The old (∼450-year isolation age) and near-homogenous deep waters of the Canada Basin (CBDW), that are found below ∼2700 m, warmed at a rate of ∼0.0004°C yr−1 between 1993 and 2010. This rate is slightly less than expected from the reported geothermal heat flux (Fg ∼ 50 mW m−2). A deep temperature minimum Tmin layer overlies CBDW within the basin and is also warming at approximately the same rate, suggesting that some geothermal heat escapes vertically through a multi-stepped, ∼300-m-thick deep transitional layer. Double diffusive convection and thermobaric instabilities are identified as possible mechanisms governing this vertical heat transfer. The CBDW found above the lower continental slope of the deep basin maintains higher temperatures than those in the basin interior, consistent with geothermal heat being distributed through a shallower water column, and suggests that heat from the basin interior does not diffuse laterally and escape at the edges.
 The deep waters of the Arctic Ocean form in the Nordic seas, and enter the Arctic Ocean through Fram Strait [cf. Rudels et al., 2012]. The Arctic Ocean itself contains two main basins; the Eurasian and Canadian, separated by the Lomonosov Ridge with a sill depth of ∼2000 m (Figure 1a). The Canadian Basin has the largest volume [Aagaard et al., 1985] and contains the oldest deep water, with 14C isolation age estimates of ∼450 yr, approximately 200 yr older than those in the Eurasian Basin [Schlosser et al., 1997; Macdonald et al., 1993]. In turn, the Canadian Basin is separated by the Alpha-Mendeleyev ridge complex (sill depth ∼2400 m) into the Makarov and Canada Basins [Swift et al., 1997; Timmermans and Garrett, 2006]. The deep waters of the Canada Basin (CBDW) are near-homogeneous, varying in potential temperature (θ) by less than 0.001°C between ∼2700 m and the bottom, a feature that Timmermans et al.  ascribe to geothermal heating and vertical convection. Langseth et al.  reported geothermal heat flux in the Canadian Basin of 40–60 mW/m2, and we use 50 mW/m2 here as a reference flux. Salinity in the Canadian Basin increases with depth to ∼2700 m, but, like θ, is nearly constant below this depth, at S = 34.957 psu.
Björk and Winsor  also observed near-homogenous bottom waters in the Eurasian Basin, noting a temperature change in the bottom layer between 1991, 1996 and 2001, which they attributed to geothermal heating. Using sparse data collected from 1990 to 2001 Timmermans et al.  described Canada Basin bottom waters and concluded that temperatures within the CBDW were near-constant with time, that little heat escaped vertically through the overlying deep transitional layer and that excess heat was lost mainly around the perimeter of the basin. Here, using data from annual surveys of the southern Canada Basin between 2002 and 2010, we re-examine role of geothermal heating in CBDW, describe temporal and spatial patterns in water mass properties, and propose heat exchange mechanisms that involve diffusive and thermobaric instabilities.
2. Study Area and Methods
 Physical data were collected in 1993 and 1997 and yearly from 2002 to 2010 during summer surveys conducted from the CCGS Louis S. St-Laurent for the Joint Ocean Ice Studies (JOIS) program [McLaughlin et al., 2011; Proshutinsky et al., 2009] (Figure 1b). Water samples were collected using a 24-bottle rosette with a Seabird CTD (SBE 911 plus). Temperature sensors were calibrated yearly and conductivity sensor data were calibrated with bottle data [McLaughlin et al., 2009]; salinity samples were analyzed on a Guildline Autosalinometer calibrated daily with IAPSO standard seawater. Nominal sensor accuracies are ±0.001°C for temperature; ±0.002 psu for salinity and 2 m for depth; instrument resolutions were about 0.0003°C for temperature; 0.0002 psu for salinity and 0.2 m for depth. Potential temperature (θ) and density relative to a given pressure are calculated from Fofonoff and Millard . The real accuracy, rather than nominal accuracy, of the temperature sensor is very likely better than ±0.001°C as our annual calibration of the temperature sensor reveals changes in calibration of <0.0003°C from one year to the next within the range of temperatures of the deep Canada Basin.
3. Observations and Discussion
3.1. Structure of the Deep Waters
 A schematic of the deep Canada Basin and the processes affecting the water there is shown in Figure 2a. CBDW appears as a thick, near-homogenous bottom layer extending from ∼2700 m to the bottom (Figure 2b). Above lies an ∼300-m-thick deep transitional layer (DTL) that is characterized by a temperature–salinity step structure, and the top of this layer is marked by a temperature minimum (Tmin) at ∼2400 m, the sill depth at Cooperation Gap on the Alpha-Mendeleyev Ridge complex. The staircase structure, through which both the mean temperature and mean salinity increase with depth, is observed at all stations across the entire basin and has been a persistent feature for at least two decades and is shown in Figure 2c for 2003 and 2010. The step structure is typically characterized by three to four mixed layers that are 10–60 m thick and are separated by 2–20 m-thick interfaces over which changes are δθ ∼ 0.003°C and δS ∼ 0.0007 [see Timmermans et al., 2003]. The θ/S properties of the deep basin (Figure 2d) show that stratification below the Tmin is marginally stable with respect to density calculated at 3000 m and that those of the Tmin itself closely match those of the Makarov Basin at sill depth, pointing to this as the likely source maintaining the Tmin layer [Timmermans et al., 2003]. For any given year the θ of CBDW within the deep basin is laterally near-uniform, varying by less than 0.0007°C across the full study area, reflecting the isolation of the basin and also the ubiquity of geothermal heating. The θ of the Tmin, however, shows greater spatial variability, as discussed below within contexts of circulation and renewal.
3.2. Temporal Trends and Spatial Variation
 The time sequence of potential temperature (θ) profiles in the deep Canada Basin from 2002 to 2010 reveals the steady increase in temperature of the CBDW (Figure 3). As shown below, this rate of warming is consistent with, but slightly less than, the geothermal heat flux. A least-squares fit to all available CBDW temperature data from 1993–2010 for stations deeper than 3000 m gives a rate of warming of ∼0.0004°C/yr; a similar rate of warming is observed in the Tmin, although there is a much larger spatial variation (Figure 4a). A least-squares fit to the associated salinity data from 1993 to 2010 shows no discernable trend for either the CDBW or the Tmin (Figure 4b).
 Stations above the slope in the south and east of the Canada Basin show CBDW temperatures warmer than those in the deep basin (Figure 5). In this plot temperatures have been normalized by removing the temporal trend and are shown as an anomaly referenced to temperatures below 3500 m. This analysis shows that stations with bottom depths of 3000 m are ∼0.0020°C warmer than the deep basin and stations with bottom depths of 2700 m are ∼0.0032°C warmer. This is because, for a uniform heat flux, a shallower bottom layer will warm faster. The key point here is that heat from the deep basin will not diffuse laterally into the shallower, warmer water over the slope.
3.3. Heat Budget Considerations
 The homogeneity of CBDW suggests a relatively short time-scale for vertical convection. Wüest and Carmack  give a rotating convective vertical velocity scale w* = 2.4(gαFg/ρcpf)0.5, where g is acceleration due to gravity, α is the thermal expansion coefficient, cp is the specific heat capacity, ρ is density, and f is the Coriolis parameter. Applying this formula yields a w* of 0.8 mm s−1; thus for the 1000 m-thick CBDW the time-scale for convection is calculated to be ∼15 days. This value suggests rapid mixing of the CBDW, particularly in relation to its isolation age.
 If geothermal heat were to remain trapped within the bottom layer, the layer would either become thicker or the temperature of the layer would increase. Examination of data from 2002 to 2010 (not shown) reveals no significant change in CBDW thickness. This is to be expected as the overlying DTL is primarily salt-stratified. If an input of geothermal heat Fg ∼ 50 mWm−2 [Langseth et al., 1990] remains in the bottom layer, the potential temperature of this layer θB with a thickness HB ∼ 1000 m (the volume-weighted mean thickness [see, e.g., Aagaard et al., 1985]), will increase according to dθB/dt = Fg/ρcpHB, where ρ = 1040 kg m−3 is the density and cp = 3900 J kg−1 °C−1 is the specific heat capacity. Applying this equation, and using the estimated geothermal heat flux of 50 mWm−2, suggests that θB will increase by ∼0.0004°C yr−1, or about 0.0032°C between 2003 and 2010. This rate of warming is consistent with the best fit of temperature data in CBDW shown in Figure 4, implying that the substantial fraction of geothermal heat remains in the deep layer. The potential temperatures at Tmin (θM) and within the DTL also increase by a similar amount, and thus the same potential temperature difference is maintained between θB and θM from year to year (Figure 4). Applying the above equation to the ∼300 m thick DTL we find that an additional heat flux of 10–15 mWm−2 is required to explain the observed warming. Thus, by these calculations, a total geothermal heat flux of 60–65 mWm−2 is needed, and this is within the range of 40–60 mWm−2 given by Langseth et al. . Hence, we can conclude that the reported geothermal heat flux can be accounted for by warming of the bottom and deep transitional layers.
3.4. Double Diffusive and Thermobaric Convection
 The temperature and salinity structure of the deep transitional layer (DTL), with cool fresher water above warmer and more saline water is conducive to double diffusive convection and thus to the formation of the observed staircase structure [Turner, 1968; Timmermans et al., 2003]. The stability of double-diffusive interfaces is defined in terms of the density ratio Rρ = βδS/αδθ; where α = −ρ−1δρ/δT, β = −ρ−1δρ/δS and ρ is density [Turner, 1973]. As noted by Timmermans et al.  the mean Rρ value for interfaces within the deep transitional layer is ∼1.6, a value conducive to strong double diffusive convection. Values of Rρ at the base of the DTL are smaller yet and thus conducive to even larger double diffusive fluxes.
 In the absence of direct, microstructure-based heat flux measurements, heat flux FH through a diffusive layer interface may be estimated using laboratory-derived flux laws [e.g., see Kelley, 1990; Polyakov et al., 2012] in which heat flux is given by the product of an empirically derived function of Rρ and the change of potential temperature across the diffusive interface δθ to the 4/3 power. Kelley  proposed the following relationship , where Pr = ν/κ is the Prandtl number, ν = 1.89 × 10−6 m2 s−1 is the kinematic viscosity, κ = 1.28 × 10−7 m2 s−1 is the thermal diffusivity, and g = 9.8 m s−2. From over 200 profiles collected between 2003–2010 in the Canada Basin the interface thickness through the DTL ranged from between 2 and 20 m, over which δθ averages ∼0.003°C and δS averages ∼0.0007 psu, while mixed-layer thicknesses are between 10 and 60 m. This yields average double diffusive heat fluxes through the multiple layers of the DTL of FH ∼ 45 mWm−2, which supports the argument that some of the vertical heat loss from the geothermally heated CBDW escapes by double diffusion. However, these values are even larger than required to match the observed warming rate of the DTL, being almost equivalent to the full geothermal heat flux estimate and more three times that which is observed in terms of DTL warming. They arise from the relatively low values of Rρ used above. However, as noted above and by Timmermans et al. , the observed values of interface thickness are much greater than typical values reported from microstructure profiles through active DL interfaces, raising some doubt that the diffusive instability is continuously active [cf. Padman and Dillon, 1987; Howard et al., 2004]. We thus assume that the double diffusion is intermittent or is influenced by additional processes such as shear instabilities or differential diffusion [Polyakov et al., 2012; Gargett and Ferron, 1996] .
 Nonetheless, the step structure is robust, as steps are found in nearly every profile, and so they are being maintained in spite of the tendency of diffusion to smooth them. If the processes creating the steps were not active, the steps would be eradicated by diffusion over a time scale τ = (H/2)2/2κ, where H is the step thickness. Using H = 40 m, and the molecular value for thermal diffusivity given above, yields a time scale of 50 years. However, the thermal diffusivity may be higher than molecular in the deep basin. For example 14C data lead to a diffusivity estimate of 4 × 10−5 m2s−1 [Macdonald et al., 1993], which gives a time scale of 2 months to eradicate the steps.
 Because the temperatures of CBDW and the DTL are increasing at approximately the same rate and salinities are remaining constant, the δθ/δS slope through the DTL has remained near-constant throughout the period of observation. This could be either coincidence or could signal a feedback mechanism that regulates vertical heat flux. Also, because the δθ/δS slope is very weakly stable for densities referenced to 3000 m, the potential exists for a convective process arising from the differential compressibility of water [Fofonoff, 1961; McDougall, 1987]. Since cold water is more compressible than warm water, a conditional instability (the thermobaric instability) occurs when a temperature interface where colder water overlying warmer water is displaced to a depth where the density of the underlying warm water is less dense than the cold water above [cf. Carmack and Weiss, 1991; Adkins et al., 2005]. Further, an interface can be preconditioned for the onset of thermobaric instability by either cooling from above or heating from below. In the present case, a two stage process may be required to trigger the thermobaric instability: first, steady warming of the CBDW end member relative to the base of the overlying DTL to decrease its density (referenced to local pressure) relative to that of the overlying water; and second, a descending motion of the interface to compress the overlying cold water more than the underlying warm water, thus making it denser and triggering thermobaric instability.
 Are geothermal heating rates and vertical displacements within the DTL sufficient to trigger the thermobaric instability? Figure 2d shows that a warming of ∼0.001°C of CBDW, requiring only 2 to 3 years of geothermal heating, would be required to trigger the thermobaric instability, assuming that the DTL remained unchanged. Vertical displacements within the DTL were measured by Timmermans et al.  using recording thermometers deployed within the deep transitional layer and they found typical vertical excursions at near-inertial frequency of 10–20 m, which they attributed to the very weak stratification. Timmermans et al.  noted larger excursions of up to 1040 m associated with topographic Rossby waves. Thus as temperatures within the bottom layer increase, and tilt the deep δθ/δS slopes closer to the critical gradient for instability, these vertical motions may be sufficient to trigger intermittent thermobaric overturning cells. The thermobaric overturning mechanism may also sharpen interface gradients, and may thus enhance conditions for double diffusive transfer.
 The quiescent deep water of the Canada Basin is warmed and mixed by geothermal heating; it is capped by a salt-stratified, multi-stepped deep transitional layer which is bounded above by a Tmin that marks the depth of exchange with the adjacent Makarov Basin (see Figure 2a). Using high precision and accuracy data collected with a SBE 911 we observe that (1) CBDW is warming at a measurable rate of ∼0.0004°C yr−1, (2) heat is not diffusing laterally and escaping along the basin perimeter since deepwater temperatures are higher along the edges, and (3) some fraction of this heat escapes through the multi-stepped deep transitional layer (DTL). The heat flux required to match the warming of the DTL is within the range of bulk estimates of double diffusive flux. We also note that because temperatures within the overlying multi-stepped DTL are increasing at approximately the same rate as the CBDW, the θ/S slope within the DTL is maintained at near-neutral stability for isolines of potential density calculated at pressure 3000 m. This in turn suggests a feedback mechanism whereby gradients within the DTL are maintained by intermittent triggering of thermobaric instabilities and subsequent adjustment. Note that temperature gradients within the Tmin are also constrained by lateral advection of cold water from the Makarov Basin, and thus there must be a balance between the inflow of cold water and upward flux of geothermal heat.
 While geothermal heating appears to explain the warmer temperatures of CBDW, a fact remains that the salinity of the CBDW is higher by ∼0.02 psu than its presumed upstream source region in the Eurasian Basin. This requires a local source of salt and the most likely source is via shelf drainage of saline plumes, either as an ongoing steady renewal of approximately 0.5% of CBDW volume per year or as major episodic renewal events over decades or centuries (see discussions by Aagaard et al. , Jones et al. , Aagaard and Carmack  and Rudels et al. ). However, since geothermal heating and boundary convection are two independent processes they can, together, control the evolution of CBDW: shelf drainage and slope convection supplies higher salinity water and perhaps additional heat, whereas geothermal heating warms and stirs the bottom layer. While our work has focused on the Canada Basin, the observations of Björk and Winsor  show geothermal heating to be a ubiquitous forcing element of the Arctic Ocean.
 Our appreciation is extended to Jane Eert and Rick Krishfield, to other scientific staff who carried out the JOIS program and the Beaufort Gyre Exploration Project, and to the officers and crew of the CCGS Louis S. St Laurent. This work was supported by Fisheries and Oceans Canada, by the Canada International Polar Year Office and by the National Science Foundation Office of Polar Programs grant OPP-0424864.
 The Editor thanks an anonymous reviewer for their assistance in evaluating this paper.