al 2013, Liu et al 2016, Long and Perrie 2017. Here, we focus on the covariability among clouds, atmospheric circulation, and sea ice in a coupled system. Clouds play a critical role in the global climate system because they influence the radiation balance (Wang and Key 2005, Li et al 2011, Cox et al 2015, Tricht et al 2016. Clouds force sea ice in two ways: they promote sea ice growth by blocking shortwave radiation and cooling the surface; and they reduce sea ice growth by emitting longwave radiation and warming the surface (Wang and Key 2003, Graversen et al 2008, Lee 2011, Bennartz et al 2013, Lee et al 2017. In polar winter, interference from clouds with shortwave radiation is minimal, so their longwave radiation dominates the clouds' influence on the surface. The unique Antarctic environment, with extremely low temperature, low moisture, and low aerosol concentration, makes cloud formation different from that in the other regions, including the Arctic (Bromwich et al 2012). Also, the low temperature and bright ice surface limit cloud detection by passive microwave techniques, resulting in a high noise level in satellite observations (Liu et al 2010). Moreover, current stateof-the-art climate models have tremendous intermodel discrepancies in cloud simulations, reflected by their annual and seasonal means of total cloud fractions (Klein et al 2013, Vignesh et al 2020. The lack of knowledge of cloud processes in the extreme polar environment (Bromwich et al 2012, Scott and Lubin 2016, Scott et al 2017, Dong 2018 introduces errors in simulations of the atmosphere and sea ice interaction in climate models. Clouds actively engage in the coupled polar climate system. A case study by Wang et al (2019b) suggested that the negative downward longwave radiative forcing (surface heat loss) from clouds alone in winter 2011 was capable of growing an additional 30 cm of sea ice in the Weddell Sea. That is a substantial impact given that winter Antarctic sea ice thickness (SIT) is only about 70-100 cm on average (Worby et al 2008, Zwally et al 2008, Kurtz and Markus 2012. In this study, we focus on the low-to-mid troposphere, and reveal a coupled mode in clouds, atmospheric circulation, and sea ice in Antarctic winter. We also address cloud formation processes associated with the coupled mode. We analyze reanalysis and satellite data during the austral winter (June, July, and August) when the shortwave radiation in the study area is at its annual minimum.

Data
By comparing with satellite observations, previous studies show that NASA's Modern-Era Retrospective analysis for Research and Applications reanalysis-2 (MERRA-2, GMAO 2015) is capable of capturing anomaly patterns of cloud-fraction and cloud-radiation physics in polar regions (Hinkelman 2019, Wang et al 2019a, 2019b. The MERRA-2 reanalysis covers the period from 1980 to present at a resolution of 0.5 • latitude by 0.625 • longitude. The reanalysis uses the updated Goddard Earth Observing System Model (GEOS-5) and a new analysis scheme (Rienecker et al 2011, Molod et al 2015. The 40 years of cloud data from MERRA-2 have advantages over satellite observations since longer time series allow the multivariate empirical orthogonal functions (MEOFs) analysis to capture more reliable modes. Based on the observations at the Barrow and Eureka observation sites in the Arctic and space radar-lidar observations, Liu et al (2017) revealed that clouds are mainly distributed below 400 hPa. Therefore, we calculated cloud fractions at two levels: low-level cloud fraction (C low ) occurring below 700 hPa, and mid-level cloud fraction (C mid ) occurring between 700 and 400 hPa. In addition, we used the monthly surface downward longwave radiative fluxes (SLRFs) for all-sky conditions and cloudfree conditions from the MERRA-2 to calculate the cloud radiative forcing on the Earth's surface.
To verify the reliability of MERRA-2 cloud modes, we also used multi-level cloud data from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) active satellite observations (Winker et al 2003) for the period from 2006 to 2019, and the total cloud cover data from the International Satellite Cloud Climatology Project (ISCCP-H) for the period from 1983 to 2017.
The daily and monthly mean sea ice concentration (SIC), generated using a bootstrap algorithm, is obtained from the National Snow and Ice Data Center (Comiso 2017). The monthly sea-surface temperature (SST) from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) (Rayner et al 2003) are used to show the potential effect of the ocean on sea ice. The SIT is from the Global Ice-Ocean Modeling and Assimilation System (GIOMAS) Data, which is based on the parallel ocean model coupling with a 12-category thickness and enthalpy distribution (TED) ice model (Zhang and Rothrock 2003). The TED model simulates sea-ice ridging processes explicitly following Hibler (1980) and Thorndike et al (1975). The atmospheric variables consist of geopotential height (GPH), wind, and vertical pressure velocity at 850 and 500 hPa, and sensible and latent heat fluxes at the surface. These atmospheric variables are from the latest reanalysis of the European Center for Medium-Range Weather Forecasts, namely, ERA5 (Tetzner et al 2019, Dong et al 2020 with a grid of 0.25 • × 0.25 • and from the MERRA-2, to represent the low-and mid-level atmospheric circulations. Since the results from the two datasets are consistent, we only show the results from the ERA5 in this paper. Note that the daily data are only used in a lagged correlation and all other results are calculated from the monthly time series.

Methods
Monthly anomalies are calculated by removing the climatology at each variable's grid point. Winter time series consist of monthly anomalies in June, July, and August of all years. For SIC, open water, land, and grids with SIC below 15% were masked and excluded from all calculations. All anomaly fields were areaweighted. Here, we use the MEOF analysis (see text S1 in the supporting information available online at stacks.iop.org/ERL/17/044053/mmedia) to identify covariability in the fields of cloud fraction, GPH, wind vector, and SIC. The maximum covariance analysis (MCA, Mo 2003) (see text S1) is also carried out to investigate the correlation between two variable fields. Lagged correlations are further used to assess causality among atmospheric circulation, cloud, and sea ice.
To confirm the relationship between cloud and SIC revealed by the MEOF, we first employ a linear correlation between total cloud and SIC anomaly time series generated by averaging both data sets in a 5 • longitude bin and from the coast to the ice edge in latitude. A temporal significance test and a spatial significance test are then used to assess linear correlation results. The methods of temporal and spatial significance tests are documented in text S1. We also quantify the effects of clouds on sea ice by using a sea ice growth equation. Because of the absence of sunlight in winter, Antarctic cloud forcing is dominated by the longwave radiative effect. So, the cloud radiative effect is the surface downward longwave radiation flux difference between all-sky and cloud-free conditions, as follows, where Q represents the SLRF in cloud conditions, F all and F clear represent the surface downward longwave cloud radiative fluxes for all-sky and cloud-free conditions, respectively. A simple sea ice growth equation (Thorndike 1992, Eisenman 2012) is used to quantify the cloud radiative effect on sea ice: where △h is sea ice growth, △t is time step, ρ is the density of sea ice (914 kg m −3 ), and L is the latent heat of fusion of sea ice (333.4 kJ kg −1 ). ε is a measure of efficiency, γ is the thermodynamic scale thickness (0.7 m; (Eisenman 2012), h is the initial SIT. Thus, we estimate sea ice growth controlled by cloud forcing via equation (2).

Relationships between low-level cloud, atmospheric circulation and sea ice
The MEOF analysis of C low , SIC, and 850 hPa GPH reveals that both the Southern Annular Mode (SAM) (Thompson and Wallace 2000) and wave-3 (Raphael 2004) stand out in the leading pattern of the atmospheric circulation (figure 1(b)). The SAM and wave-3 pattern are dominant climate modes in the Southern Ocean, significantly influencing sea ice (Yuan andLi 2008, Eayrs et al 2021). The profound anomalous circulation in the Southeast Pacific is closely related to the Amundsen Sea low (ASL) , Raphael et al 2016, which produced a robust Antarctic dipole pattern (ADP) in sea ice (Yuan 2004, Yuan andMartinson 2000). ASL has deepened over the last half-century due to ozone depletion (England et al 2016). ASL and its deepening are closely tied to ENSO variability (Yuan 2004, Clem et al 2017, Yuan et al 2018. Although the wave-3 pattern dominates the circulation, it does not stand out in the cloud distribution ( figure 1(a)). Furthermore, the phases of C low and SIC anomalies coincide in space, except over the eastern Indian Ocean (figures 1(a) and (c)). This character of the relationship in the first MEOF mode is also reflected in the full fields. The colocated correlation between the total C low and SIC anomalies around the Antarctic confirms the similar and statistically significant result: more clouds correspond to more sea ice (figure S1), which is counterintuitive. Generally, we expect that more clouds emit more longwave radiation to the sea surface, resulting in less sea ice (as is the case in the eastern Indian Ocean). Thus, this cloud mode is expected to result in an antiphase sea ice field rather than the pattern in figure 1(c). It suggests that the leading C low mode cannot be the determinative driver of the wave-3 pattern in sea ice. Figure 1 shows that SIC anomalies correspond well to the atmospheric circulation in winter. Driven by the cold southerly wind, sea ice tends to move equatorward, and new ice forms in leads and coastal polynyas, replenishing open waters. That is conducive to the sea ice edge expansion dynamically and thermodynamically in the western Indian Ocean, western Pacific, and the Amundsen Sea. In contrast, sea ice is compacted by the northerly wind, and ice formation is discouraged by warm advection in the Weddell Sea, the western Ross Sea, and the eastern Indian Ocean. The red and blue arrows in figure 1 schematically illustrate these physical processes. Evidently, atmospheric circulation and sea ice anomalies are highly coupled by the wave-3 pattern, as suggested by early studies (Raphael 2004, Yuan and Li 2008, Renwick et al 2012. The moisture transported poleward would condense into clouds when encountering the cold polar environment, as positive cloud anomalies appear over the Indian Ocean. On the other hand, negative cloud anomalies associated with the equatorward flow of cold and dry Antarctic air occur in the western Pacific. However, the cloud anomalies in the other regions of the Antarctic do not correspond well with the moisture transport process associated with the wave-3 pattern. Other processes must play a dominant role in the cloud distribution of the lower atmosphere. For example, the cloud formation due to orography does respond to the circulation change but does not necessarily contribute to the coupled wave-3 pattern in the leading mode. When the Andes and Antarctic Peninsula obstruct the anomalous easterly wind, uplifted moisture forms more clouds on the mountain ridges' west side. The lee-side sinking air leads to negative cloud anomalies over the Weddell Sea, which may overcome the poleward moisture transport, as marked by the red arrow in figure 1(a). Other complex processes may also be involved in determining the low-atmosphere's cloud distribution, which is not the focus of this study. Thus, the low-level atmospheric circulation anomalies likely drive the sea ice anomalies, suppressing the radiative effect of the leading C low mode. This mode accounts for 10% of the total variance. The red (blue) arrow represents the warm (cold) advection driven by pressure anomalies.

The coupled mode of mid-level clouds and atmospheric circulation
The MEOF analysis of C mid , SIC, and 500 hPa atmospheric variables, including GPH, wind, and vertical velocity, shows that the wave-3 pattern dominates the anomalous circulation in the first mode (figure 2). It is worth noting that the C mid and SIC patterns are almost unchanged after reducing the circulation weight in the MEOF analysis by removing the wind vector from the MEOF (not shown). The consistent circulation patterns in figures 1 and 2 suggest the barotropic nature of the wave-3 mode in Antarctic winter. So, the SIC patterns associated with pressure anomalies at the low-and mid-levels are similar.
Unlike C low , the C mid leading mode shows a more intuitive relationship with sea ice. The positive (negative) C mid anomalies correspond well to the negative (positive) SIC anomalies around the Antarctic (figures 2(a) and (d)). The positive cloud anomalies emit more longwave radiation to the surface, increase surface net radiation flux, and hinder sea ice growth, resulting in negative sea ice anomaly. The negative cloud anomalies decrease surface net radiation flux in winter, and lead to more heat loss from the surface, promoting the sea ice formation in the Amundsen Sea, western Pacific, and the Indian Ocean. Simple linear regression between C mid and SIC, which represents full anomalous fields instead of the leading MEOF mode, also shows the same result in these areas: more clouds correspond to less sea ice ( figure S2).
In addition, the maximum cloud anomalies do not occur in the areas of maximum northerly wind anomalies, where moisture is transported into the polar region. Instead, cloud anomalies are shifted to the west by 5 • -15 • longitude relative to northerly maxima, especially in the area north of 65 • S. This indicates other processes must be involved in addition to horizontal moisture transport. Furthermore, the maximum cloud anomalies also do not occur in the areas of low-pressure centers where the clouds can form due to convection. On average, cloud anomaly centers shift eastward by 15 • -25 • relative to the lowpressure centers.
To illustrate the relationships among clouds, anomalous vertical and horizontal air motions, we superimpose the results of circulation and clouds to form a 3D schematic circulation (figure 2(f)). It shows that maximum cloud anomalies are formed between the low-pressure centers and maximum northerly wind, as a result of both cloud formation mechanisms: uplifted air and horizontal moisture transport. Similarly, the minimum cloud anomalies occur between the high-pressure center and maximum southerly wind. We conclude that both horizontal transport and vertical motion induced cloud formation processes contribute to the wave-3 C mid distribution, resulting in maximum mid-level cloud anomalies between meridional wind and pressure anomalies. Vertical motion mainly represents a local moisture source for clouds, while horizontal transport represents a remote moisture source from middle latitudes. Thus, the discovery provides a better understanding of the winter cloud distribution and moisture source in the Antarctic.

Impacts of cloud radiative forcing related to wave-3 on sea ice
We calculate the surface radiation budget controlled by cloud to isolate its radiative effect on sea ice in the coupled mode. We decompose the SLRF and C mid by MCA to isolate the SLRF pattern associated with C mid anomalies. The result shows that the first SLRF mode (figure S3) has a similar spatial distribution as the first C mid mode (figure S3), with a high temporal correlation of 0.9. So, the SLRF mode is considered the signal of dominant variability from cloud related to wave-3. In other words, the cloud radiative forcing associated with the wave-3 pattern, best represented in the leading mode of C mid , exists in the total cloud and is reflected in SLRF. Although the leading mode in figure 2 only accounts for 11%  of the total covariability in the coupled system, the SLRF related to wave-3 cloud pattern can account for 20%-30% of radiation variance in some key regions such as the Ross Sea and Bellingshausen Sea (figure S3(c)). We combine the eigenvector of the first SLRF mode and its principal component to reconstruct the associated cloud forcing anomalies for the period of 1980-2019. To highlight the potential influence of the SLRF mode on sea ice, we choose a strong sea ice anomaly case in August 2018 to show the contribution of cloud radiative forcing related to wave-3 pattern. Figure 3(a) shows that the cloud forcing anomaly is approximately −15 W m −2 over the Amundsen Sea, 12 W m −2 over the northern Ross Sea, and 9 W m −2 over the northern Weddell Sea. Overall, the magnitude of cloud forcing corresponding to the wave-3 pattern is approximately 25% of that of turbulent heat flux, which is mainly related to surface air temperature and wind and represents the thermodynamic forcing of the low-level atmospheric circulation (figures 3(a) and (d)).
Negative anomalies of surface cloud forcing (the ocean losing heat) favor sea ice growth. Here, we use equation (2) to estimate the change in SIT due to cloud radiative flux associated with the wave-3 pattern. The results show that the cloud radiative effect could generate sizable SIT anomalies in the Weddell Sea (−10 cm), in the Amundsen Sea (12 cm), and the northern Ross Sea (−12 cm). This radiative effect is relatively weak in the other areas but still creates 3-6 cm sea ice anomalies. The magnitude of sea ice growth resulting from cloud forcing is approximately 20% of that of SIT total anomaly (figures 3(b) and (e)). Furthermore, we calculated SIT anomalies due to the cloud forcing within the black box in the Amundsen Sea (shown in figure 3(c)) using the entire time series. The areal mean of SIT anomalies shows that the wave-3 associated cloud anomalies produce −12 to 12 cm of SIT anomalies in the winter months. The above results suggest that the winter cloud controlled by the wave-3 pattern has an indispensable radiative influence on SIC. The radiative effect would reenforce the sea ice variability forced by low-level atmospheric circulation since the sea ice growth is in accordance with the sea ice pattern in the coupled mode shown in figures 1(c) and 2(d).

Discussion
Although cloud in the coupled mode has a substantial influence on SIC, distinctions between the sea ice growth pattern controlled by cloud forcing and leading SIC pattern also exist in some areas, such as in the eastern Atlantic (0 • E-45 • E) (figures 2(d) and 3(b)). The correlation between total C mid and SIC anomalies is also not significant. A possible explanation is that the sea ice variability might be dominated by other factors, especially by interactions with the ocean in these regions. These sea ice variability drivers include SST (Kusahara et al 2017, Blanchard-Wrigglesworth et al 2021, Zhang et al 2021 and polynyas (Tamura et al 2008(Tamura et al , 2016. Figure S4 shows that SIC and SST are well correlated in the eastern Atlantic (r = −0.81) and SIC lags SST by about half a day. Statistically, this means SST can account for more than 66% of the total SIC variance in the area, playing a critical role in SIC variability. Besides, large amounts of sea ice are produced at Mertz Glacier Polynya (172 km 3 a −1 ) (Tamura et al 2016), which is also an indispensable factor in sea ice variability. Despite the abovementioned processes, the clouds' radiative forcing on sea ice stands out significantly in the ADP regions and the Ross Sea. Moreover, sea ice anomalies could feed back positively to cloud anomalies through modifying evaporation, which needs to be addressed in future studies.
We also conducted an MCA between C mid and SIC, 850 hPa GPH and SIC, and 500 hPa GPH and C mid to show the direct connections between these paired fields. The results show that the spatial patterns of the leading MCA mode are similar to those of MEOF (figures 2 and S5). For example, the positive (negative) C mid anomalies correspond well to the negative (positive) SIC anomalies around the Antarctic (figures S5(a) and (d)). This mode can account for 24% of the total squared covariance between SIC and C mid . Heterogeneous correlations between the C mid time series of the first MCA mode and total SIC anomalies range −0.3 to −0.5 in the Amundsen Sea, eastern Weddell Gyre, and western Pacific. Thus these analyses support MEOF results and suggest that the method used here is reliable.
For the cloud data used in this paper, we admit that the MERRA-2 cloud has its own limitations. Previous studies showed that the MERRA-2 underestimates near-surface clouds (Rozenhaimer et al 2018) and its mean bias is −2.73% over the Arctic compared with NASA CERES-MODIS (Huang et al 2017). To verify whether the cloud pattern in the coupled mode derived from the MERRA-2 data (figure 2(a)) represents a real pattern in observations, we use the MCA to analyze the similarities between the clouds from MERRA-2 and satellite observations from CALIPSO (figure S6). The results show that the wave-3 pattern appears in the leading modes of both cloud products, although the anomalies are weaker in CALIPSO compared to MERRA-2. The leading patterns of C low , C mid , and C high from MCA between two cloud datasets accounted for 24%, 19%, and 32% of the total squared covariance respectively, and the associated time series were correlated at 0.84, 0.94, and 0.97 respectively. C high shows the highest consistency between the two data sets. C high here is just for the verification of the MERRA-2 cloud because the high-level clouds are thin and have a limited impact on sea ice (Liu et al 2017). It indicates that the leading cloud modes calculated from MERRA-2 at different levels are the real patterns in satellite observations.
We also use the total cloud fraction of satellite data ISCCP to extract the cloud fraction modes coupled with 500 hPa meridional wind using the MEOF analysis. The result shows that the MERRA-2 C mid anomaly pattern (figure 2(a)) mostly resembles the ISCCP second mode of the MEOF (figure S7), although there are some regional differences. Considering significant differences in cloud data sets from different sources (Bromwich et al 2012), the consistent cloud patterns between MERRA-2 and two satellite data sets suggest that the wave-3 pattern in C mid is a predominant mode in the total cloud and C mid and the MERRA-2 is capable of capturing it.
To examine possible causality relationships among investigated parameters, daily SIC anomalies, C mid anomalies, and the PC of the first 850 hPa GPH EOF mode are used in the lagged correlation. The PC represents the low-level atmospheric circulation variabilities associated with the wave-3 pattern. Although Correlations as a function of lead and lag days between daily SIC total anomalies and the principal component (PC) of the first 850 hPa GPH EOF mode (black), SIC and C mid total anomalies (red), and C mid and the PC of the first 850 hPa GPH EOF mode (green). The circles represent confidence levels above 95%. Negative lags represent the first variable of each pair marked in figure lagging. Here we consider the time series of the first 850 hPa GPH EOF mode as the low-level atmospheric circulation daily variabilities that are associated with the wave-3 pattern. the circulation and cloud anomalies occur simultaneously, the circulation leads SIC by 3 d while the cloud leads SIC by only 1 d, suggested by the maximum correlations (figure 4). It suggests that circulation and clouds drove sea ice anomalies independently, and cloud radiative forcing related to wave-3 has its contributions to SIC variabilities.

Summary and conclusions
We found a well-defined coupled mode associated with the wave-3 pattern in winter Antarctic atmospheric circulation, cloud, and sea ice. In the coupled mode, both horizontal and vertical moisture transports contribute to the cloud's wave-3 distribution at the troposphere mid-level, indicating that the moisture sources of cloud come from both local and remote regions. The sea ice variability in winter is driven by both low-level atmospheric circulation and cloud radiative forcing related to the wave-3 pattern. The atmospheric circulation controls C mid distribution. In turn, cloud radiative forcing further strengthens the sea ice anomalies generated by the low-level atmospheric circulation. However, the leading C low mode dominated by orography effects and atmospheric circulation anomalies does not match the wave-3 pattern, of which the radiative forcing on sea ice is suppressed by the direct dynamic and thermodynamic forcing of the circulation.
The cloud radiative forcing associated with the wave-3 pattern, best represented in the leading mode of C mid , exists in the total cloud and is reflected in SLRF. Despite the fact that the leading MEOF mode only accounts for 11% of the total variance in the coupled system of mid-level circulation, cloud, and sea ice, the total cloud radiative forcing related to the wave-3 pattern can account for 20%-30% of radiation variance in some key regions such as the Ross Sea and Bellingshausen Sea. This cloud radiative forcing can contribute −15 W m −2 to the surface radiation budget in the Amundsen Sea, 12 W m −2 in the northern Ross Sea, and 9 W m −2 in the northern Weddell Sea in August 2018, producing up to −12, 12, and 10 cm SIT anomalies, respectively, in these regions, which accounts for approximately 20% of the SIT total anomaly. This study suggests that the cloud radiative effect on sea ice is not an isolated event as presented in the case study by Wang et al (2019b), but plays a nonnegligible role in sea ice variability consistently.
In addition to the wave-3 pattern, SIC variability can be attributed to other climate modes, such as SAM, ENSO, and ASL, and many factors, such as ocean temperature, ocean currents, ice drift, and polynyas. Here, we only focused on the coupled mode associated with the wave-3 pattern, which yields a better understanding of clouds' roles in the Antarctic climate system and provides evidence for climate model validation. The study also reveals the importance of understanding cloud processes in the polar environment for improving Earth system models in the future.

Data availability statement
No new data were created or analyzed in this study.