Atmospheric controls on the Terra Nova Bay polynya occurrence in Antarctica

Polynyas, or ice-free regions within the sea ice pack, are a common occurrence around Antarctica. A recurrent and often large polynya is the Terra Nova Bay Polynya (TNBP), located on the western side of the Ross Sea just off Victoria Land. In this study, we investigate the atmospheric conditions leading to the occurrence of the TNBP and its spatial variability, as estimated using satellite-derived ice surface temperature and sea ice concentration data. A cluster analysis revealed that katabatic winds descending the Transantarctic Mountains, account for about 45% of the days when the TNBP exceeded its 2010–2017 mean extent plus one standard deviation. Warmer and more moist air intrusions from lower-latitudes from the Pacific Ocean, which are favoured in the negative phase of the Southern Annular Mode, play a role in its expansion in the remaining days. This is more frequent in the transition seasons, when such events are more likely to reach Antarctica and contribute to the occurrence and the widening of the polynya. In-situ weather data confirmed the effects of the mid-latitude air intrusions, while sea ice drifts of up to 25 km day−1 cleared the ice offshore and promoted the widening of the polynya starting from the coastal areas. Knowing the atmospheric factors involved in the occurrence of coastal polynyas around Antarctica is essential as it helps in improving their representation and predictability in climate models and hence advance the models’ capabilities in projecting Antarctic sea ice variability.


Introduction
Polynyas are areas of open water surrounded by sea ice.Two main mechanisms can lead to their formation (Smith and Barber 2007a, b;Bruneau et al. 2021): (i) upwelling of warm waters from below, which inhibits the formation of ice through above freezing temperatures and/or vertical mixing (known as sensible heat polynyas, thermodynamically driven), and (ii) persistent winds and/or ocean currents that drive the ice away from a fixed boundary such as a coastline (termed as latent heat polynyas).An example of a sensible heat polynya is the Maud Rise or Weddell Sea polynya, located near an area of elevated topography (Francis et al. 2019(Francis et al. , 2020;;Zhou et al. 2022).Such polynyas are normally observed in the open ocean away from coastlines, and their formation is fostered by (i) the presence of deep cyclones, as the clockwise (in the Southern Hemisphere) and anticlockwise (in the Northern Hemisphere) flow around its center promotes ice divergence and subsequently its break-up, and/ or (ii) atmospheric rivers (ARs), which bring the warmer and more moist tropical air into higher latitudes, helping to thin and eventually break the ice (Francis et al. 2020).Coastal polynyas, such as the Terra Nova Bay Polynya (TNBP) in the Ross Sea of Antarctica (Kurtz and Bromwich 1985;Fusco et al. 2002;Guest 2021a, b), are typically dynamically driven and are also recurring events, while open-water polynyas are much rarer, taking place at decadal or longer time-scales (e.g.Mohrmann et al. 2021).
Polynyas play an important role in the local, regional and potentially global meteorological and oceanic conditions: the strong temperature gradient between the exposed ocean surface and the underlying atmosphere, which can reach 40 °C (Wenta and Cassano 2020), leads to heat fluxes of the order of thousands of W m −2 (e.g.Budillon et al. 2000;Fusco et al. 2009;Guest 2021b), promoting convection (Rusciano et al. 2013), while the vertical mixing of the water impacts the circulation underneath (Budillon and Spezie 2000;Morales Maqueda et al. 2004;Fusco et al. 2009).Despite this, and the recent major improvements in model physics and dynamics (Sansiviero et al. 2017), polynyas continue to be poorly captured by numerical models (Mohrmann et al. 2021).Given the important role they play in atmospheric and ocean dynamics, which may be exacerbated in a warming world (e.g.Smith Jr. and Barber 2007a), an understanding of the mechanisms behind their formation is vital so as to better predict them in weather and climate models.
Through the analysis of infrared satellite images taken in the austral winter of 1979, Kurtz and Bromwich (1983) reported that the TNBP is quasi-periodic with a period of 15-20 days, and with an average area of 1000 km 2 that peaked at about 5000 km 2 .The TNBP is essentially driven by katabatic winds, closing off in the presence of strong and persistent easterly (onshore) winds, and opening up in a westerly (offshore) or weak easterly wind regime, with the adiabatic warming experienced by the downslope winds (e.g., Francis et al. 2023) further aiding its formation.A later study (Bromwich and Kurtz 1984) confirmed this, and stressed the role of the nearby Drygalski Ice Tongue in preventing sea ice from the south from coming into the bay which allows the polynya to form.As the air descends the steep slopes and encounters colder air below it flows horizontally, and may extend over a large distance offshore.For example, Parish and Bromwich (1989) found that katabatic winds originating from the Reeves Glacier upwind of the TNB were detected 250 km offshore in aircraft missions in early November 1987.Being a cold season event (sea ice is not present in the region in the warmer months), when cyclogenesis is more frequent in the Ross Sea (Simmonds et al. 2003), the TNBP area shows considerable seasonal variability (Aulicino et al. 2014(Aulicino et al. , 2018)), decreasing from April to August as the regional temperature decreases (Ding et al. 2020).The regular sea ice formation in the TNBP leads to brine rejection, with the salt left behind increasing the density of the water column, resulting in the formation of the High Salinity Shelf Water (HSSW), the densest water mass of the Southern Ocean, and contributing to the formation of the Antarctic Bottom Water (Kurtz and Bromwich 1985;Fusco et al. 2009;Budillon et al. 2011;Castagno et al. 2019).
An in-depth analysis of a katabatic wind event at the TNB is presented in Guest (2021a;b) using weather station measurements and those collected by a ship sailing away from the coast from which weather balloons were launched.Hurricane-force wind speeds of 35 m s −1 were observed at a weather station located at the coast on 08 May 2017, decreasing to 18 m s −1 around 100 km oceanward.The katabatic wind layer extended up to 700 m above the surface, comprising four temperature inversions and four peaks in wind speed.As the air moved over the polynya it warmed and moistened, with a 10°C and 0.5 g kg −1 increase in boundary layer air temperature and specific humidity, respectively, roughly 100 km downwind.The large air-sea temperature gradients led to sensible heat fluxes of up to 2000 W m −2 over the open water regions cleared of sea ice by the katabatic winds.Over pancake ice, the heat fluxes were roughly half as large, decreasing to 5% of that magnitude over snow-covered young ice flows.
Katabatic winds can be intensified through the interaction with synoptic or mesoscale cyclones, leading to strong offshore speeds and sea ice drift velocities, as noted by Wang et al. (2021) for coastal polynyas in East Antarctica.Besides synoptic-scale cyclones associated with mid-latitude storm tracks, mesocyclones are a regular occurrence around Antarctica, in particular in the Ross Sea/Ross Ice Shelf (Carrasco et al. 2003;Sinclair et al. 2010;Scarchilli et al. 2020), and can also intensify katabatic winds outbreaks.
While katabatic winds are the primary driver of the variability in the TNBP area (Wenta and Cassano 2020), they do not solely explain all TNBP occurrences.Wang et al. (2022) investigated the large-scale circulation in the Ross Sea in the austral spring season, and stressed the role the Amundsen Sea Low (ASL; Raphael et al. 2016), a semipermanent low pressure system located in the Amundsen-Bellingshausen Seas to the west of the Antarctic Peninsula, played in the atmospheric circulation in the region.Vagnoni et al. (2021), using back-trajectories, showed that, particularly in the warmer months, the air mass around Tasmania can be advected into the TNB.The work of Scarchilli et al. (2011) stressed that air masses from the East Pacific are also transported towards the TNB.The potential role of the advection of warmer and more moist low-latitude air from the Pacific Ocean to the TNBP has not been fully explored, which prevents a full understanding of the mechanisms behind its variability.Therefore, the objective of this study is to identify the atmospheric factors from local to regional scale leading to the occurrence and the widening of the polynya.This is achieved by analyzing satellite data in conjunction with atmospheric reanalysis and in-situ data.
This study is structured as follows.In Sect.2, the methodology used to estimate the TNBP area using satellite data and the reanalysis and observational datasets used in this work are described.In Sect.3, a climatology of the TNBP area is presented, with the dominant large-scale mechanisms behind its variability being the focus of Sect. 4. Two case studies are discussed in Sect.5: an exceptional case of katabatic winds, and an event for which the polynya widened also as a result of the advection of a lower latitude air mass into the region.The findings of this work are summarized in Sect.6.

Satellite data
The TNBP, which is characterized by a lack of sea ice in the TNB, is typically detected using changes in brightness temperature or ice surface temperature (IST).As noted by Ding et al. (2010), the 1 km × 1 km sea ice concentration (SIC) estimated by the ISTs measured by the Moderate Resolution Image Spectroradiometer (MODIS; Hall et al. 2004) instrument is more accurate than that estimated from the brightness temperature measured by the Advanced Microwave Scanning Radiometer (AMSR; Comiso et al. 2003;Karvonen 2014) instrument at 6.25 km × 6.25 km spatial resolution.However, the MODIS IST is more likely to be affected by clouds making it harder to detect the polynya, even though the finer resolution provided by MODIS allows for its better delineation (Paul et al. 2015).
The methodology of Aulicino et al. (2018), based on MODIS IST, is used in this work to estimate the area of the TNBP.Three MODIS products, over the domain 75.6-74.5°Sand 163-168°E, are used: (i) MYD02, to retrieve the IST; (ii) MYD03, to extract the geolocation of each 1 km satellite swath; (iii) MYD35, which provides cloud information used to discard scenes partially or completely covered by clouds.Only measurements collected by the MODIS instrument onboard the National Aeronautic and Space Administration's (NASA's) Aqua satellite are considered.Pixels for which IST ≤ 255 K are classified as sea ice and those for which IST ≥ 265 K are classified as open water, with intermediate values assigned using the pseudo-ship neighborhood analysis.This combination is more efficient than using a given single threshold to provide a correct estimation of the TNBP extent.In fact, due to its impact on airsea heat flux exchange and ice production, the polynya extent estimated using this method also accounts for the ocean surface that is partially covered by newly formed thin ice such as pancake and frazil ice around the open water (Aulicino et al. 2019;Mäkynen et al. 2020), whose temperatures can change from scene to scene in MODIS imagery.The TNBP area is then estimated as the sum of the open water and thin ice pixels, and is compared with the previous and following values for consistency.This procedure is applied to MODIS overpasses for April-October 2010 to 2017, as sea ice is not present in the region in the summer months.
The TNBP's spatial extent is also computed using the 3.125 km × 3.125 km SIC maps provided by the University of Bremen (UoB 2022) for comparison with that given by the MODIS IST.The 70% threshold used to separate ice from open ocean regions employed by several authors (e.g.Parmiggiani 2011;Ding et al. 2020;Wenta and Cassano 2020) is used here.This dataset provides daily SIC from 2002 to present estimated using the measurements collected by two instruments: The Advanced Microwave Scanning Radiometer onboard NASA's Earth Observing System (AMSR-E) from 2002 to 2011, and the Advanced Microwave Scanning Radiometer 2 (AMSR2) onboard the Japan Aerospace Exploration Agency Global Change Observation Mission 1st-Water "Shizuku" (GCOM-W1) satellite from 2012 to present.The SIC maps are derived from the University of Bremen ARTIST Sea Ice (ASI) algorithm, which is based on the polarization difference (vertical minus horizontal) of the brightness temperature, as ice has a much smaller polarization difference than open water (Spreen et al. 2008).For a given day, the number of pixels in the domain 75.6-74.5°Sand 163-168°E for which SIC < 70% is counted and the TNBP area is then estimated.This procedure is applied to all days in the period April-October from 2010 to 2017, the same for which MODIS IST data is available.
Sea ice motion vectors and SIC are also plotted to better visualize the evolution of the TNBP's spatial extent.Sea ice motion vectors are visualized using the low resolution (62.5 km) sea ice drift product of the European Organization for the Exploitation of Meteorological Satellites (EUMET-SAT) Ocean and Sea Ice Satellite Application Facility (OSI SAF; Lavergne et al. 2010).This product uses a combination of passive and active microwave satellite observations from the Special Sensor Microwave Imager/Sounder (SSMIS), the AMSR-E and the Advanced Scatterometer (ASCAT) instruments to derive sea ice drift over the preceding 48 h with daily temporal resolution.

Reanalysis data
ERA-5 is the latest reanalysis dataset generated by the European Centre for Medium Range Weather Forecasts (Hersbach et al. 2020).It is available from 1950 to present on an hourly basis and on a 0.25° × 0.25° (~ 27 km) grid.This product is selected due to its (i) high spatial and temporal resolution, which are needed to understand the processes responsible for the variability of the TNBP, and (ii) good performance in comparison with station observations and with respect to other reanalysis datasets (Gossart et al. 2019), including over the TNB (Ding et al. 2020).
ARs are identified in ERA-5 based on the Integrated Water Vapor Transport (IVT), which is computed following Eq.(1) below, where q is the specific humidity, V is the horizontal velocity vector, g is the gravitational acceleration of 9.80665 m s −2 , and dp is the difference in pressure between adjacent pressure levels: The algorithm employed to detect ARs, described in Francis et al. (2020), works as follows: (i) contiguous regions for which the IVT values exceed the 85th percentile (for the period 2008-2017) and 100 kg m −1 s −1 at each grid cell are first extracted; (ii) a filter is applied so as to retain areas for which the geometrical and directional requirements of ARs are met, in particular the objects must have a length of at least 1500 km and a length-width ratio that must exceed 1.5.More details about the AR detection and classification can be found in Mattingly et al. (2018Mattingly et al. ( , 2020)).The IVT threshold used in the AR algorithm, 85th percentile, is smaller than the 98 th percentile employed by Willie et al. ( 2021) that is more restrictive, but similar to that used in other Antarctic studies which investigated ARs (e.g.Francis et al. 2020;Clem et al. 2022).While both methodologies do not detect ARs in the TNBP region (163-168°E; 75.6-74.5°S),the warmer and more moist air mass associated with the ARs does impact the TNB, as highlighted in the April 2017 case study considered in Sect.5.2.
In order to identify mesocyclones, which are a regular occurrence in the TNB and have an impact on the TNBP's spatial extent (Sinclair et al. 2010), two fields in which they have an imprint are considered: the 850 hPa relative vorticity and maximum eddy growth rate (MEGR), the latter a measure of baroclinicity (Hoskins and Valdes 1990).The MEGR is defined in Eq. ( 2), where f is the Coriolis parameter, u is the zonal wind speed, z is the vertical coordinate, N is the Brunt-Vaisala frequency, is the potential temperature and g is the gravitational acceleration.A threshold of 0.6 day −1 , considered in Hoskins and Valdes (1990), is used to identify regions of high baroclinicity.
The prevailing mechanisms responsible for the variability of the TNBP are identified by applying the k-mean clustering technique (Steinley 2006) to the TNBP spatial extent as estimated by the MODIS IST and UoB SIC data.The k-means clustering technique initially assigns each data point to a cluster and then successively updates the clusters so as to make them as homogeneous and distinct from the other clusters as possible.In practical terms this means minimizing the sum of the squares of the distances between each data point and the respective cluster's centroid while maximizing the variability between clusters.A set of surface and pressurelevel fields from ERA-5 is selected as discussed in Sect.4, and a range of clusters from 1 to 5 is considered.The optimal number of clusters is identified through the silhouette technique (Rousseeux 1987), applied to sea-level pressure as it is a critical field in determining the spatial extent of the TNBP given its close association with the strength of the wind flow.The silhouette scores range from − 1 to + 1, with a higher value indicating a better performance.

Observational data
Besides satellite and reanalysis data, in-situ measurements at automatic weather stations (AWSs) in the TNB are considered to assess the strength of the katabatic winds and quantify changes in temperature and water vapour mixing ratio in association with warmer and more moist air intrusions.

Climatology of TNBP occurrence
Figure 1a shows the polynya's spatial extent for the April-October days of 2010 to 2017 obtained with MODIS IST data.The monthly climatology is given in Fig. 1c.The seasonal variability of the TNBP, with a smaller area in the middle of the winter and larger values in the transition seasons (April-May and September-October), is in line with that reported by other authors (e.g.Kurtz and Bromwich 1985;Ding et al. 2020).The TNBP's spatial extent is around 1000 km 2 in April and October, with a minimum of ~ 380 km 2 in July.The lower temperatures in the coldest months make it harder for the polynya to develop (Ding et al. 2020), with warmer and more moist air intrusions from the lower latitudes also less frequent compared to the transition season months.This also explains the larger standard deviation in the warmer months.The diurnal variability of the TNBP is controlled by the frequency of occurrence and strength of the katabatic winds (Aulicino et al. 2018).The highest estimated area in the eight-year period is ~ 12,836 km 2 on 31 October 2010, an event discussed in Sect.5.1.Despite a similar temporal variability, as shown in Fig. 1c, the TNBP spatial extent estimated with UoB SIC data is generally higher than that obtained using MODIS IST data and with a larger spread as well (cf.Figure 1a, b).In particular, with MODIS IST data the eight-year mean is about 710 km 2 , while using the UoB SIC data the TNBP's average spatial extent is roughly 2460 km 2 , the latter in the range of the 2400 ± 500 km 2 reported by Ding et al. (2020) who employed the same methodology for the period 2005-2015.The contrasting magnitudes of the two estimates has been noted by other authors.e.g., Ciappa et al. (2012), who used MODIS data for 2005-2010, estimated the TNBP spatial extent to be of ~ 900 km 2 with annual means ranging from 800 to 1000 km 2 , in line with the 1008 km 2 and 878 km 2 reported by Aulicino et al. (2018) for 2010 and 2011, respectively.On the other hand, Martin et al. (2007), who detected the polynya using sea ice thickness data at 25 km resolution, found the average TNBP area over 1992-2002 to be 3000 ± 750 km 2 , while Kern (2009), who made use of the brightness temperature difference between two frequencies (37 GHz and 85.5 GHz) at 5 km resolution, reported an averaged value over the same 11-year period of 4,200 ± 800 km 2 .Besides the different techniques used to estimate the TNBP area, the fact that (i) the spatial resolution of the MODIS IST (1 km) is roughly three times higher than that of the UoB SIC data (3.125 km) for a polynya whose spatial extent is generally less than 10,000 km 2 or roughly 100 × 100 km (Fig. 1a,  b), and (ii) the higher temporal resolution of the MODIS IST data (sub-daily vs. daily UoB SIC data) explain the discrepancy in magnitude between the two estimates.This stresses the need for frequent high-resolution satellite data such as the Synthetic Aperture Radar (SAR; e.g.Yastika et al. 2019) and fine-tuned algorithms to assess the spatial extent of coastal polynyas such as the TNBP that are typically small in size and exhibit large variability.Having said that, the monthly means of the TNBP area are generally within 300 km 2 of each other, Fig. 1c, and when accounting for the standard deviation the two estimates can be regarded to be in agreement.

Large-scale atmospheric conditions
In order to better understand the mechanisms responsible for the variability of the TNBP, the k-means clustering technique (Steinley 2006) is employed considering five pressure-level fields (namely the 500 hPa geopotential height and zonal and meridional wind speed; 850 hPa relative vorticity and MEGR) and seven 2D fields (sealevel pressure; 10-m zonal and meridional wind speed; precipitable water; vertical integral of eastward and northward water vapour flux; surface total energy flux) over all the April-October days in the 8-year period for which the polynya's spatial extent, as estimated from the MODIS IST and UoB SIC data, exceeded one standard deviation from the respective mean (total of 273 and 274 days, respectively).A different number of clusters from one to five is considered, with the silhouette analysis (Rousseeux 1987), applied to the sea-level pressure, indicating the highest scores for two clusters for both (Figs.S1a-b).For the MODIS IST clusters, clusters #1 and #2 account for about 45% and 55% of the days when the TNBP area exceeds one standard deviation from its 2010-2017 mean, respectively, and are characterized in Fig. 2. The monthly occurrence is shown in Fig. 2a, in which the number of days corresponding to each cluster that falls in a given month over April-October 2010-2017 is added.The daily occurrence is given by the vertical lines in Fig. 1a.The clusters obtained with the TNBP area estimated using UoB SIC data are given in Fig. S2 with the time-series presented in Fig. 1b.
Cluster #1, Fig. 2b-d, comprises the prevailing katabatic wind episodes.In particular, warmer and more moist low-latitude air arrives into Victoria Land from the Pacific Ocean and dries out as it moves over the higher terrain, with actually negative precipitable water anomalies at the TNBP of around − 0.2 kg m −2 (Fig. 2c).The combination of the ridge over northeastern Victoria Land and a trough to the east, over the Ross Ice Shelf and West Antarctica, leads to offshore winds over the TNB.The drier conditions in the TNB indicate these offshore winds are likely katabatic winds as the latter are characterized by an increase in air temperature and a decrease in humidity as the air experiences adiabatic compression as it descends from the Transantarctic Mountains (Nylen et al. 2004).The ASL in this cluster is displaced westwards with respect to its climatological location (Raphael et al. 2016), leading to a steeper pressure gradient around the TNB strengthening the near-surface winds.The large-scale pattern given in Fig. 2b comprises a wavenumber #4, which is the dominant mode of intra-seasonal variability in the Southern Hemisphere mid-latitudes (Kidson 1999).It can be triggered by anomalous convection in the subtropical Western Fig. 2 a Monthly occurrence of clusters #1 (red) and #2 (blue) obtained using the TNBP area estimated from MODIS IST data for the period April-October 2010-2017.b Geopotential height (shading; m) and horizontal wind vector (arrows; m s −1 ) at 500 hPa, and zoomed-in plots of the c precipitable water (shading; kg m −2 ) and the Integrated Water Vapor Transport (IVT) vectors (arrows; kg m −1 s −1 ) and d 850 hPa relative vorticity (shading; s −1 ) and maximum eddy growth rate, hatching if ≥ 0.6 day −1 , the threshold used in Hoskins and Valdes (1990), for the first cluster, which accounts for ~ 45% of the days when the area of the TNBP area is higher than its April-October 2010-2017 mean plus one standard deviation (273 days in total).The cross in b highlights the approximate location of the TNBP while in c, d the TNBP region (163°-168°E; 75.5°-74.5°S)is plotted as a black rectangle.In (d), areas where the 850 hPa pressure level is below the surface are shaded in grey, with the location of Victoria Land and the Ross Ice Shelf also given.The fields are extracted from ERA-5 with the anomalies in b, c computed with respect to the 1979-2020 climatology.e-g are as b-d but for the second cluster, which accounts for about 55% of the days ◂ Pacific near New Zealand (Senapati et al. 2021), and has also been shown to play a role in marine heat waves and cool spells (Chiswell 2021).At a more local scale, and as seen in Fig. 2d, the circulation in cluster #1 promotes the occurrence of cyclonic vorticity (as noted by the negative relative vorticity values) and baroclinicity (as noted by the MEGR values in excess of 0.6 day −1 ) over the TNB, both of which promote the development of mesocyclones in the region (Sinclair et al. 2010).The relative vorticity anomalies in the TNB (not shown) are in phase with the full values presented in Fig. 2d, indicating an expected strengthening of the cyclonic circulation in response to a more active katabatic wind regime.The correspondent cluster obtained with the UoB SIC data, Figs.S2b-d, exhibits a similar large-scale and regional-scale pattern, even though the ridge over Victoria Land is weaker and extends further into Antarctica, resulting in a weaker pressure gradient around the TNB.
Cluster #2 (Fig. 2e-g) is associated with the advection of warmer and more moist air from the eastern Pacific into the Ross Sea and the TNB through West Antarctica, a common pathway as given by the back-trajectories computed by Scarchilli et al. (2011).This occurs in association with a ridge that extends over the whole continent and a more active ASL, Fig. 2e.It is important to note that the transient nature of the mid-latitude baroclinic systems means the actual source of moisture may not be the region from which it is being advected.For example, in the case study corresponding to cluster #2 discussed in Sect.5.2, the warmer and more moist lower-latitude air mass coming into the Ross Sea is advected from the east to northeast, as in Fig. 2f, but in fact it originates from the Tasman Sea and the New Zealand sector.The large-scale circulation around the South Pole projects onto the negative phase of the Southern Annular Mode (SAM; Fogt et al. 2012;Zheng and Li 2022), the leading mode of extratropical variability in the Southern Hemisphere.The negative phase of the SAM is associated with an equatorward shift of the polar jet, and is known to promote the occurrence of ARs in the Ross Sea (Wille et al. 2021).Negative precipitable water anomalies prevail over much of West Antarctica, with a strip of positive anomalies with a magnitude of ~ 0.1-0.3kg m −2 extending from the Weddell Sea and Ronne Ice Shelf across the interior of Antarctica and into the western Ross Ice Shelf (Fig. 2f), which is in contrast with the drier conditions seen here in cluster #1 (cf. Figure 2c).The prevailing flow also opposes the background offshore winds in the western Ross Sea, which explains the reduced values of cyclonic vorticity in the TNB (Fig. 2g) compared to that of cluster #1 (cf. Figure 2d), with overall smaller values of the MEGR as well.Cluster #2 prevails in the transition seasons, Fig. 2a, in particular in September-October and April-May, when warmer and more moist air intrusions into the region are more frequent (Nicolas and Bromwich 2011).As opposed to cluster #1, the cluster #2 maps obtained with MODIS IST and UoB SIC data are largely in agreement, with a reduced moistening over the Ross Sea/Ross Ice Shelf in the latter as a result of a more spatially confined (albeit slightly stronger) ASL.In any case, the strip of positive precipitable water anomalies extending from the Weddell Sea into the Ross Ice Shelf extends over Victoria Land into the adjacent Southern Ocean in the UoB SIC cluster #2 (Fig. S2f).
The results of the cluster analysis indicate that, besides katabatic winds promoted by the steep pressure gradient between a ridge over Victoria Land in Antarctica and a westward displaced ASL, the advection of warmer and more moist air from the Pacific Ocean can also foster the formation and widening of the TNBP.As seen in Fig. 2a, such episodes are more frequent in the transition seasons.

Regional-scale atmospheric circulation patterns
In this section, one event representative of each of the clusters presented in Sect. 4 is analyzed in detail, using a combination of satellite-derived products, in situ measurements and reanalysis data.Despite their regular occurrence in the TNB, no mesocyclones are identified in the polynya region in the two case studies discussed below.

Cluster 1: katabatic wind regime
On 31 October 2010, the largest spatial extent of the TNBP, as given by MODIS IST data for the period April-October 2010 to 2017, was observed.Figure 3 highlights the dramatic increase in the polynya's area from ~ 1800 km 2 on 30 October (Fig. 3b) to ~ 12,836 km 2 on 31 October (Fig. 3c), with values in excess of 10,000 km 2 on 01 November (Fig. 5f).
The polynya remained wide, when compared to that in early October (Fig. 3a), well into November 2010 (Fig. 3d).The role of the Drygalski Ice Tongue in maintaining the polynya noted by Bromwich and Kurtz (1984) is also seen in Fig. 3, as it prevents sea ice from the south from moving into the TNB.The ice filaments seen in particular in Fig. 3b-d correspond to ice streaks, which develop in the presence of strong winds (Kampf and Backhaus 1999), fading over time as the winds slacken (cf. Figure 3c and d).Similar observations using MODIS data have been made by Scarchilli et al. (2010) for katabatic wind events in the region in 2007 and 2008.
Figure 4 summarizes the large-scale conditions on 30-31 October 2010, while the observations collected at the location of the AWSs from 29 October to 01 November 2010 are given in Fig. 5a-d.No measurements were collected at the Virginia AWS during this period.Air temperature and mixing ratio data at the Rita AWS and wind observations at the Manuela AWS are also not available during 29 October to 01 November 2010.The polynya-averaged values from ERA-5 are plotted as a black line.Figure 5e shows the polynya-averaged surface net radiation flux, R net , total energy flux, F net , and downward longwave radiation flux, LW down , from the reanalysis dataset, with positive values indicating downward pointing fluxes.The TNBP area, as given by the MODIS IST and UoB SIC data, is given in Fig. 5f.
On 30 October, Fig. 4a, a low pressure over northeastern Victoria Land weakened the background offshore flow and advected slightly more moist mid-latitude air into the TNB.This is seen in the station data, Fig. 5b-d, which show weak wind speeds, generally below 10 m s −1 and at times blowing from an onshore direction, mostly from 06 to 18 UTC on 29 and 30 October, accompanied by an increase in the mixing ratio by up to 0.5 g kg −1 .The surface downward longwave radiation flux increased from about 140 W m −2 to 230 W m −2 as a result of the more moist and cloudy weather conditions.The low pressure shifted eastwards and strengthened on 31 October, Fig. 4a, b, with a ridge from the Pacific extending into northern Victoria Land and bringing a warmer and more moist air mass into the region that dried out as it moved through, a set up that corresponds to cluster #1 (Fig. 2b-d).At the same time, the polar high over the Antarctic continent (Kakegawa et al. 1986) also moved into northern Victoria Land.The interaction of these three systems led to a pronounced increase in the wind speed from 30 to 31 October 2010, Fig. 5d, with values up to 40 m s −1 at the Rita AWS.As noted by Wang et al. (2021), the interaction between the local topography and synoptic-or mesoscale systems can lead to the intensification of offshore winds, as found for coastal polynyas in East Antarctica.For the 08 May 2017 (Guest 2021a) and 21 September 2012 (Wenta and Cassano 2020) katabatic wind events, speeds of 35 m s −1 were observed at the Manuela AWS, for which data is not available for this period, comparable to those observed at the Rita AWS on 31 October 2010.The winds also blew persistently from a westerly direction, Fig. 5c, with the increase in air temperature (by up to 6 °C, Fig. 5a) and October 2010 from 00 to 23 UTC, with the daily mean value drawn as a thick purple line.f Vertical cross-section at 75°S of zonal wind speed (shading; m s −1 ), the arrows give the direction, and potential temperature (solid black contours; 5 K) on 31 October 2010 at 18 UTC.Regions below the orography are shaded in grey, while the vertical line gives the approximate location of the center of the TNBP drop in mixing ratio (by up to 0.2 g kg −1 , Fig. 5b) in line with a katabatic wind episode, as the air coming from the Transantarctic Mountains experiences adiabatic compression while descending the slopes (Nylen et al. 2004).The drier conditions (Fig. 5b) and the clear skies (Fig. 4d) led to a drop in the downward longwave radiation flux from 230 W m −2 to 145 W m −2 during 30-31 October (Fig. 5e).While the largest TNBP extent, as given by the MODIS IST data, is on 31 October at roughly 05 UTC, Fig. 5f, when the winds start to pick up, the air temperature begins to increase and the mixing ratio starts to drop, it is just a single occurrence.Persistent high values are seen at 03-05 UTC on 01 November, the same day when the TNBP area, as given by the UoB data, achieves its maximum magnitude in the period.This is after the katabatic wind event, highlighting its role in the widening of the polynya.As noted in Aulicino et al. (2018), katabatic winds have to be strong (with speeds in excess of 20-25 m s −1 ) and persistent (lasting for more than 8 h) to have an impact on the polynya.What is more, the peak in the TNBP extent is roughly 12 h after the katabatic winds reached their maximum strength, which is consistent with the typical lags between the wind forcing and the TNBP response of 9 to 14 h found by Aulicino et al. (2018).Despite the strong winds, the sea ice drift speeds in the 48 h period ending on 01 and 02 November 2010 are rather small of up to 20 km, or about 0.12 m s −1 (Figs.S3a-b), typical ice drift speeds in the western Ross Sea (Jacobs and Comiso 1989;Farooq et al. 2020).A possible explanation is that the katabatic wind episode on 31 October, although remarkable, was a pulse event, with rather weak wind speeds with a more variable direction on 29 and 30 October and after the event on 01 November (Fig. 5c, d).
Figure 4c shows the tight pressure gradient over the Western Ross Sea and 10-m wind gusts in excess of 30 m s −1 in the reanalysis data on 31 October at 18 UTC, around the time when they peaked in intensity.The tongue-shape of the wind gusts in the TNB reflects the blocking effect of the Drygalski Ice Tongue to the south and Cape Washington to the north (Fig. 3).Actual wind gusts are likely to be of an even higher magnitude, as the still coarse (0.25° or ~ 27 km) spatial resolution of ERA-5 precludes a full representation of the complex topography of the region, in particular of the Transantarctic Mountains, and therefore it incorrectly captures the tail of the wind speed distribution.The ERA-5 winds for this event can be quantitatively compared with those predicted for the full 1979-2021 period.This is done in Fig. 4e, where the histogram shows the frequency of occurrence of polynya-averaged wind speeds in the range 0-25 m s −1 in the 43-year period, with those for 31 October 2010 highlighted with the coloured vertical bars.Some of the speeds on 31 October are in the top 1% of the climatological distribution, with the daily-averaged speed (thick purple line) in the top 5% of the distribution.This underscores the extreme nature of this event, whose impact on the polynya's extent is confirmed in the satellite images given in Fig. 3.
The shading in Fig. 4d gives the surface total energy flux (F net ), defined as the sum of the radiation and heat fluxes, and the hatching denotes regions where the total cloud cover exceeds 25% on 31 October at 18 UTC.Clear skies are seen in the region around the TNBP, consistent with a katabatic wind event (Bromwich 1989).The F net values reached a peak of roughly −350 W m −2 in the TNB, indicating a strong heat flux from the surface to the atmosphere, as expected in the case of a polynya where the open ocean is exposed to the very cold near-surface air.Guest (2021b), and from instruments deployed on a ship moving away from the coast, estimated total fluxes of up to 2500 W m −2 just off the ice shelf, decreasing to below 500 W m −2 roughly 50 km away from the coast.The time-series of the polynya-averaged R net and F net given in Fig. 5e shows a pronounced diurnal cycle, with a drop in magnitudes from 30 to 31 October in response to the katabatic wind event, also seen in LW down .The R net (and consequently the F net ) values are positive during daytime (local time is roughly UTC + 11), in particular on 31 October, owing to the clear skies (Fig. 4d) and heating of the surface by the Sun in mid-austral spring (the midnight Sun at the site starts in early November).The polynya-averaged F net values on 31 October 2010, which dropped below − 150 W m −2 (Fig. 5e), are comparable to those estimated in the September 2017 Weddell Sea polynya (Francis et al. 2020).
Further insight into the katabatic wind event can be gained by inspecting vertical cross-sections, one of which is presented in Fig. 4f.Here, the zonal wind and potential temperature are plotted at 75°S, which is roughly the center latitude of the TNBP, with its center longitude highlighted by the black vertical line.Despite its coarse resolution to represent a downslope wind event, the ERA-5 cross-section shows the winds descending along the isentropes and accelerating as they do so, with speeds in excess of 30 m s −1 extending throughout the polynya and up to about 5° (or ~ 555 km) downstream, in line with the tongue of high wind speeds shown in Fig. 4c.The distance over which the katabatic winds reached while maintaining their characteristics in this case is roughly five times larger than the ~ 100 km found by Guest (2021a).However, the polynya's area on 08 May 2017, as estimated using MODIS IST data, peaked at about 2228 km 2 , which is roughly six times less than the maximum spatial extent on 31 October 2010.It appears therefore that the stronger the katabatic wind event, the wider the TNBP will be, and the further away from the coast the winds of katabatic origin will reach.

Cluster 2: warmer and more moist air intrusion
A case of a warmer and more moist air intrusion from the mid-latitudes into the TNB took place on 27-30 April 2017, and is summarized in Figs. 6,7,8,9.The air mass source is the Tasman Sea/New Zealand sector, one of the source regions of air masses coming into the TNB in particular during the warmer months (Vagnoni et al. 2021), even though it reaches the Western Ross Sea indirectly from the northeast (Fig. 6c-f).The TNBP's spatial extent reached 8,096 km 2 on 28 April and 7,861 km 2 on 29 April 2017 (Fig. 8f) which, although lower than that in the previous event, is still more than twice as large as that in the katabatic wind cases discussed in Guest (2021a, b) and Wenta and Cassano (2020).Two low pressure systems moved eastwards north of Victoria Land and the Ross Sea, one on 26-27 April (Fig. 6a) and another on 28-29 April (Fig. 6b), with a ridge over Antarctica leading to the advection of warmer and more moist air into the Ross Sea (Fig. 6c-f).The largescale pattern during this event (Fig. 6a, b), with a low pressure to the southeast of New Zealand and another extending from the northeastern Ross Sea to the Bellingshausen Sea and a ridge over Antarctica, is a typical set up for cluster #2 events (Fig. 2e).As opposed to the October 2010 event, there is a much reduced pressure gradient over the western Ross Sea (cf.Figures 4c, 6d, f), which is consistent with the weaker wind speeds observed at the location of the AWS stations, generally below 20 m s −1 (Fig. 8d).In fact, the polynya-averaged ERA-5 wind speeds on 28 April 2017, comparable to those observed at the Eneide, Maria and Rita AWSs, are roughly in the middle of the April climatological (1979-2021) distribution (Fig. S3c).Offshore winds prevailed at the location of the stations during this event, blowing consistently from a westerly to northerly direction (Fig. 8c).However, the reduced speeds stress the smaller role played by the katabatic winds in the widening of the polynya during this event, in particular as they do not meet the conditions stated in Aulicino et al. (2018), namely a speed in excess of 20-25 m s −1 for at least 8 h, to have an impact on the TNBP, with a typical time-lag between the wind forcing and the polynya's response of 9-14 h.This is particularly true on 28 and 29 April, when the wind speeds were the lowest in the four-day period (Fig. 8e).However, it is important to note that the referred thresholds are inferred from two-years of data, and both the strength of the wind and the response of the polynya to the wind flux exhibit a large variability in the area (Ciappa et al. 2012;Rusciano et al. 2013).
Figure 7 shows two ARs, one from the Tasman Sea on 26-27 April and another to the east of New Zealand on 28-29 April.The moisture levels associated with these ARs around Victoria Land and the Ross Sea, of up to 150 kg m −1 s −1 , are typical of ARs reaching the Antarctic Fig. 6 Sea-level pressure (shading; hPa) and 10-m wind vectors (arrows; m s −1 ) on (a) 27 April and (b) 29 April 2017 at 12 UTC.Zoomed-in view of the c precipitable water (shading; kg m −2 ) and 2-m temperature (brown contours; every 10 °C) and d surface total energy flux (F net ; shading; W m −2 ) and sea-level pressure (brown solid contours; every 10 hPa) on 27 April 2017 at 12 UTC.e, f are as c, d but on 29 April 2017 at 12 UTC.The conventions are as in Fig. 4a-d coast (Francis et al. 2020;Gorodetskaya et al. 2020).The first warmer and more moist air intrusion is the most prominent (cf. Figure 7a, b with 7c, d), and is the main contributor to the increase in air temperature (by as much as 7 °C) and mixing ratio (which roughly tripled at the Eneide and Rita AWSs to around 0.75 g kg −1 and doubled at the location of the remaining three stations) at the AWSs from 27 to 29 April (Fig. 8a, b).This is seen in Fig. 6c and e, with values of precipitable water of about 3-4 kg m −2 and air temperatures in the range − 15° to − 10°C around the TNB on 29 April at 12 UTC, roughly when the second peak in the polynya's spatial extent is reached (Fig. 8f).The warmer, more moist and cloudier conditions led to an increase in the surface downward longwave radiation flux from 160 W m −2 to 240 W m −2 during 28-29 April, remaining high until the end of 30 April when drier conditions started to develop (Fig. 8e).The roughly 80 W m −2 increase in LW down is comparable to that observed at a weather station at Pine Island Glacier in West Antarctica in response to a warm air intrusion (Djoumna and Holland 2021).It is interesting to see that in this case the rise in the downward longwave radiation flux is accompanied by an increase in the polynya's spatial extent (Fig. 8e, f), while in the October 2010 event the opposite is true (cf. Figure 5e, f).This highlights the effects of the warmer and more moist air intrusion on the polynya's area.ARs are a regular occurrence around Antarctica (Francis et al. 2021(Francis et al. , 2022)), and while the Ross Sea is one of the least likely regions to be affected by ARs, they do occur here (Wille et al. 2021).ARs have not been detected in the TNBP region during April-October 2010-2017 both using the algorithm considered here and that of Willie et al. (2021) (not shown).However, the advection of a warmer and more moist air mass, which can occur in association with an AR as in the April 2017 event, plays a role in the variability of the polynya's spatial extent.The fact that the measurements at the Virginia AWS are similar to those in the Manuela AWS is not surprising, as the two stations are close to each other located at the Inexpressible Island (Fig. 3a).Together with the Maria AWS located at Mount Browning in the southern part of the TNB, the three sites record the coldest and driest conditions in the region.On 30 April, the weather conditions turned drier, with an increase in wind speed at the Manuela and Virginia AWSs, suggesting a return to the katabatic wind regime.These changes explain the drop in the polynya-averaged surface fluxes at the end of 30 April (Fig. 8e).As opposed to the late October 2010 event, both the polynya-averaged R net and F net are below zero (i.e.pointing upwards towards the atmosphere) during 27-30 April 2017 (cf. Figure 8e with 5e).This is explained by the fact that by late April the site is about to enter the polar night (i.e.no direct sunlight) with up to roughly 3 h of direct sunlight, whereas in late October it has about 20-22 h of direct sunlight, just before the midnight Sun which begins in early November.However, in both, as in LW down , the effects of the warmer and more moist low-latitude air intrusion during 28-29 April is present through the steady increase in their magnitudes, dropping at the end when the katabatic regime began to establish itself.The fact that the three radiation fluxes exhibit a similar temporal variability stresses the important role played by the downward longwave radiation in the surface energy budget.Hourly data for the Virginia weather station, located at the Inexpressible Island (Fig. 3a), is available for this period and is given by the orange line The effects of the poleward flow into the Ross Sea can be seen in the sea ice drift maps (Fig. 9a, b).The speeds peaked at 25 km day −1 (or ~ 0.29 m s −1 ) on 28 April just east of Victoria Land.The southward sea ice drift is consistent with the westward low-level atmospheric flow that prevailed in the Ross Sea in particular on 27 April, when a deep low pressure moved eastwards over the northern part of the basin (Fig. 6a and d; note that the sea ice drift velocity is estimated over the 48 h period ending at the given time).As a result of the ASL, the Ross Sea is a major exporter of sea ice (Holland and Kwok 2012), with an average sea ice drift of roughly 0.2 m s −1 over April-October 1992-2010 away from the region into the Southern Ocean.There is, however, considerable interannual variability, partially linked with ENSO (Kwok et al. 2017;Pope et al. 2017;Silvano et al. 2020), with the inflow of ice remaining approximately constant from 1992 to 2008 (Comiso et al. 2011).A comparison with the ice drift speeds in the western Ross Sea shows that a value of 0.3 m s −1 is at the higher end of those observed in the region (Jacobs and Comiso 1989;Farooq et al. 2020).Further, the southward transport of ice also promoted the widening of some of the coastal polynyas: e.g.note the oceanward expansion of the pink line, which corresponds to a SIC of 85%, in the TNBP and the two polynyas to its north, from 27 to 28 April (Fig. 9a, b).The F net values reached down to -520 W m −2 (-300 W m −2 at the TNB) during this event, Figs.6d and 8 e, a smaller magnitude compared to that in the October 2010 katabatic wind episode.The regions where F net is most negative, the northwestern edge of the Ross Ice Shelf and the eastern side of Victoria Land including in the TNB, largely correspond to the areas where sea ice melting took place, as evidenced by the lower SIC values in the region (Fig. 9c, d).These areas exhibit large sea ice gradients (Rack et al. 2021), which increased owing to the substantial coastal melting, with the SIC dropping below 15% in some parts.The southward movement of sea ice in the western side of the Ross Sea towards the ice shelf, which peaked on 28 April (Fig. 9b) but continued until 29 April, however, replenished some of the melted ice.

Conclusions
Polynyas, or "holes in the ice", are a common occurrence at higher-latitudes, triggered by the upward advection of warmer water, strong winds and/or ocean currents that push the ice away from a coastline.While mid-sea polynyas are generally larger in size, coastal ones are a more regular occurrence, and hence changes in their variability over time can potentially be an indication of climate change.One of such features is the Terra Nova Bay Polynya (TNBP), located on the eastern side of Victoria Land facing the Ross Sea and on the lee side of the Transantarctic Mountains.While the downslope (katabatic) winds are its main driver (e.g.Bromwich and Kurtz 1984;Aulicino et al. 2018;Wenta and Cassano 2020), little attention has been given to other mechanisms, such as warmer and more moist air intrusions that are known to contribute to the formation and widening of polynyas (e.g.Francis et al. 2020).This is particularly important given the crucial role polynyas play in the atmospheric and oceanic circulation, which can be more pronounced in a warming world (Smith Jr. and Barber 2007b), and the fact that they are still poorly represented by numerical models (Mohrmann et al. 2021).In this study, the main atmospheric drivers of the TNBP are investigated using ERA-5 reanalysis, satellite-derived and in-situ data.
The spatial extent of the TNBP is estimated using the 1 km ice surface temperature (IST) data collected by the Moderate Resolution Image Spectroradiometer (MODIS) instrument onboard the Aqua satellite.An analysis of the April-October 2010-2017 TNBP area revealed an annual mean of about 700 km 2 , with lower values in the bulk of the winter season (~ 380 km 2 in July as opposed to ~ 1000 km 2 in April and October) owing to the colder temperatures and reduced influence of warmer and more moist air masses.The TNBP spatial extent is also computed using the 3.125 km sea ice concentration (SIC) daily maps provided by the University of Bremen (UoB) for the same 8-year period, and the 70% threshold employed by several authors (e.g., Parmiggiani 2011;Wenta and Cassano 2020) to distinguish between ice and ice-free regions.The annual mean spatial extent is about 2460 km 2 , in line with the results of Ding et al. (2020) who considered the same data for the period 2005-2015.This stresses the sensitivity of the TNBP area estimates to the resolution of the satellite-derived data used and to the methodology employed to detect it, and is likely applicable to other coastal polynyas around Antarctica that are generally smaller in size and exhibit a pronounced temporal variability.However, when accounting for the standard deviation, the monthly mean of the TNBP area estimated using MODIS IST data is generally in agreement with that computed using the UoB SIC data, with both exhibiting the same annual cycle.
A cluster analysis of all the days when the polynya's spatial extent exceeded its eight-year mean plus one standard deviation revealed that katabatic wind episodes' and warmer and more moist low-latitude air intrusions are the dominant processes that control the variability of the TNBP's area.The former accounts for ~ 45% of the 273 days, being associated with a westward-displaced Amundsen Sea Low (ASL) and a ridge over Antarctica, with the resulting increased pressure gradient around the TNB strengthening the offshore winds.The large-scale pattern for this cluster features a wavenumber #4, the dominant mode of intra-seasonal variability in the Southern Hemisphere mid-latitudes (Kidson 1999).In 55% of the days, warmer and more moist air intrusions from the Pacific Ocean can play an important role.Such occurrences are associated with a ridge over Antarctica and a more active ASL, a pattern that projects onto the negative phase of the Southern Annular Mode (SAM), and are more frequent in the transition seasons (April-May and September-October).
One event corresponding to each cluster is analyzed in detail.For a katabatic wind episode, the one that took place in late October 2010, when the TNBP's spatial extent, as given by the MODIS IST data, reached its highest value in the period April-October 2010-2017, ~ 12,836 km 2 on 31 October, is investigated.The ASL was particularly strong on this day, with a daily-mean central pressure below 945 hPa.The pressure gradient with respect to the polar high, which was sitting over Victoria Land with a central pressure of roughly 1020 hPa, gave rise to strong downslope winds.10-m wind speeds as high as 40 m s −1 are observed at the location of weather stations in the region, with the polynya-averaged 10-m wind speeds from ERA-5 on 31 October being in the top 1% of the 1979-2021 October wind speed distribution.The strengthening of the offshore winds was accompanied by an increase in the air temperature and a reduction in the mixing ratio, as expected for katabatic winds that experience adiabatic compression as they descend the Antarctic Plateau towards the coastal areas (Nylen et al. 2004).The surface total energy flux (F net ) reached about -350 W m −2 at the TNB during this event, comparable to that estimated for the Weddell Sea polynya in September 2017 (Francis et al. 2020), with the upward pointing fluxes reflecting the exposure of the open water to the very cold polar air.
A warm and moist air intrusion took place on 26-29 April 2017, with two cyclones moving eastwards north of Victoria Land and a ridge over Antarctica injecting a lower-latitude air mass into the Ross Sea.The effects of the warmer and more moist air intrusion are seen at the location of five AWSs in the TNB, with an up to 7 °C increase in the air temperature and up to a tripling of the mixing ratio to 0.75 g kg −1 from 27 to 29 April.The polynya-averaged surface downward longwave radiation flux of the ERA-5 reanalysis increased by about 80 W m −2 in response to the warmer, more moist and cloudier conditions.Satellite data showed a pronounced decrease in SIC along the northwestern edge of the Ross Sea and the TNB, with ice drift speeds of up 25 km day −1 on 27 April.The southward transport of ice towards the Ross Ice Shelf, likely a response to the easterly low-level atmospheric flow, cleared some of the offshore ice off the east coast of Victoria Land which promoted the expansion of coastal polynyas in the region.Further research is needed into ice movements in the Ross Sea as it has wide implications for the ice shelf and polynyas.Wind speed measurements at the weather stations in the TNB are generally below 20 m s −1 during 28-29 April, with the polynya-averaged reanalysis speeds roughly in the middle of the April 1979-2021 distribution.This confirms the reduced role played by katabatic winds in the widening of the TNBP.As found by Aulicino et al. (2018), katabatic winds have to be strong (with speeds in excess of 20-25 m s −1 ) and persistent (lasting for at least 8 h) to have an impact on the TNBP, which is not the case in this event.However, the referred thresholds are obtained from two years of measurements, with both the strength of the wind and the spatial extent of the polynya exhibiting a large variability.
The findings of this work confirm that, despite their close proximity to steep topographic features adjacent to the Antarctic landmass, coastal polynyas around the continent are not solely driven by katabatic winds.While the Ross Sea is one of the least likely regions of Antarctica to be affected by low-latitude air intrusions, given the presence of a topography-induced cyclonic eddy that drives the extratropical cyclones and associated moisture plumes eastwards towards the Amundsen Sea (Baines and Fraedrich 1989), the frequency of atmospheric rivers (ARs) has been increasing here (Wille et al. 2021).This suggests that warmer and more moist air advection from the mid-latitudes, which can occur in association with an AR even if the AR does not move directly over the polynya, is likely to play a larger role in the variability of coastal polynyas, such as the TNBP, in a warming climate.

Fig. 1 a
Fig. 1 a Time-series of the TNBP area (km 2 ) for the April-October days of 2010 to 2017 as measured by the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument onboard the National Aeronautics and Space Administration's (NASA's) Aqua satellite.The black line shows the estimated value and the pink shading gives one standard deviation from the mean.The red and blue vertical lines indicate the days that belong to clusters #1 and #2, respectively, with the percentage of data points each cluster comprises given in the legend.b is as a but for the TNBP spatial extent estimated from the 3.125 km sea ice concentration (SIC) dataset provided by the Uni-

Fig. 3
Fig. 3 MODIS visible imagery on a 01, b 30 and c 31 October and d 10 November 2010.The spatial extent of the TNBP for 28 October-01 November is given in Fig. 5f.The red triangle in the inset gives the location of the polynya, while the red circles in a indicate the five AWs considered in this study: Manuela (74.946°S;

Fig. 5
Fig.5a 2-m air temperature (°C), b water vapour mixing ratio (g kg −1 ) and 10-m wind c direction (°) and d speed (m s −1 ) at the Eneide (red curve), Maria (green), Rita (blue) and Manuela (purple) stations, defined in Fig.3a, and ERA-5 data area-averaged over the TNBP (163°-168°E; 75.5°-74.5°S;black).e Surface net radiation flux, R net (blue; W m −2 ), total energy flux, F net (red; W m −2 ), and downward longwave radiation flux, LW down (green; W m −2 ), averaged over the TNBP region from ERA-5 (positive values indicate downward pointing fluxes).All fields are shown at hourly intervals for the period 29 October 2010 at 00 UTC to 02 November 2010 at 00 UTC.f TNBP spatial extent (km 2 ) estimated using MODIS IST (blue dots) and the UoB SIC (red dots) data.The vertical brown line indicates the time when the TNBP area, as given by the MODIS IST data, was persistently the highest (single occurrence on 31 October)

Fig. 7
Fig. 7 IVT (kg m −1 s −1 ), the shading showing the magnitude and the arrows giving the vectors, on a 26 April 2017 at 12 UTC, b 27 April at 06 UTC, c 28 April at 12 UTC and d 29 April at 12 UTC.The

Fig. 8
Fig. 8 As Fig. 5 but for the period 27-30 April 2017.Hourly data for the Virginia weather station, located at the Inexpressible Island (Fig.3a), is available for this period and is given by the orange line

Fig. 9 a
Fig. 9 a Satellite-derived sea ice drift velocity (in colors) and direction (in vectors) over the 48 h ending on 27 April 2017 at 00 UTC.The drift is defined as the movement of the sea ice during the 48 h prior to the nominal day.The solid pink contour is the 0% sea ice concentration (SIC) contour, and the solid purple contour is the 85% SIC contour.Light blue shading denotes open ocean while ice shelves FiveAWSs in the vicinity of the TNBP were operating during 2010-2017: two at the Inexpressible Island in northern TNB, Manuela (74.946°S; 163.687°E) and Virginia (74.949°S; 163.685°E), and three in southern TNB near Cape Wash- ington: Eneide (74.696°S; 164.092°E);Rita (74.725°S; 164.033°E) at Enigma Lake; Maria (74.626°S; 164.011°E) at Mount Browning.The Virginia, Eneide, Rita and Maria AWS are maintained by the Italian National Programme of Antarctic Research, while the Manuela AWS data is made available as part of the University of Wisconsin-Madison Automatic Weather Station Program.All AWSs measure air temperature, relative humidity, surface pressure and horizontal wind direction and speed at least at hour intervals, with the data freely available online.