Antarctic ice shelves are thinning rapidly due primarily to basal melting, allowing the ice sheets to accelerate and lose mass (e.g. Pritchard et al., 2012) to significantly contribute to the future sea-level rise (e.g. Bamber et al., 2019; Golledge et al., 2019; DeConto et al., 2021). The highest thinning rate has been observed among those ice shelves draining towards the Amundsen Sea (e.g. Rignot et al., 2019; Paolo et al., 2015), where relatively warm modified Circumpolar Deep Water (mCDW) intrudes onto the Amundsen Sea continental shelf via bathymetric troughs, allowing it to come into direct contact with the base of ice shelves (e.g. Rignot et al., 2019; Heywood et al., 2016). The flux of the mCDW entering the ice shelf cavity determines the rate of ice shelf melting (e.g. Jacobs et al., 2011; Dutrieux et al., 2014). Therefore, understanding the local circulation that determines the flow of mCDW and its associated heat transport toward the ice shelves is crucial for a better prediction of ice shelf melting, future sea level and climate.

Models and observations have revealed the presence of a cyclonic gyre in the centre of Pine Island Bay (PIB; hereafter PIB gyre for this gyre). The gyre is well defined between the sea surface and about 700 m depth, in front of Pine Island Ice Shelf (Thurnherr et al., 2014; Heywood et al., 2016). Gyres play an important role in local ocean circulation, distributing heat and enhancing water mass exchange in the Amundsen Sea (Zheng et al., 2021; Schodlok et al., 2012). Schodlok et al. (2012) use a high-resolution model to infer that the strength of this small PIB gyre can be the main determinant of heat transport toward the ice shelf and the associated glacial melt rate (Schodlok et al., 2012). Zheng et al. (2021), Mankoff et al. (2012) and Tortell et al. (2012) also suggest that the PIB gyre entrains water as it exits the ice cavity, contributing to the spreading of glacial meltwater and its associated heat, nutrients and freshwater. Yoon et al. (2022) further mention that the PIB gyre can modulate heat delivery to the Pine Island Ice Shelf.

Despite the importance of gyres near ice shelves, they are poorly observed and modelled as polar oceans are often ice-covered and their bathymetry largely unknown, which limits the regular observations and high-resolution models that resolve small gyres. The formation of the PIB gyre has been attributed to the wind forcing and the meltwater outflow in the southeastern Amundsen Sea (e.g. Thurnherr et al., 2014). Model results from Heimbach and Losch (2012) show that the PIB gyre would be weaker by more than two-thirds if wind forcing were absent. From October 2011 to May 2013, moored current meters in PIB revealed a reversal of the ocean current velocity (Webber et al., 2017), potentially indicating a change of PIB gyre direction from cyclonic to anticyclonic. However, the driver of any reversal is still uncertain as the limited available observations do not show a change in the sign of the wind stress curl field or in meltwater outflow locations.

Sea ice may play a role in circulation in polar oceans by regulating the heat and momentum exchange between the ocean and the atmosphere (e.g. Meneghello et al., 2018). Meneghello et al. (2018) suggest that the interplay among sea ice, wind and ocean can affect the wind-driven Beaufort Gyre dynamics through the influence of sea ice on dampening ocean surface currents, the so-called ice–ocean stress governor. This mechanism focuses mainly on mobile ice and the processes occurring under ice cover (e.g. Meneghello et al., 2018; Elvidge et al., 2016). Mobile ice may drag the ocean and increase the wind stress field felt by the ocean through sea ice (i.e. ocean surface stress, hereafter OSS), while fixed ice, including fast ice and ice shelves, will reduce or entirely block OSS below ice. Nevertheless, relatively little attention has been paid to fixed ice and processes near the boundary of ice coverage in the context of wind-driven polar gyres.

The sparse spatial coverage of observations in these often-ice-covered regions limits our understanding of the mechanisms regulating these gyres, which motivates this study. In Sect. 2, we present a set of new observations of another gyre near Thwaites Ice Shelf (hereafter Thwaites gyre; Fig. 1). Similar to the PIB gyre, Thwaites gyre is influenced by a climatologically cyclonic wind field (which favours cyclonic gyres); however, it is anticyclonic, raising the intriguing question of what mechanism(s) controls the direction of these gyres. We note the different sea ice coverage over the Thwaites and PIB gyres and hence hypothesise that sea ice can influence the OSS to alter the gyre direction. Considering the sparse observations in the Amundsen Sea, we adopt a new approach to this complex question. In Sect. 3, we introduce an idealised model designed to explore the roles of wind and sea ice in determining the gyre features. Section 4 presents the results of the idealised model when different ice conditions and wind stress fields are applied. We discuss the limitations and applications of the results and summarise the results in Sect. 5.

From 25 February to 4 March 2019, the RV Nathaniel B. Palmer collected the first hydrographic dataset to the west of Thwaites Ice Tongue as part of the International Thwaites Glacier Collaboration: Thwaites-Amundsen Regional Survey and Network (ITGC: TARSAN) project. Temperature and salinity profiles were obtained using a Sea-Bird SBE 911+ CTD (conductivity, temperature, depth) profiler with two pairs of conductivity and temperature sensors and then vertically averaged into 1 dbar bins. Ocean current velocities over the upper 980 m of the water column were obtained using a 75 kHz ocean surveyor (OS75) and a 38 kHz ocean surveyor (OS38) shipboard acoustic Doppler current profiler (sADCP). The amount of good-quality data reduced significantly, if not completely, below a depth depending on the sADCP frequency, rolling of the ship, water turbidity and bubbles passing below the sADCP. During this campaign in the Thwaites gyre region, these depths are 400 m for OS75 and 760 m for OS38. Velocity measurements taken below these depths are removed. All velocity measurements are then horizontally averaged into 2km×2km bins. No de-tiding has been applied to the velocity measurements presented in this study because the moored current meters and models in the region suggest that tidal currents are less than 2 cm s-1 (Jourdain et al., 2019), and bathymetry is so poorly known in this region that uncertainties in predicted tidal model currents are of a similar magnitude.

This region is habitually covered by sea ice year-round and has only opened twice since 2000 (Worldview Aqua/MODIS corrected reflectance). The newly obtained velocity dataset reveals the previously unreported Thwaites gyre (blue arrows in Fig. 2). Based on the observations, we identify the centre of Thwaites gyre at -75∘ S, -107.55∘ W. We then calculate the tangential components of ocean current around concentric circles centred on Thwaites gyre and average them into 1 km radius bins (Fig. 3a). Thwaites gyre has an approximate radius of 13 km and can be well identified in the 30 to 430 m depth range covered by the sADCP (Fig. 3a), recirculating about 0.2 Sv (i.e. 106 m3 s-1) of water. The ship CTD survey reveals a detectable density gradient across the gyre (Fig. 3b, c, d). The tangential velocity increases with depth from the near surface to about 130 m, where it reaches its highest speed (about 10 cm s-1, Fig. 3a) and then decreases with depth to zero at about 620 m. We calculate the average vertical shear of the tangential velocities between 3–7 km from the gyre centre to quantify the baroclinicity of the gyre, defined as the vertical gradient of the tangential velocity between the velocity maximum (130 m) and minimum (620 m). The gyre velocity decreases with depth at a relatively constant rate between these depths. The calculated averaged vertical shear is 2×10-4 s-1, i.e. a change of 0.1 m s-1 over 490 m. Note that Thwaites gyre is anticyclonic, despite the local cyclonic wind stress curl, evident in both the climatology (Fig. 1) and contemporaneous observations (Fig. 4), favouring cyclonic gyres.

To identify the role of the gyre in transporting meltwater, and test if meltwater outflow can help to explain the gyre rotation, we calculate meltwater content from temperature and salinity profiles in the gyre region using the composite-tracer method (Jenkins, 1999; Pink dots in Fig. 2). In this calculation, we use three water masses including mCDW, Winter Water and glacial meltwater, and the two tracers conservative temperature (Θ) and absolute salinity (SA), defined following the Thermodynamic Equations of Seawater-10 standard (McDougall and Barker, 2011). Both tracers are assumed to be conservative for all observations. We chose the endpoints following previously published research. The endpoints of mCDW (Θ=1.044∘ and SA=34.8795) are consistent with Wåhlin et al. (2021), and the endpoints of Winter Water (Θ=-1.86∘, SA=34.32) and glacial meltwater (Θ=-90.8∘, SA=0) are the same as used by Zheng et al. (2021) and Biddle et al. (2019). The fraction of meltwater can be derived from observations with the equation below: 1φmeltwater=Θobserved-ΘmCDW-SAobserved-SAmCDW×ΘWW-ΘmCDWSAWW-SAmCDWΘmeltwater-ΘmCDW-SAmeltwater-SAmCDW×ΘWW-ΘmCDWSAWW-SAmCDW, where φmeltwater is the meltwater fraction, and Θ and SA with subscripts define the conservative temperature and absolute salinity endpoints of each water mass.

The highest meltwater content is detected in the southeast of the Thwaites gyre (Fig. 2). This is consistent with observations collected by an autonomous underwater vehicle presented in Wåhlin et al. (2021) that suggest a north-westward meltwater-rich outflow emanating from the cavity beneath Thwaites Ice Tongue. The Thwaites gyre may entrain this meltwater plume and thus play a role in circulating meltwater near Thwaites Ice Shelf and boost water-mass mixing.

Although a previous study suggests that the buoyancy of glacial meltwater at depth may facilitate gyres (Mathiot et al., 2017), Thwaites gyre is not likely to be meltwater-driven. As glacial meltwater plumes coming out from the base of glacier are more buoyant than the ambient water, they rise and turn left due to Coriolis force, as seen in Pine Island Bay (e.g. Thurnherr et al., 2014; Zheng et al., 2021). Glacial meltwater coming out from Thwaites Ice Tongue to the southwest of the Thwaites gyre will therefore impede the anticyclonic gyre, rather than accelerate it. Hence, neither the cyclonic wind stress curl shown in Fig. 2 nor the meltwater discharge can directly generate an anticyclonic gyre. Therefore, we explore other factors that might explain this apparent contradiction.

Sea ice coverage is often remarkably different between the PIB and Thwaites gyre regions (Fig. 1). Satellite imagery shows that the PIB gyre region was generally open during the whole summer of 2009 (Worldview Aqua/MODIS corrected reflectance), when the PIB gyre was firstly observed (Thurnherr et al., 2014). At the same time, fast ice covered most of the Thwaites gyre region, and the sea ice did not open until late January 2019 (Fig. 4a), about a month before the sADCP survey revealed the gyre (in late February to early March 2019). During February, ice coverage in the Thwaites gyre region changed from covering the western part of the Thwaites gyre (1 February 2019, Fig. 4b) to completely open (12 February 2019, Fig. 4c). Sea ice covered the western part of the Thwaites gyre again (23 February 2019, Fig. 4d) 2 d before the start of the sADCP data collection.

The presence of sea ice alters OSS (e.g. Elvidge et al., 2016; Meneghello et al., 2018). Thus, we hypothesise that the sea ice coverage may mediate the OSS and the resulting surface stress curl felt by the ocean (i.e. ocean surface stress curl, hereafter OSSC) sufficiently to reverse a gyre, leading to the different PIB and Thwaites gyre directions. To test this hypothesis, we use an idealised model to reproduce wind-driven gyres and run a set of conceptual experiments to simulate the response of wind-driven gyres to different sea ice coverages.

Regional circulation models have often been used to study ocean gyres (e.g. Meneghello et al., 2021; Regan et al., 2020). However, to comprehensively explore the impact of the ice coverage on the gyre formation through its modification of the stress imparted to the ocean, we need to isolate the individual forcings and conduct a very large number of experiments to examine how they interact. We therefore invoke an idealised model that represents the surface forcing in a simple manner, excluding other features of the real ocean such as baroclinicity, ice-shelf processes and topography. Some biases may occur due to the lack of those mechanisms. For example, the meltwater injection that is thought to facilitate gyres (Mathiot et al., 2017) is not included, which may explain the differences in the magnitude of Ψ between the simulated gyres (0.58Sv for PIB and -0.42Sv for Thwaites) and gyre observations (about 1.5Sv for PIB, Thurnherr et al., 2014, and -0.27Sv for Thwaites). Our argument is not to dispute the effect of meltwater, but rather to highlight the role of wind forcing and the sea ice conditions.

Since our model has much lower computational costs than other regional models, we can run our model hundreds of times to test the gyre response in different wind–ice combinations and apply the results in different polar oceans under varying conditions. All of our simulated gyres reached steady state within 2 months and respond to changed surface conditions on a similar timescale. We also tested a 1.5-layer reduced-gravity model as a comparison for the barotropic case presented here, with all forcings and model design remaining the same. Although the baroclinic model produced gyres with more intensified surface currents and a slightly longer spin-up time, the gyres in baroclinic and barotropic model cases have the same direction and similar sizes and transports. To explore the sensitivity of the model results to the width of this marginal ice zone, we created two new ice conditions with the smaller gradients (not shown) in surface stress and weaker OSSC over three grid points (3 km) and four grid points (4 km). The simulated gyres have almost the same strengths (both about -0.41 Sv) and same shapes as the simulated gyre from simulations without a wider marginal ice zone. This indicates that the width of the marginal ice zone is not important for gyre generation. Hence, our model reproduces the observed gyre direction and response to ice coverage well.

The influence of mobile ice on changing OSS has been well studied (e.g. Meneghello et al., 2018) while little progress has been made towards fixed ice. We have considered both the increase and decrease in OSS, which included scenarios caused by both mobile ice and fixed ice altering OSS. We further discussed the significant effects of the associated OSSC along the ice boundary on gyre formation and gyre features. Our results are especially useful in Antarctic continental shelf seas where climatological winds often blow offshore from the ice shelves/fast ice to the ocean, which allows the gyres near fixed ice to fully develop.

In Antarctic continental shelf seas, gyres near ice shelves contribute significantly to the spreading of glacial meltwater and its associated heat and nutrients (Zheng et al., 2021; Mankoff et al., 2012). Meltwater can impede sea ice formation, causing polynyas, or alter the locations and timing of hotspots of marine productivity (Zheng et al., 2021; Mathis et al., 2007; Mankoff et al., 2012). Gyres may transport heat, which is carried by warm water entrained into the gyre, towards ice shelf cavities (Schodlok et al., 2012). The existence of gyres can cause isopycnal displacement. For cyclonic gyres, isopycnals shoal in the gyre centre, as regularly observed in Pine Island Bay (e.g. Thurnherr et al., 2014; Zheng et al., 2021; Dutrieux et al., 2014; Heywood et al., 2016); for anticyclonic gyres, isopycnals deepen in the gyre centre. Below warm-cavity ice shelves, the water is stratified with a fresh meltwater-rich upper layer and a warm yet salty mCDW lower layer. This isopycnal displacement may allow warmer mCDW to enter the ice cavity and melt ice shelves (Yoon et al., 2022). The intrusion of warm water into the base of warm-cavity ice shelves is via the dense lower layer, the so-called “salt wedge” (Robel et al., 2022). The water mass exchanges due to gyres may impact the stratification in front of the ice shelve and hence affect this salt wedge and the related intrusions of warm water to ice shelves. Similarly, gyres formed near the sea ice edge can circulate ambient water masses. The redistribution of water masses and the heat and freshwater might feed back onto sea ice formation to affect the ice formation and regeneration of ice cover in the next winter. By changing both the ocean stratification and the sea ice cover, gyres will affect heat fluxes in the vicinity of ice shelves, which may in turn influence the heat available for basal melting (e.g. St-Laurent et al., 2015; Webber et al., 2017). Therefore, it is important to understand the conditions that can generate or modify such gyres.

Due to the importance of gyres for the regional ocean environment, more observations and simulations are required to better understand the relationship between surface conditions and gyre strength and direction. We highlight the importance of wind–ice–ocean interactions, especially the wind stress curl at the ice edge, to polar ocean gyres. These interactions occur at small scales (of the order of tens of kilometres) that will be poorly resolved by global coupled models. Simulating these processes is also dependent on accurate sea ice conditions including the representation of fast ice and polynyas. Gyres in typically ice-covered regions (such as Thwaites gyre) present extreme challenges for repeat ship-based surveys. We show that the sea ice coverage can change rapidly (Fig. 4) and has a large influence on the ocean surface stress (Figs. 10–13); however, it is yet difficult to monitor. The PIB gyre changes its location interannually and seasonally (Heywood et al., 2016; Zheng et al., 2021), further demonstrating the need for long-term continuous monitoring. Such continuous gyre observations would also be useful for model evaluation and could be obtained with high-resolution sea ice motion from satellites, which can clearly show the surface current through sea ice movement. We urge that further effort is needed for improving the quality of sea ice satellite products, especially the data coverage, operating frequency and spatial resolution. Future work should also include quantifying the effect of different types of sea ice on enhancing or reducing the ocean surface stress, which may be made by the combination of ice-tethered profilers providing under-ice current velocity and autonomous surface vehicles providing the near-surface wind speed and heat fluxes.

Despite the important role that gyres play, small wind-driven gyres have received limited attention in the Antarctic continental shelf seas. There are a few other gyres documented in the Antarctic continental shelf seas, such as the Prydz Bay Gyre (Smith et al., 1984), and the cyclonic gyre in front of Filchner Ice Shelf (Foldvik et al., 1985), but none of them are primarily driven by local wind, so they are not discussed here. Model representations of ocean currents show the existence of some small gyres with a radius of about 15–30 km, including PIB gyre, in Antarctic continental shelf seas (e.g. Fig. 4c, d in Nakayama et al., 2019). However, except for the PIB gyre, which has been observed in several different years, little attention has been paid to such gyres and their formation mechanisms. This is partly because polar observations often cover either too short a time period or too small an area to provide in situ verification for the gyres found in models. Our study provides a possible mechanism to explain the formation of the gyres formed near the ice–ocean boundary that might be explored for other small gyres shown in high-resolution ocean model results in Antarctic continental shelf seas.

This study uses new observations to identify the Thwaites gyre for the first time, located in a habitually ice-covered region of the southeastern Amundsen Sea. This gyre rotates anticyclonically, despite the climatological cyclonic wind forcing that implies the gyre should rotate cyclonically, as is the case for the only other gyre reported in the eastern Amundsen Sea, PIB gyre (e.g. Thurnherr et al., 2014; Heywood et al., 2016). To investigate this apparent discrepancy, we use a barotropic model with idealised sea ice and wind forcing only to simulate gyres similar to those observed in the vicinity of ice shelves. Our model suggests that sea ice plays a key role in mediating the wind stress transferred to the ocean and hence determines the direction and strength of the gyre rotation. The percentage of wind stress transferred to the ocean, and the angle between the wind direction and sea ice edge, can alter the OSSC over ice-covered regions and along sea ice edges sufficiently to reverse gyres. Although the simulated gyres are slower than those observed, we demonstrate the potential of sea ice to control gyre direction and intensity. We suggest that these processes may explain gyre formation and reversal in polar oceans, for example, the PIB gyre reversal hypothesised by Webber et al. (2017). We further suggest that this wind–ice–ocean interaction may contribute to the development of gyre features throughout the polar oceans.