Ocean convection linked to the recent ice edge retreat along east Greenland

Warm subtropical-origin Atlantic water flows northward across the Greenland-Scotland Ridge into the Nordic Seas, where it relinquishes heat to the atmosphere and gradually transforms into dense Atlantic-origin water. Returning southward along east Greenland, this water mass is situated beneath a layer of cold, fresh surface water and sea ice. Here we show, using measurements from autonomous ocean gliders, that the Atlantic-origin water was re-ventilated while transiting the western Iceland Sea during winter. This re-ventilation is a recent phenomenon made possible by the retreat of the ice edge toward Greenland. The fresh surface layer that characterises this region in summer is diverted onto the Greenland shelf by enhanced onshore Ekman transport induced by stronger northerly winds in fall and winter. Severe heat loss from the ocean offshore of the ice edge subsequently triggers convection, which further transforms the Atlantic-origin water. This re-ventilation is a counterintuitive occurrence in a warming climate, and highlights the difficulties inherent in predicting the behaviour of the complex coupled climate system.

W ithout redistribution of heat by the atmospheric circulation and ocean currents, only a small portion of the Earth's surface would be habitable. In the Atlantic Ocean the poleward transport of heat is largely accomplished by the Atlantic meridional overturning circulation (AMOC), in which warm waters are transported northward near the surface and cold waters are returned to the south at depth 1 . Most of the North Atlantic deep waters originate from the Nordic Seas 2 , as result of a warm-to-cold transformation that takes place primarily in the eastern part of that region [3][4][5] . The resulting product, referred to as Atlantic-origin water 6 , is returned southward by the East Greenland Current 7 (Fig. 1). This is a main source of dense water to the overflow plume that passes between Iceland and Greenland through Denmark Strait 8 and provides the largest and densest contribution to the lower limb of the AMOC 9 .
The other major contribution to the Denmark Strait overflow is Arctic-origin water formed in the interior Iceland and Greenland Seas 10 . Proximity to the ice edge, where the highest ocean to atmosphere fluxes of heat occur due to frequent intense cold air outbreaks in which frigid, dry air masses are advected over comparatively warm surface water, is an important factor [11][12][13][14] . Diminished heat loss from this region owing to a retreat of the ice edge and different rates of warming of the ocean and the atmosphere will likely reduce the formation of Arctic-origin water 15 . In the Iceland Sea this water mass is primarily produced on the northwestern outskirts of the cyclonic gyre, where the atmospheric forcing is more intense 13,16 . Farther to the west (within the region outlined in black in Fig. 1), the influence of fresh, low-density surface water becomes dominant 17,18 . The presence of this water mass is known to inhibit convection 19,20 , but thus far a lack of wintertime measurements from the westernmost part of the Iceland Sea has prevented verification 13 . Here we show that this fresh, low-density surface layer is not present in the western Iceland Sea in winter, which permits the formation of deep mixed layers offshore of the ice edge. As the ice edge retreats toward Greenland, water masses including the Atlantic-origin water transported by the East Greenland Current that were previously insulated from the atmosphere underneath sea ice are now being ventilated.   visible (θ ≥ 0°C). Instead, the western end of the February 2016 glider transect has an approximately 400 m deep mixed layer with a density of σ θ = 28.01-28.02 kg/m 3 , more than sufficient to supply the dense Denmark Strait overflow plume (Figs. 2b and 3b). As this was still relatively early in the convective season, most likely the depth and density of the mixed layer increased further into winter due to continued heat loss to the atmosphere 13 . The second glider encountered a re-stratifying water column farther to the south in late April, at the end of the convective season, that also shows similarly modified Atlantic-origin water (Figs. 2c and 3c). Both gliders were turned around due to the proximity of sea ice prior to reaching the core of the East Greenland Current. Hence, the extent to which the bulk of the Atlantic-origin water mass was further transformed by convection and how readily the modified waters were transported toward Denmark Strait remain uncertain. However, the glider transects clearly demonstrate that Depth (m) 28 28.03  The western Iceland Sea is a sparsely sampled region, in particular in winter. Nonetheless, using newly available data from instrumented seals 21 in combination with historical measurements 13 , the monthly mean hydrographic properties averaged over the upper 50 m of the water column demonstrated a pattern that is consistent with the shipboard and glider measurements (Fig. 4). The fresh surface layer starts developing in May or June and is most pronounced between July and August with a salinity well below 34. In October the surface layer becomes substantially more saline, and by November a well-defined surface layer is no longer present. At that time the water column has an~80 m deep mixed layer and is well preconditioned for convection.
Local formation of deep mixed layers. In order to investigate whether the deep mixed layers observed at the western turnaround point of the February glider transect may have formed locally, a one-dimensional model (see the Methods section and ref. 22 for details of the model) was employed. The August 2012 profile closest to the turnaround point was used as initial conditions and a constant heat loss of 120 W/m 2 , typical for winter 2015-2016 (Fig. 5b), was applied from November to mid-February. As expected from previous work in this region 19,20 , the fresh surface layer prevented convection, resulting in effect instead in the formation of sea ice as soon as the cooling commenced. This demonstrates that the disappearance of the fresh water is not the result of vertical mixing caused by heat loss to the atmosphere. Another mechanism is instead required to remove the freshwater prior to the onset of convection.
Strong northerly winds are prevalent along the east coast of Greenland in fall and winter 23   transport induced by these northerly winds by assuming an Ekman layer depth of 50 m, and find a pronounced seasonal cycle (Fig. 5a). While there is no substantial Ekman transport in summer, it increases considerably through fall. We estimate that the Ekman transport is sufficient to flush the fresh, low-density layer onto the Greenland shelf, thereby preconditioning the western Iceland Sea region for convection before the atmospheric forcing peaks between December and March (Fig. 5b). The integrated Ekman transport distance through the strong forcing season of order 300 km sets the horizontal scale of the band that is preconditioned for convection. We note that such northerly barrier winds, which often occur along the entire east coast of Greenland in winter 24,25 , inhibit the offshore diversion of freshwater and hence limit its impact on convection in the interior Nordic and Irminger Seas. In 2015-2016 the non-linear combination of unusually strong northerly winds and high heat fluxes at the beginning of winter, along with nearly ice-free conditions, were particularly conducive for convection (Fig. 5).
Melting of sea ice in summer, when onshore Ekman transport is negligible, helps replenish this fresh surface layer 26 .
A new simulation with the one-dimensional model was initialised in which the fresh surface layer was replaced by a mixed layer extending from 80 m depth to the surface, in agreement with the mean November profile from the combined seal and historical hydrographic data set. The resulting simulated profile (red trace in Fig. 6) corresponds very closely to the profile recorded by the glider at the turnaround point in February 2016 (blue trace in Fig. 6), even though the initial conditions were obtained from summertime measurements 4 years prior to the glider transect. This implies that the deep and dense mixed layer recorded by the glider in February 2016 is consistent with local formation as the result of convection offshore of the ice edge.

Discussion
A consequence of the re-ventilation of Atlantic-origin water east of Greenland may have been detected farther downstream. At the Denmark Strait sill there is no discernible seasonal variability in the overflow transport, but a significant seasonal signal in temperature. In particular, the temperature of the Denmark Strait overflow is minimum in September 27 , which, assuming an advective speed of 4-5 cm/s, is consistent with the transport of colder Atlantic-origin water that was re-ventilated the previous winter in the Iceland Sea. Taking into account that the reventilation may also have taken place farther north along the ice edge east of Greenland, this is in reasonable agreement with observed velocities of the Atlantic-origin water in the East Greenland Current 7,8 .
The sea ice concentration in the western Iceland Sea has diminished substantially over the last few decades 15 . There was, in particular, very little sea ice in winter 2015-2016 relative to climatological values (Fig. 5c). As a consequence, increasing areas along the Greenland continental slope, and hence also of the East Greenland Current, are no longer insulated by sea ice from interaction with the atmosphere in winter. If the ice edge continues to retreat toward Greenland, we may expect to see more reventilation of Atlantic-origin water and higher heat fluxes into the atmosphere along the pathway of the East Greenland Current from Fram Strait to Denmark Strait. If such re-ventilation becomes more pronounced, this could further modify the properties of the Denmark Strait overflow and thereby impact the lower limb of the AMOC. While large-scale sea ice loss may lead to a weakening of the AMOC 28 , the ice edge retreat toward Greenland causing increased ventilation of Atlantic water in the Nordic Seas during a warming climate is unexpected and contrary  to the tendency of reduced convection generally predicted by climate models 29 .

Methods
Hydrographic data. The wintertime transects were obtained by autonomous, buoyancy-driven Seagliders that are capable of diving to 1000 m depth 30 . The gliders profile in a sawtooth pattern with a typical vertical to horizontal glide ratio of 1:3 that results in a speed through water of about 20 cm/s. Conductivity and temperature were sampled at 20-80 s intervals, resulting in a vertical resolution of 1-5 m, enhanced within the surface mixed layer. Following a tuning of the flight model parameters for each glider, the data were reprocessed and corrections were applied to compensate for thermal-inertia and flushing speed issues arising from usage of unpumped sensors 31 . Temperature and salinity measurements outside the expected range of values in the Nordic Seas (−2-20°C and 20-36, respectively) were discarded. Each dive/climb cycle was subsequently inspected for density inversions, and measurements causing inversions exceeding 0.05 kg/m 3 were excluded 18,32,33 . The sensors were laboratory calibrated prior to deployment. In addition, the glider data were calibrated against shipboard measurements at the times of deployment and recovery. A rendezvous partway through the deployment was used for data intercomparison. Corrections were estimated in potential conductivity-potential temperature space 34 . Offsets corresponding to 0.006 and 0.016 in salinity were applied to two of the gliders (sg559 and sg562, respectively), while a drift over the entire deployment period corresponding to 0.0008 in salinity was corrected for one glider (sg559). No corrections were applied to the conductivity sensor of the third glider (sg564) or any of the temperature sensors. Vertical sections of potential temperature and salinity were constructed using Laplacian-spline interpolation 7,35 .
Details about the August 2012 shipboard transect can be found in ref. 7 and the calibrated seal data set in ref. 21 . Estimates of wind stress and heat fluxes were derived from short-range forecasts averaged over 6-hourly intervals (using forecast steps for 9-21 h). Only ice-free grid cells, with a sea ice concentration of less than 0.5, were considered.
Ekman transport. The Ekman transport distance was estimated from where τ(t) is the 6-hourly average mean wind stress in the ice-free portion of the western Iceland Sea region (Fig. 1), ρ = 1025 kg/m 3 is the reference density of sea water, f is the coriolis parameter, and n ⊥ is the box mean unit vector perpendicular to Greenland's shelf. For the depth of the Ekman layer a constant value of h = 50 m was assumed, which corresponds approximately to the typical depth of the sharp pycnocline that is associated with the summertime fresh surface layer (Fig. 6). We note that the precise depth of the Ekman layer cannot be determined based on the available observations but the value of 50 m is in line with depth estimates based on the similarity height (e.g., ref. 36 ). Since the Ekman transport distance is inversely proportional to Ekman depth, a decrease of h by 50% would result in an increase of the estimated distance by a factor of 2, whereas an equivalent increase of h would result in a lower distance by one-third.
Mixed-layer model. To implement the one-dimensional PWP 22 mixed-layer model, turbulent heat fluxes, which provide the dominant contribution to mixedlayer deepening 37 , were imposed at the surface at each time step. Initial conditions were obtained from the August 2012 shipboard and February 2016 glider measurements. The depth and properties of the mixed layer were then adjusted until three stability criteria, chief among which is static stability, were satisfied.
Data availability. The glider data can be accessed at the Pangaea repository (https://doi.org/10.1594/PANGAEA.884339). The shipboard data can be obtained from http://kogur.whoi.edu. The seal data can be found in the MEOP database (www.meop.net). The ERA-Interim reanalysis data were obtained from the European Centre for Medium-Range Weather Forecasts and the passive microwave sea ice product 38