Geothermal heating and episodic cold-seawater intrusions into an isolated ridge-flank basin near the Mid-Atlantic Ridge

Six-year records of ocean bottom water temperatures at two locations in an isolated, sedimented deep-water (∼4500 m) basin on the western flank of the mid-Atlantic Ridge reveal long periods (months to >1 year) of slow temperature rises punctuated by more rapid (∼1 month) cooling events. The temperature rises are consistent with a combination of gradual heating by the geothermal flux through the basin and by diapycnal mixing, while the sharper cooling events indicate displacement of heated bottom waters by incursions of cold, dense bottom water over the deepest part of the sill bounding the basin. Profiles of bottom water temperature, salinity, and oxygen content collected just before and after a cooling event show a distinct change in the water mass suggestive of an incursion of diluted Antarctic Bottom Water from the west. Our results reveal details of a mechanism for the transfer of geothermal heat and bottom water renewal that may be common on mid-ocean ridge flanks. Bottom water renewal on the western flank of the mid-Atlantic Ridge is driven by rapid intrusion of cold, dense water enabled by gradual warming due to geothermal heat flux and diapycnal mixing, as revealed by six-year observational records.

A number of past studies have presented evidence for an important role for geothermal heating of ocean bottom waters in hydrothermal, biological, and physical oceanographic processes associated with hydrothermal systems in oceanic crust. These have been documented particularly well in closed or semi-restricted axial rift segments on the Mid-Atlantic Ridge 1,2 and Juan de Fuca Ridge 3 . They have also been documented in two semi-restricted, well sedimented young ridge-flank systems where geothermal fluxes are high: the eastern flank of the Juan de Fuca Ridge, where there are known off-axis vents at some volcanic edifices not covered by sediments 4 , and the Panama Basin, which encompasses three spreading center segments and young ridge flanks 5,6 . Buoyancy-driven overturn of geothermally heated bottom waters has been suggested to be an important contributor to the global meridional overturning circulation [7][8][9][10] . There remains debate about the magnitude of the geothermal effects, and detailed time-series records of the process have not been reported to date. Here, we present long-term records (from 2011 to 2017) of bottom-water temperatures in an enclosed, sedimented basin called "North Pond" located on the western flank of the Mid-Atlantic Ridge, along with conductivitytemperature-depth (CTD) profiles collected in the center of the basin during an expedition in 2014. These data show evidence for slow geothermal heating of the deep bottom waters, likely combined with a downward heat flux by diapycnal mixing induced by flow over the rough seafloor topography, that is interrupted at seasonal to interannual time scales by gravity currents delivering cold dense bottom water into the basin from the west. This kind of process may be generally important on the flanks of mid-ocean ridges for the transfer of geothermal heat to, and renewal of, ocean bottom waters. In the North Pond setting, we infer that the gravity currents transport diluted Antarctic Bottom Water (AABW) into the basin, so the results may also have implications for the AABW budget in the western North Atlantic Ocean.
North Pond geological, hydrographic, and geothermal setting. First mapped and named in 1974 11 , North Pond is a roughly 8 × 15 km basement depression filled with sediments up to 200-300 m thick on crust 7-8 million years old located on the western flank of the Mid-Atlantic Ridge at ∼23°N (Fig. 1). The basin is elongated in a south-southwest to north-northeast direction, subparallel to the mid-Atlantic Ridge, and is surrounded by thinly sedimented basement topographic highs with patches of exposed basement. The deepest sill is on the southwest side, elevated ∼100-200 m relative to the 4400-4500 m average sedimented seafloor depth within North Pond. A series of deep basins runs from the southwest sill area to the west (Fig. 1) and represents a deep pathway for cold, dense bottom water to reach as far as North Pond from deeper water to the west. This chain of basins is comparable to the ridge-normal canyons and valleys that cross-cut the western flank of the Mid-Atlantic Ridge in the South Atlantic and have been shown to allow flow of deep waters from the west toward the ridge-crest 12,13 . In the region west of North Pond, the densest global bottom water mass in the oceans, Antarctic Bottom Water (AABW), has been documented to flow northward to at least 30°N in water depths a few hundred meters deeper than North Pond [14][15][16] . Some vertical mixing between the AABW and the overlying North Atlantic Deep Water (NADW) is likely in the deep waters of the North Atlantic at the depth of the North Pond sill at the North Pond latitude 14 . In the Kane Fracture Zone ∼110 km north of North Pond, AABW has been mapped to flow eastward to North Pond longitude at depths as shallow as 4600 m 17 .
In the mid-1970s, the Deep Sea Drilling Project cored 576 m of basement near the southeast edge of the pond beneath 92 m of sediment in Hole 395A 18 . Since then, North Pond has been revisited multiple times for geophysical surveys and further scientific ocean drilling, and it has now become an important type location for studies of low-temperature ridge-flank hydrothermal circulation in thinly sedimented crust formed at slow spreading rates. Over 100 sediment probe measurements of seafloor heat flux distributed throughout the sediment pond [19][20][21][22] have documented an average heat flux of ∼40 mW m −2 , much less than the ∼180-190 mW m −2 predicted by conductive plate cooling models for crust of its age 23 . Higher values were measured near the northwest and northern edges of the sediment pond, and two values greater than the predicted value were measured on the slopes of the basin, one below sill depth 24 . These results, combined with two decades of observations of flow of cold ocean bottom water down Hole 395A into permeable uppermost basement [25][26][27][28] , suggest a vigorous low-temperature lateral circulation system within the permeable section of basement beneath the North Pond sediment cap, generally running from south to north, with advective release probably occurring through the thinly sedimented borders of the sediment pond 19,20,22,24 .
Since 1997, the North Pond low-temperature ridge-flank hydrothermal circulation has been investigated with four sealed borehole Circulation Obviation Retrofit Kit (CORK) hydrogeological observatories 29,30 focused primarily on subseafloor pressure and temperature measurements and geomicrobiological sampling from basement formations. Three of the installations 31,32 provided long time-series measurements of bottom water temperatures using platinum resistance temperature detectors (RTDs) from 2011-2017 that we analyze in this paper. We also utilize CTD data 46  collected during a submersible revisit to the CORKs in spring of 2014. Further details are provided in the Methods section.

A: Effects of geothermal heating on North Pond bottom water.
The six-year records of bottom-water in-situ temperatures from the main CORK installations at Holes U1382A and U1383C ( Fig. 2) show evidence of a broad range of time scales in North Pond bottom-water temperature variability, with very similar magnitudes and patterns in the temporal variations at the two sites. Because the holes are about 6 km apart, we infer that this is a basin-wide effect. Much of the difference in absolute values of the bottom water temperatures recorded at the two holes is likely due to RTD calibration uncertainties described in the Methods section, but relative variations are resolved much more accurately. In addition, over North Pond there is small adiabatic increase of in-situ seawater temperatures below ∼4100 m 24 that would produce a difference of in-situ temperature of the order of 0.01°C between the two sites. Nearly identical variations to those at Hole U1383C were also recorded by the temperature sensor in the Hole U1383B data logger located ∼25 m from Hole U1383C, but these are not shown to avoid cluttering Fig. 2.
The general pattern includes long periods (up to 1½ years) of gradual heating, interspersed with abrupt drops in temperature over 2-4 weeks. There appear to be four major episodes of abrupt temperature drops in the record, mainly in late spring or early summer, as well as several smaller drops at other times of the year. The rates of temperature fall average roughly 0.25°C yr −1 , whereas the rates of temperature rise during the periods of gradual heating average about 0.02°C yr −1 , over an order of magnitude less (Fig. 2). Using values of ∼3870 J kg −1°C−1 for the specific heat, C p , and ∼1048 kg m −3 for the in-situ density, ρ, of North Pond bottom waters calculated from the CTD data described in the next section, the measured warming rate would require a heat input of ∼2.6 mW m −3 of heated bottom water.
If we assume horizontally uniform heating within the basin, the time rate of change of temperature, T(z, t), over the depth range, Δz = h, above the seafloor is given by the one-dimensional heat balance where Q sf is the conductive seafloor heat flux, Q h is the vertical heat flux at elevation, h, above the seafloor, and ρC p = 4.1 × 10 6 J°C −1 m −3 . Using (1) and assuming, for now, that there is zero flux from processes within the overlying water column (i.e., Q h = 0), the basin-wide average Q sf ≅ 40 mW m −2 and the observed rate of temperature rise 0.02°C yr −1 are consistent with a uniform geothermal heating of the deepest ∼15 m of the basin, easily spanning the depths of the CORK data loggers. However, the average measured conductive flux is almost certainly a lower limit to the total geothermal heating of the bottom waters below the deepest sill depth to the southwest of North Pond, given that it is much less than the predicted plate cooling value of 180-190 mW m −2 . Lateral flow of formation fluids in the permeable basement is likely to advect a significantly greater heat flux to the surrounding thinly sedimented basement exposures surrounding North Pond, where heat can be transferred to the ocean by higher conductive flux and by advection, e.g., as at the locations of two areas of diffuse lowtemperature venting respectively on and 3 km from the northwest edge of the sediment pond mapped in 2014 24 . The salinity of the vented fluids is essentially identical to that of the bottom water because it is a cool seafloor hydrothermal system 33 , in which the temperature increase relative to bottom water is small, lateral flow is rapid, and the residence time in the subseafloor is short. At this location, fluids collected from the basement aquifer using the CORKs did not deviate from major ion concentrations in the bottom seawater 34 . Therefore, any venting fluids in this system will have very little effect on the salinity of the bottom water. Some of the advective flow may be released via surrounding basement highs that are shallower than the North Pond sill depth, so we cannot resolve the true magnitude of the heat that is absorbed by the North Pond bottom waters below sill depth, but an upper limit is provided by applying the total predicted lithospheric heat flux to the region below sill depth at and around North Pond. This would raise the estimate of the thickness, h, of the solely geothermally heated water layer in North Pond by a factor of nearly five, to h = 70-75 m.
There is most likely an additional component of heating of the North Pond waters below sill depth by mixing from above, as reported for a ∼300 m thick geothermal boundary layer on the flank of the Juan de Fuca Ridge in Cascadia Basin 4  internal waves, shear-instability and other topographically induced processes 9 , the heat flux, Q hs at the sill depth, h s = 150 m (a representative depth for North Pond) that is needed to support uniform heating of 0.02°C yr −1 within the basin, is given by For the low-end seafloor heat flux Q sf = 40 mW m −2 , the estimated downward heat flux associated with diapyncal mixing, Q hs = −350 mW m −2 , while for the more likely high-end seafloor heat flux, Q sf = 190 mW m −2 , the estimate for diapycnal mixing decreases to Q hs = −200 mW m −2 . Thus, to maintain a temperature rise of 0.02°C yr −1 throughout the basin requires a downward heat flux from diapycnal mixing that is, at minimum, comparable to the geothermal heating. On the basis of this analysis, we propose that both seafloor heating and diapycnal mixing are contributing significantly to warming of the North Pond water following a cold-water intrusion. Given the uncertainty in the advective geothermal flux into North Pond, we cannot rule out the possibility that diapycnal mixing is more important than geothermal heating but, in any reasonable scenario, it is likely that most of the geothermal heating component would remain trapped in the North Pond bottom waters until there is an influx of cold, dense deep water into the basin.
B. CTD evidence for renewal of North Pond bottom waters in spring 2014. Although none of the four major drops in temperature shown in Fig. 2 occurred during expeditions to the location that conducted CTD casts, there was a smaller fall in temperature observed in early April 2014 between CTD casts on 29 March and 10 April. These casts (Fig. 3a) showed a clear decrease in Conservative Temperature, similar in magnitude to the decrease of in-situ temperature seen in the time series (Fig. 3b), and increases in neutral density and oxygen saturation in the deepest ∼130 m of the basin up to 4330 m water depth. These observations are all consistent with the interpretation of an influx of cold, dense water over the sill.
Especially noteworthy in the pre-incursion 29 March CTD data are the nearly uniform profile of Conservative Temperature up to the sill depth and a small but significant change in gradient in oxygen saturation at sill depth. The 29 March profile of neutral density is also quite uniform up to sill depth, which is not unexpected given that both salinity and temperature are uniform with depth in the lower layer. In sum, the 29 March profiles indicate that the sill depth marks the upper boundary of the warmed, mixed bottom layer in North Pond prior to the influx of cold, dense bottom water. The 10 April profiles indicate an upward displacement of the warmed, mixed layer by about 25-50 m; however, this may represent a one-time snapshot of a temporally variable response of the structure during the incursion event, but not the post-incursion equilibrium state.
The Absolute Salinity traces show little relative change, except for a slight increase in the bottom part of the 10 April profile that is coincident with the Conservative Temperature decrease below ∼4330 m water depth. The general offset between the two Absolute Salinity traces (0.004 g kg −1 or roughly one part in 10 4 ) is within the normal error range for salinity between separate casts. As neutral density and Absolute Salinity both depend mainly on practical salinity and location, the neutral density traces also are offset but we focus here on relative changes, which are well resolved. Between renewal events, North Pond bottom water temperatures are continuously modified by the processes described above and oxygen may be consumed. However, salinity probably remains more constant, except for small increases during any incursions, given that venting low-temperature hydrothermal fluids are essentially equivalent to bottom seawater as noted above.
The properties of the inflowing waters are likely intermediate between end-member AABW and lower-bound NADW water masses, and would suggest a mixed origin in this instance. The intermediate Conservative Temperature, the elevated dissolved oxygen content, and the location of the site suggest a contribution from AABW that has been found to extend eastward immediately to the west of the North Pond location, with potential temperatures of up to ∼2°C 17 . In the Discussion, we return to the issue of the identity of the cold intruding water.
Whether the cold-water events are "pulled" or "pushed" is an important distinction to be made 35 . A "pulled" mechanism was provided by Turner 36 , who described the release of discrete thermals above a horizontal surface heated from below: "…for most of the time the process of transfer near the boundary is one of conduction, followed by a comparatively short interval during which the conditions are locally restored to the original uniform state by the removal of the buoyant fluid as a plume or thermal." This might suggest that the buoyant rise of discrete thermals in North Pond pulled in colder bottom water over the bounding sill. However, a more detailed analysis of the records described in the next section strongly suggests a more dynamic "push", whereby density-driven inflows of colder bottom water flush the deepest waters within North Pond, displacing the heated bottom waters upward.
C: Hydrodynamical processes and gravity current inflows. The thermal records from Holes U1382A and U1383C (Fig. 2) consist of temporal (t) variations in bottom water temperature, T(t), ranging from semi-diurnal (M 2 ) tidal periods to seasonal and interannual time scales. A correlation analysis of the 6-year records reveals a distinct peak time-lag of ∼1.6 days between the two sites, with U1382A leading U1383C (Fig. 4a). However, the central peak of the correlation function, C(τ), as a function of time lag, τ, is quite broad (symmetrical, with a half-width of 7 days) owing to contributions from tidal currents, near-inertial motions, noise, and other processes. Consequently, we undertook a visual examination of the time lags between the two sites on an event-by-event basis (which is straightforward given the limited number of major and intermediate cold-water events) and found a consistent lag of around 2 days, with a few exceptions such as the time difference of 3.4 days for the main event in 2016 ( Fig. 4b-d). After adjusting for the average time-lag between the two sites, temperature variations at U1383C track those at U1382A in considerable detail, indicative of a basin-wide response at time scales ranging from days to seasonal.
Based on the strong tidal signals in the temperature records, it is readily apparent that the high frequency temperature variations are caused by currents, u, that transport the bottom water back and forth past the sensors in the presence of a horizontal temperature gradient, ∇T, with the functional relationship between current and changing water temperature being ΔT = (u·∇T) for each 2-min time step, Δt. (Thermal sensors can only detect currents where there is a resolvable spatial gradient in water temperature, which means that temperature fluctuations at frequencies higher than primary tidal frequencies are poorly resolved during periods dominated by gradual geothermal warming when the heat is most uniformly distributed in the basin.) Superimposed on the tidally induced variations are large amplitude quasi-annual temperature changes consisting of a slow rise in temperature, followed by abrupt declines in temperature roughly mid-way through the calendar year. This pattern was observed in all years except 2012 and 2015, when there were no major disruptions in temperature rise. The pronounced seasonal to interannual variability indicates that the flow dynamics in North Pond are linked to large-scale oceanic circulation processes.
Each major temperature event was followed by smaller amplitude intra-seasonal (3-4 month) fluctuations that mimic the profile of the longer events. As noted above, temperature increases occurred over durations of several months to over a year at an equivalent heating rate of ∼0.02°C yr −1 . In contrast, temperature declines had short durations of about a month at an Close examination of the temperature time series during major events (when the flow dynamics are made most apparent because of the formation of a bottom water temperature gradient) also shows temperature modulation at the spring-neap (fortnightly) tidal period of 14-days. The presence of a significant fortnightly period in the tides is confirmed by the bottom pressure records from the basin (Fig. 3b). Thus, bottom water movements in North Pond appear to consist of fortnightly modulated semidiurnal tidal currents on which are superimposed major event-like flows of relatively cold water at intervals of months to years. During major cold-water events, we also observe weaker ∼7-day variations in temperature within each fortnightly cycle that appear to be related to the dynamics of the cold-water flows.
Our interpretation of the temperature observations is that they comprise a background state, consisting of extended periods of gradual warming of the bottom water by the seafloor heat flux and diapycnal mixing, that is interrupted by a renewal state, during which the basin is flushed by intrusions of colder, denser bottom water that has spilled over the deepest part of the sill into North Pond from a source farther west on the flank of the Mid Atlantic Ridge (Fig. 5). In this model, bottom water renewal can be triggered once the dense water from the west reaches the elevation of the deepest gap in the sill in the southwest corner of the basin, and/or once the top of the heated, vertically mixed bottom layer in the North Pond reaches the elevation of the sill. For either mechanism, gravity-driven inflows occur once the density difference between water inside and outside the basin is sufficiently high to establish an inward pressure gradient. Two observational features support the concept of intrusive dense water inflow through the deepest part of the sill at the southwest corner of the basin: (1) all events are first observed at U1382A and then at U1383C 1.5-3.5 days later; and (2) the temperature drops observed at U1382A are considerably greater than at U1383C, indicating that the bottom water mixes with warmer overlying water as it advances slowly from southwest to northeast, or circles in a counterclockwise direction around the bottom of the basin. For example, the initial temperature drop near the beginning of 2016 (Fig. 2) was 0.0077°C at U1382A but had diminished to 0.0023°C by the time the density flow reached U1383C about 3.4 days later (average travel speed = 0.02 m s −1 ). Many shortperiod events in the temperature decreases can be correlated between U1382A and U1383C (Fig. 4). These typically show lower amplitudes by factors of 2-3 and time lags of a few days at U1383C relative to U1382A. Average travel speeds for the observed time lags of 1.5 to 3.5 days range from 0.046 to 0.020 m s −1 , respectively.
The quasi-weekly and quasi-fortnightly modulations of the intrusive events in North Pond are likely caused by variations in Fig. 4 Correlation function and example time lags between bottom-water temperatures at Holes U1382A and U1383C. a The correlation function between temperatures at the two holes as a function of time lag. b One month record of in-situ bottom water temperatures at Holes U1382A and U1383C from late March to late April, 2014, spanning the early April cooling event and two CTD casts. DG = data gap on data download, with brief spike on resumption of data logging. c, d Expanded three-month views of the major decreases in bottom-water temperatures in 2013 and 2017, respectively, illustrating typical 2-to 3.5-day lags of events at Hole U1383C. In b-d, the U1382A temperature record is shown in red, the U1383C record is shown in blue, and vertical arrows indicate example events seen at both locations.
tidal current mixing over the deepest portion of the sill leading into the basin from the southwest. As in coastal waterways, density currents readily flow over entrance sills into adjoining basins during neap tides when vertical mixing is weak, but are blocked from crossing the sill during spring tides when vertical mixing is strong 37,38 . The presence of a weekly signal in the temperature records for the two CORK sites during an intrusion adds further support to the tidal mixing concept in that it indicates that dense-water inflow during neap tides occurs during both the turn to an ebb current and the turn to a flood current. Because the neap tidal currents would be so weak (possibly near zero velocity) during the turns, dense bottom water would have no difficulty slipping over the sill into the adjoining basin at 7-day intervals. Similar processes have been proposed for the density currents observed at 4300 m depth at the southern end of the Middle America Trench off Costa Rica 39,40 , although in that case, primary forcing originates from equatorial Kelvin waves, and more importantly, turbidity is involved and heat is carried down to deeper levels.
Once the gravity current begins to descend into North Pond, its speed, v g , would be determined by its thickness, h, the water depth, H, and the density difference, Δρ, between the intrusion and the ambient water. For the present case, H>> h, whereby where g 0 ¼ g Δρ=ρ is reduced gravity, g is the acceleration of gravity, ρ is the density of the ambient bottom water, and F r is the Froude number (which is of order unity for weakly stratified  (2) the presence of cold, denser bottom water that has shoaled to depths of around 4300 m along the western flank of the ridge. The arrival of dense bottom water outside the sill in late spring and early summer in 4 of the 6 years investigated is clearly related to larger scale processes in the Atlantic Ocean. As noted earlier, the properties of the April 2014 density current and the regional hydrography suggests that the observed inflow consists of AABW (with elevated oxygen and low Conservative Temperature) that had mixed with overlying NADW in the deeper region immediately to the west of the ridge. From geostrophic dynamics, we would expect the Coriolis effect to cause the northward flowing AABW to favor an eastward flow up the ridge flank. In contrast, the southward-flowing NADW would tend to be forced to the west, away from the ridge flank. If our inference of gravity-driven inflow of modified AABW is correct, then the absence of renewal events in 2012 and 2015 suggests that the AABW source water had not reached as far north and east as North Pond in those two years or was not dense enough to intrude to the bottom of North Pond. In the other four of the six years recorded, there were large-scale intrusions midway through the calendar year that were preceded by one or two shorter duration, smaller amplitude intrusive events; these "precursor" events could indicate that northward and eastward advance of the AABW was still underway. As noted earlier for the Cascadia Basin in the northeast Pacific, heating of the bottom waters by off-axis hydrothermal plumes, the background geothermal flux of∼0.3 W m −2 , and mixing by turbulent motions produces a 250-350 m thick geothermal boundary layer that is clearly observed in CTD data and characterized by reduced vertical stability 4 . Our CTD data from North Pond indicate an analogous geothermal boundary layer up to the deepest sill depth on 29 March 2014, before its displacement by the early April incursion of cold, dense bottom water. Given its very low vertical gradient in neutral density, the section up to roughly 25 to 50 m above sill depth would seem to correspond to the weakly stratified bottom boundary layer as defined for the global ocean 42 . If so, the thickness of that boundary layer would be on the order of 150-250 m above the North Pond seafloor.
Our results do not provide quantitative constraints on the magnitude of the contribution of the geothermal flux from the flank of the Mid-Atlantic Ridge to the global ocean overturning circulation. However, they do document important details of the transfer of the buoyancy arising from geothermal heating to the waters overlying a well-studied basin on the young flank of a slow-spreading ridge, and associated episodes of renewal of basin bottom waters by gravity flows of colder, denser regional bottom waters over the basin sill. That kind of topographic structure is typical on the flanks of slow-spreading ridges in the Atlantic and Indian oceans, so these processes may be common globally. In addition, our interpretation suggests that temporal variations of bottom water properties in North Pond and the series of basins to its west could be diagnostic of the strength of the northward flow of AABW in the North Atlantic west of the Mid-Atlantic Ridge. In that respect, North Pond and basins to its west seem to be intriguing candidates for further temporal monitoring with a more extensive set of instrumentation. This should include arrays of current meters with temperature and salinity sensors, spanning the deepest 200-300 m of the water column, to better delineate the dynamical processes implicated by our limited datasets. Such data could help constrain the degree to which AABW plays a role in the intrusions we have documented, and further elucidate the processes of geothermal heating of the basin bottom waters and their renewal by gravity current intrusions of cold, dense regional bottom waters over basin sills.

Methods
A. CORK bottom-water temperature measurements. The long-term CORK hydrological observatories are designed mainly to sample and monitor subseafloor state and hydrological processes. The configurations of the CORKs installed from the scientific drillship JOIDES Resolution in Holes U1382A and U1383C and "CORK-Lite" deployed in Hole U1383B by remotely operated vehicle (ROV) are described in detail elsewhere 31,32 . For the installations at Holes U1382A and U1383C, the primary CORK pressure and temperature data loggers are mounted ∼4 m above seafloor, or at water depths of ∼4479 m at Site U1382 and ∼4409 m at Site U1383 (Fig. 1b). Hole U1383B lies ∼25 m from Hole U1383C, and its data logger lies about 2.5 m above seafloor. All three data loggers include RTD temperature sensors thermally coupled to the endcaps of the pressure cases that were sampled at 2-minute intervals to provide the long-term 2011-2017 records of bottom water temperatures used in this paper. Data were recovered from data logger memory to a shipboard computer during ROV operations in 2012, 2014, and 2017. Offsets among the RTD readings almost certainly resulted from calibration errors, as calibrations before deployment were done only in an ice bath and at room-temperature, but resolution of relative temporal variations is 0.001°C in our measurement range. Data logger clock drifts were assessed at the times of the ROV data recovery operations and the clocks were re-synchronized with UTC in 2014 and 2017. Clock drifts averaged ∼1-2 min yr −1 , with very similar values at Sites U1382A and U1383C, so they are not significant to the findings reported here. At the end of 2017 ROV operations, the data logger memories were cleared and the sample intervals were changed to 20 min, such that sufficient battery power remained for at least a decade more of continued recording of signals such as the bottom water temperature variations reported here.
B. CTD data. The 2014 CTD data used in this study were obtained with a Sea-Bird SBE 9plus CTD profiler using factory calibrations. It included redundant sensors for temperature and conductivity; for each of those parameters, the redundant sensor readings were quite consistent. The lowering speeds and sampling frequency resulted in a data sample every ∼0.03 m, so some of the data shown in Fig. 3a were smoothed using a ∼0.5 m box filter. The CTD data were processed using the TEOS-10 Gibbs Seawater toolkit 43 to produce the profiles of Conservative Temperature and Absolute Salinity shown in Fig. 3a and the values of specific heat and density used in the Results section A. To produce the neutral density profile in Fig. 3a, the data were processed using the legacy EOS-80 neutral density function 44 .

Data availability
Long-term temperature and pressure data from the CORK installations at Holes U1382A, U1383B, and U1383C are freely available from the Marine Geoscience Data System (MGDS) 45-50 at www.marine-geo.org. North Pond CTD data from 2014 used in this paper are available from PANGAEA 51,52 at www.pangaea.de. Earlier CTD data from North Pond collected in 2009 and 2012 are also available from PANGAEA [53][54][55] . Further CTD data from North Pond collected in 2017 are available via the Biological and Chemical Oceanography Data Management Office at https://www.bco-dmo.org/datasetdeployment/757723.