1. Introduction

Warm and salty Atlantic Water (AW) originating in the North Atlantic is transported at intermediate depths (~150–900 m) through the deep basins of the Arctic Ocean by topographically steered pan-Arctic boundary currents (e.g., Aagaard 1989; Rudels et al. 1994; McLaughlin et al. 2009). The role and relative importance of AW heat in shaping the Arctic Ocean's ice cover is still under debate [see discussion in Polyakov et al. (2012a)]. One significant source of uncertainty is the impact on diapycnal fluxes of the relatively cold halocline layer (CHL) that separates the fresh and cold surface mixed layer (SML) from AW (Fig. 1); see, for example, Aagaard et al. (1981), Pfirman et al. (1994), and Schauer et al. (1997, 2002). The stratification of the CHL—strong vertical gradients of salinity S and potential density σ_θ but a negligible gradient of potential temperature θ—impedes vertical mixing and the upward transport of AW heat (e.g., Rudels et al. 1996). For completeness, but not discussed here, we also note that in the Canadian Basin the lateral injection of relatively fresh Pacific-origin waters at intermediate (60–220 m) depths further strengthens stratification to inhibit heat exchange between the AW and the SML (McLaughlin et al. 2004; Steele et al. 2004).

View Full Size

Profiles of (a) θ , (b) S, and (c) θ–S derived from the EWG (1997) winter climatology for the two locations of the ITP-37 drift trajectory used for calculation of heat content and fluxes. Limits of the SML, CHL, and upper permanent pycnocline (UPP) are shown by horizontal green lines in (a) and (b), and by green circles in (c).

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

Limited observations suggest that upward heat fluxes from the AW over much of the interior of the Arctic Ocean's deep basins are weak (<1 W m⁻²; e.g., Padman and Dillon 1987; Rainville and Winsor 2008; Fer 2009; Timmermans et al. 2008a), but that heat fluxes over steep topography and basin margins can often be much larger (e.g., Padman and Dillon 1991; Lenn et al. 2009; Shaw et al. 2009; Sirevaag and Fer 2009; Polyakov et al. 2012b). We also expect seasonal variability in mixing, at least in the upper ocean, as surface forcing, sea ice state, and stratification all vary. Understanding the contribution of AW heat to the upper Arctic Ocean, including sea ice, requires identifying the causes of spatial and temporal variability in mixing from the AW layer to the surface.

Here, we use one year of hydrographic data from Ice-Tethered Profiler (ITP) drifters to quantify the release of heat from the upper pycnocline in the Eurasian Basin (EB) of the Arctic Ocean to the SML. These observations identify a direct, seasonally modulated flux of heat from the AW to the SML and thus to the sea ice in the eastern Arctic.

2. Observations

a. Hydrography

Two ITP buoys (www.whoi.edu/itp) provided twice-daily conductivity–temperature–depth (CTD) profiles in the upper ~800 m of the Eurasian Basin (Fig. 2). The ITP CTDs were equipped with SBE-41/41CP CTD sensors and had high vertical resolution (12–25 cm) and accuracy of θ (0.002°C) and S (0.002). Data processing and applied corrections are described in detail by Johnson et al. (2007). In our analysis, we used data interpolated to a 25-cm fixed vertical grid; that is, close to the original sampling interval. Buoy ITP-37 measured profiles for one year (September 2009–September 2010). However, the data quality deteriorates from mid-July 2010 as demonstrated by substantial profile-to-profile differences in θ and S in the deep part of the ITP-37 profiles (not shown), and so was not used in our analyses. Buoy ITP-36 was active for one month (September 2009) before it stopped transmitting data.

View Full Size

Locations of oceanographic profiles provided by the ITP #36 and #37 (circles, black digits show profile number). Colors used for the circles show time (see legend at bottom left). The Gakkel Ridge (GR) divides the Eurasian Basin (not marked) into the Nansen Basin (NB) and the Amundsen Basin (AB). The Lomonosov Ridge (LR), Novosibirskiye Islands (NO), Severnaya Zemlya (SZ), Franz Joseph Land (FJL), Spitsbergen (SP), and Greenland (GD) are indicated. Background shows water depth in m; color scale upper right.

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

Every second ITP-36 profile and every fourth ITP-37 profile were used in this analysis. This thinning out of ITP records was done because of the highly labor-intensive procedures involved in the raw data processing; it also eliminated problems associated with differences between the up and down profiles. Despite thinning, the reduced datasets still resolve all variability relevant to the present study; the selected ITP-37 data consists of more than a hundred θ(z) and S(z) profiles with mean horizontal separation of ~6.4 km between profiles in winter months; this sampling is sufficient to resolve seasonal changes discussed in this analysis.

3. Variability of upper Eurasian Basin hydrography derived from ITP records

A section of θ(z) and S(z) through the Eurasian Basin, obtained by concatenating the ITP-36 and ITP-37 datasets (Figs. 2 and 3), incorporates both spatial and temporal changes. However, for reasons explained below in section 4, we propose that most variability is temporal, and so we discuss the section's basic features in terms of time of data acquisition.

View Full Size

(top) θ (°C) and (middle) S (psu) from (left) ITP-36 and (right) ITP-37; (bottom) bathymetry along the ITP drifts from the General Bathymetric Chart of the Oceans (GEBCO) 1′ Digital Atlas. The x axis shows profile number complemented by approximate time in top and middle panels. White segments indicate missing data. The ranges of parameters used for color maps are shown in white inserted windows within ITP-37 panels; the first (last) color scale is used for values less (more) than the identified range (e.g., the deepest blue color for θ identifies values ≤−1.85°C). Black solid line in the top and middle panels on right shows the depth of the upper mixed layer; black broken line in bottom and right panel shows the depth of the CHL base. The white box in the top and right panel indicates the record segment used for analysis of upward heat fluxes. The same panel shows vertical dotted white lines indicating periods when the buoy trajectory made a loop; this part of the ITP-37 record was used to evaluate impact of spatial variability on estimates of heat content and vertical fluxes (see Fig. 7 and text for details).

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

At the beginning of the record in summer (August–September) 2009, the upper ocean above ~25 m was relatively warm (Fig. 3). In ITP-36 data from September, there is a well-defined near-surface temperature maximum (NSTM) near ~25–30 m. The NSTM is associated with summer solar radiation that heats the upper ocean over a depth range set by stratification and the water's optical properties (e.g., Jackson et al. 2010; Steele et al. 2011). Part of this radiative heat is then taken up at the surface by the melting of ice, lowering the SML temperature, salinity and density and leaving a temperature maximum at an intermediate depth. In recent years, the NSTM has survived winters in the Canada Basin (Jackson et al. 2011; Steele et al. 2011), suggesting negligible thermodynamic coupling between the layer below the NSTM and the SML. This was not always the case in the Canada Basin; Maykut and McPhee (1995) demonstrated the disappearance of the NSTM in data from the 1970s. We note that the NSTM is visible in the ITP-37 data from the EB at ~20-m depth in early September, but it is absent throughout the winter portion of the record (Fig. 3).

We define the depth of the SML h_SML as a change in σ_θ from the ocean surface of 0.125 kg m⁻³, following Monterey and Levitus (1997). The SML depth gradually increases through the ITP-37 record from ~20 m in September 2009 to ~60 m in March–April 2010 (Fig. 3). Below the warm near-surface layer in summer, and NSTM in September and early October, there is a cold layer of winter water (WW) that is the remnant of the previous winter's SML; see, for example, Jackson et al. (2011) and Steele et al. (2011). The WW is evident in the summer 2009 ITP-36 record as cold and increasingly salty water in the depth range ~30–50 m (Fig. 3). Within the WW were two anticyclonic eddies (from profiles 30–35 in ITP36 and profiles 540–550 in ITP37) that, although saltier, were at a similar temperature and depth as shallow cold core eddies observed north of 75°N in the Canada Basin (Timmermans et al. 2008b). Excluding these eddies, we assume that most variability in Fig. 3 is due to temporal changes, with renewal of the WW occurring by convection beginning with SML cooling in September–October, followed by salinization by brine rejection from growing sea ice starting in early December. In summer 2010 (June–July), radiative heating warms the surface water again, leaving a new subsurface WW layer.

The WW lies in the upper part of the CHL (see Fig. 3) where S increases with depth but θ remains almost constant, close to the freezing point. The lower CHL boundary is defined following Bourgain and Gascard (2011), who used an extensive collection of modern data and argued that the density ratio R_ρ = (α∂θ/∂z)/(β∂S/∂z) = 0.05 (α is the thermal expansion coefficient and β is the haline contraction coefficient) may be used to establish the depth of the CHL base. Using this definition, we found that the lower CHL boundary was at 77.8 m in early January 2010 and at 81.2 m in late April 2010 (Fig. 4). Thus, 80-m depth is a good proxy for the depth of the CHL base for the winter part of the ITP-37 record.

View Full Size

ITP-37 profiles of (a) θ (°C) and (b) S (psu) based on 5-day averages in early January and late April 2010. (c) θ–S plot for the two ITP-37 θ and S profiles. Inferred CHL and UPP layer–integrated heat flux divergences are for the temperature profiles (see text for details). The SML, CHL, and UPP limits are shown by horizontal green lines in (a) and (b), and by green circles in (c); dashed lines in (a),(b) show averaged depth of the SML.

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

Below the CHL, θ and S both increase downward through the permanent pycnocline above the core of the AW layer. The only source of heat for the UPP in this part of the Arctic Ocean is upward flux from the AW layer. In summer and early winter, the UPP (heated by the AW) is separated from the SML by the WW. During this time, the layer below the WW accumulates heat as expressed by the trend toward shallower isotherms for warm temperatures typical of the UPP (Fig. 3). Starting in late December–January, the UPP becomes colder. This process stops in May–June.

We quantify the seasonal changes in the CHL and UPP using vertically integrated heat content Q (J m⁻²), defined as

View Expanded

where θ_freezing is the freezing temperature, ρ_w is water density, c_p is specific heat of seawater, and z₁ and z₂ are depths of the upper and lower boundaries. Figure 5 shows monthly-mean Q and the corresponding heat flux divergence required to explain warming/cooling in the 65–100-m layer. The upper ocean warms and restratifies late in the record, coincident with the onset of summer but also with ITP-37 crossing the Gakkel Ridge (see Figs. 2 and 3). We discuss how changes to monthly estimated heat flux relate to diapycnal heat flux divergence in section 5.

View Full Size

(bottom) Monthly-mean heat content in the 65–100-m water layer computed using θ from ITP-37. (top) Heat flux divergence evaluated from the differences of consecutive monthly heat contents; red bars (positive heat fluxes) indicated periods when the layer was warming and blue bars (negative heat fluxes) indicate periods when the layer was losing heat. Solid colors are used for estimates of unadjusted heat fluxes Δθ₀, semitransparent colors are used for estimates of heat fluxes based on adjusted heat fluxes Δθ_AW (see text for details). Standard errors are shown in both panels by vertical black segments.

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

4. Separating spatial and temporal variability

Datasets collected by ice-mounted drifting buoys necessarily blend spatial and temporal variability (see Padman and Dillon 1991; McLaughlin et al. 2004). Here, we present a scaling argument to suggest that most of the observed upper-ocean variability is temporal, and thus can be interpreted as being caused by vertical advection and diapycnal turbulent fluxes. While some specific short-duration features (e.g., the anomalously warm SML in late April; Fig. 3) are probably associated with mesoscale variability such as eddies, most observed variability over time scales of one month or longer can be explained through diapycnal processes as explained below.

Consider a simple model that describes changes in the heat budget within a unit-area water column of the upper ocean within a depth range of thickness H = z₂ − z₁, where z₁ and z₂ are upper and lower depth limits. The vertically integrated heat budget for this box is a combination of lateral and vertical advection and divergence of turbulent heat fluxes:

View Expanded

In Eq. (2), the overbar denotes vertical mean within z₁ < H < z₂, is the change in θ over a period of time δt, U is along-trajectory horizontal velocity, and W is vertical velocity. The last two terms in Eq. (2) describe divergence of lateral and vertical heat fluxes. The coefficient K_H is the lateral diffusivity, and ΔF_h is the difference of diapycnal heat fluxes F_h at z₂ and z₁. Typically, F_h is calculated as with K_z being the diapycnal diffusivity; however, here we estimate ΔF_h from changes in vertically integrated heat content, not via measured ∂θ/∂z and estimated K_z.

We wish to explain the cooling of the upper pycnocline between January and April 2010, a period of δt = 115 days during which ITP-37 drifted a distance of L ~ 370 km at a mean speed of U ≈ 4 cm s⁻¹. This entire segment of the buoy trajectory was in the Amundsen Basin, moving roughly westward parallel to the Gakkel Ridge (Fig. 2) over relatively flat and deep bathymetry (Fig. 3).

We apply Eq. (2) to the depth range 65–100 m, chosen because it best represents the layer in which heat from the AW is stored and released. For this depth range, the average change δθ over δt = 115 days was 0.23°C. This value is much larger than the spatial temperature difference δθ_clim of ~0.03°C derived from the EWG (1997) winter climatology for the same depth range and the same portion of the ITP-37 trajectory (Fig. 1). However, estimates utilizing EWG climatology should be viewed with caution because of the very different state of the ocean in recent decades relative to the 1970–80s (see discussion for details). Note also that δθ_clim for the depth range 65–100 m is of opposite sign to the observed change in the lower permanent pycnocline in the ITP-37 dataset. The value δθ is also much larger than the observed temperature change for deeper layers where the seasonal signal is expected to be weak (e.g., Lique and Steele 2013). For the depth range of 150–250 m (the lower permanent pycnocline consisting of AW), the difference in θ between the beginning of January and end of April was δθ_AW = −0.062°C (Fig. 4).

We estimate the potential contribution to the heat budgets of vertical advection associated with Ekman pumping from the curl of the wind stress (∇ × τ), which drives a vertical displacement of

View Expanded

where ρ_w is water density and f is the Coriolis parameter. Positive (negative) values of w_e represent cyclonic (anticyclonic) wind stress exerted on the surface causing a divergence (confluence) of surface waters resulting in upwelling (downwelling). The stress τ in Eq. (3) takes into account the presence of ice via a drag coefficient that is a second-order polynomial function of ice concentration (Andreas et al. 2010). This relationship suggests that air–water momentum exchange is most efficient when sea ice concentration approaches ~50%. Typical values of w_e based on ERA-Interim winds (section 2c) are 0.01 m day⁻¹ (Fig. 6a). By integrating w_e along the drift track, we estimate maximum displacements due to ∇ × τ over the ~4-month period of <10 m and typically <3 m (Fig. 6a). Early winter is characterized by weak downwelling (w_e < 0), transitioning to upwelling later in winter. The magnitudes of these displacements are small compared with observed isopycnal variability over winter (Fig. 3). The associated potential temperature change δθ caused by Ekman pumping over the winter season (January–April; see white box in Fig. 3) is 0.069°C. The same estimate made using w_e calculated without taking into account ice concentrations produces ~10% higher δθ. We expect that instantaneous values of ∇ × τ may be significantly larger than those predicted from the coarse-grid ERA-Interim product; however, averages over weeks and months, which drive the net displacements being estimated here, are probably reasonable. Thus, we argue that vertical advection caused by ∇ × τ does not explain observed θ changes. We note, however, that even in ERA-Interim, atmospheric events of short duration such as cyclones can cause substantial distortion of isopycnal surfaces. For example, anomalous w_e in early April (Fig. 6) resulted in an upwelling of the depth of the SML (black line in Fig. 3) by O(10) m.

View Full Size

(a) Daily w_e (m s⁻¹; red) and associated integrated vertical displacement δz_e (m; blue) along the trajectory of drifter ITP-37, based on ERA-Interim. (b) Total atmospheric heat fluxes derived from ERA-Interim for two sites near the locations used for calculating winter heat fluxes; near ITP-37 profiles #230 and #505 (see Fig. 1). (c) AMSR-E satellite-based sea ice concentration at the same two locations as in (b).

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

The heat budget for the upper ocean depends, in part, on surface buoyancy fluxes that may be spatially variable because of different atmospheric or sea ice conditions. It is thus plausible that some of the upper-ocean variability along the ITP-37 drift track represents different surface buoyancy forcing histories at different locations. However, comparisons of total atmospheric fluxes (from ERA-Interim; see section 2c) and sea ice concentration (see section 2b) reveal very little change between the start and end locations of the analyzed section of ITP-37 (Figs. 6b,c). These observations suggest similar freezing rates and, therefore, brine rejection intensity, so that the surface forcing along the entire track may be approximated by spatially invariant time series. The difference of mean atmospheric heat fluxes from ERA-Interim for the two locations was ~0.5 W m⁻² (Fig. 6), which is equivalent to ice thickness change of only ~2 cm over the 115-day interval. For the ~25-m-thick upper layer with mean salinity of 33.24 psu measured at ITP-37 profile #230, and taking a lower-limit ice salinity of 3 psu, this ice thickness change would result in an upper-ocean salinity change of the order of 0.01–0.02 psu: this is a small fraction of measured seasonal salinity fluctuations in the upper mixed layer and pycnocline (Fig. 4).

Because lateral and vertical advection of gradients is usually insignificant and, in addition, modeling results suggest that advective exchanges between the Nansen and Amundsen basins are weak (Ye. Aksenov 2013, personal communication), we hypothesize that observed changes in heat content of the layer 65–100 m must be associated with turbulent mixing. An upper bound for the contribution from lateral mixing is given by . For = 0.03°C (370 km)⁻¹, K_H would have to be on the order of 10⁴ m² s⁻¹ for lateral mixing to explain the observed δθ. Because this is several orders of magnitude higher than values of K_H determined for the open ocean (see, e.g., Ledwell et al. 1993), we conclude that lateral mixing cannot explain our observations. We also note that the δ_yθ diffusive component (corresponding to a normal-to-drift north–south direction) should not lead to the observed wintertime cooling of the upper ocean; to the contrary, the general westward buoy drift trajectory should lead to warming of the water column as a result of warmer water masses in the western Nansen Basin compared to its eastern part. In the Amundsen Basin, δ_yθ is too small to contribute substantially to temperature change [e.g., δ_yθ_65–100m = 0.01°C between the North Pole Environmental Observatory 2010 CTD station #5 (87°57.380′N, 89°15.410′E) and ITP-37 station #476 separated by 172 km, both made in late April 2010]. Thus, the only term that remains to balance the observed change in upper-ocean θ is vertical divergence of diapycnal turbulent heat flux ΔF_h.

Given the above results, in the following section, we explore the heat budget for the upper ocean interpreted as being dominated by vertical turbulent processes.

5. Inferred diapycnal heat flux divergence in the upper ocean, winter 2009/10

We use the ITP-37 temperature record to estimate winter heat loss from the CHL and UPP. Estimates of heat content change ΔQ were made for the time interval when the CHL and UPP were both cooling, using time-averaged profiles (Fig. 4) for periods 1–5 January and 25–29 April 2010 to reduce the potential impacts of the short-term variability seen in Fig. 3. The change in diapycnal flux divergence required to explain the change in layer-averaged temperature from January to April is sensitive to the assumption that the lateral gradient ∂θ/∂x is small. We consider the effect of uncertainty in these gradients by applying corrections Δθ, uniform with depth, to the April profile of θ(z) based on (i) Δθ_EWG = −0.03°C from EWG (1997) winter climatology, averaged over the depth range 65–100 m; (ii) no correction (Δθ₀ = 0), and (iii) Δθ_AW = +0.062°C based on observed changes in the AW in the lower permanent pycnocline (150–250 m; Fig. 4). For these calculations, the CHL was defined to lie between the variable depth SML (see Fig. 3) and 80 m, overlying the UPP for which the lower boundary was specified as 135 m (Fig. 4).

Using Δθ_AW, the estimated values of ΔQ in the CHL and UPP were −3.9 and −30.5 MJ m⁻², respectively. In our one-dimensional model, these values are equivalent to a heat flux difference ΔF_h (~ΔQ/δt) for each layer of 0.4 and 2.7 W m⁻², respectively. As expected from Fig. 4a, ΔF_h across the CHL is much weaker than across the UPP. Note that these values are flux differences, and total fluxes may be larger than these values because of additional nondivergent heat transports that we cannot observe with this dataset.

The net divergent heat flux through the CHL and UPP based on Δθ_AW is ~3.1 W m⁻² (plus the unknown nondivergent term); a strong flux for the Arctic Ocean interior (see section 1). This value increases to 4.7 W m⁻² if no θ adjustment is used and to 5.4 W m⁻² using the Δθ_EWG offset (Table 1). We also tested the sensitivity of estimates of ΔF_h to the definition of the UPP and SML depths (Table 1). For example, an increase of the lower UPP boundary by 5 m changes ΔF_h (UPP) by ~3% whereas the use of constant h_SML=50 m increases ΔF_h (CHL) from 0.4 to 0.7 W m⁻².

Table 1.

Estimated upward divergent AW heat fluxes F_h (W m⁻²).

View Table

For a value of ∂θ/∂z=0.03°C m⁻¹ below the SML (Fig. 4), the inferred increase in K_z to explain 3 W m⁻² of seasonal additional heat flux from the UPP and CHL is ~2 × 10⁻⁵ m² s⁻¹, comparable to canonical deep-ocean values. While it is higher than most of the few direct measurements of K_z from the central Arctic Ocean, there are no such measurements from the EB during winter and so we have no a priori reason to discount this estimate.

We estimated the month-to-month change of this heat flux divergence, using Q derived from monthly-mean profiles of θ(z) (Fig. 5). Using Δθ_AW, the divergent heat flux across the combined UPP and CHL layers peaks at 5 W m⁻² in February–March with 1–2 W m⁻² in December–January and by the end of the winter season. Our winter estimates are robust: winter Q and heat fluxes using unadjusted (Δθ₀) and adjusted profiles of monthly θ(z) are within the accuracy of our statistical estimates of the monthly means (Fig. 5). However, the decline in heat content from October to November (Fig. 5) is inconsistent with the seasonal cycle of heat variability driven by divergent diapycnal fluxes. ITP-37 crossed the Gakkel Ridge (Figs. 2 and 3) in September–October, moving from the warmer AW of the Nansen Basin to cooler AW of the Amundsen Basin and the estimates for this period are, therefore, likely to be affected by spatial variability.

Specific features of the ITP-37 drift trajectory provided further insight into the relative roles of spatial and temporal variability in the observed winter cooling of the upper Amundsen Basin. Figures 7a and 7b show that, in late March and April, the buoy made a loop with three closely spaced segments. Thus, evaluation of Q for the ITP-37 profiles along these three segments provides estimates of the rate of ocean cooling with minimal contamination by large-scale buoy displacement. Estimates of Q and derived ΔF_h shown in Figs. 7c and 7d suggest that most cooling occurred in late March into early April when ΔF_h was estimated as ~9 W m⁻². By the end of April this rate dropped to ~1 W m⁻². These values are consistent with the divergent fluxes based on monthly Q (Fig. 5). By the end of April the upper ocean below the SML was cooler by ~0.1°C compared with late March (Fig. 8d).

View Full Size

(a) Bathymetry (m; solid black lines) and locations of ITP-37 profiles (blue dots, every other profile is shown). (b) Region shown by red box in (a) where blue digits indicate the numbers of ITP-37 profiles. (c) Heat content in the 65–100-m water layer for three selected segments of the ITP-37 record when the buoy drifted westward (profiles 406–425), then eastward (profiles 425–449), and again westward (profiles 449–486). Standard errors are shown by vertical black segments. (d) Heat flux divergence evaluated from the differences of consecutive heat contents; these three periods were associated with the loss of heat by the layer (negative heat fluxes).

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

(a) The squared Brunt–Väisälä frequency N² for ITP-37 406–486 profiles (20 Mar–30 Apr 2010). The white vertical dotted lines show periods when the buoy trajectory made a loop (the buoy drifted reverse to general westward direction during the central segment of the record; see Fig. 7). (b) ITP-37 drift speed. (c) θ–N² and (d) θ–S plots for the same ITP-37 profiles. Colors in (c),(d) follow rainbow notation with blue–green used for the earlier profiles and yellow–red used for the later profiles. Every other (i.e., made once per day) ITP-37 profile is shown in (a),(c),(d).

Citation: Journal of Physical Oceanography 43, 10; 10.1175/JPO-D-12-0169.1

• Download Figure
• Download figure as PowerPoint slide

Cooling in late March–April was also associated with erosion of the SML as suggested by weakening of stratification (Figs. 8a,c) and elevation of the SML base (Fig. 3). This is contrary to the deepening of the SML and sharpening of vertical gradients typically associated with convection driven by surface buoyancy fluxes (cooling and ice formation). Thus, some other mechanisms must be involved in order to establish the observed pattern of spatiotemporal variability. We will discuss potential mechanisms to explain the inferred elevated mixing below the SML in winter in the next section.

Regardless of the mechanisms for enhanced winter subsurface mixing, if we assume that mixing below the base of the UPP is constant throughout the year and that the associated heat input is balanced by heat lost from the UPP and CHL from January to April (115 days), then the estimated annual-averaged upward flux of heat from the AW layer is about 3.1(115/365) = 1.0 W m⁻².

6. Discussion and conclusions

Our analyses corroborate the existence of previously hypothesized thermodynamic coupling between the AW heat and the sea ice in the Eurasian Basin of the Arctic Ocean (e.g., Polyakov et al. 2012a). We estimate that the annual-average vertical heat flux from the AW to the base of the UPP is ~1.0 W m⁻², which is larger than most previous estimates of AW heat loss in areas away from steep topography (e.g., Padman 1995; Lenn et al. 2009; Polyakov et al. 2012b). This heat is stored in the UPP and CHL until there is enough vertical mixing through either winter ice formation and associated brine rejection or storms to entrain the heat into the winter surface mixed layer. The rate of AW heat transport is equivalent to ~10 cm yr⁻¹ of ice loss and is close to the observed imbalance of net heat flux to the sea ice required to explained observed thinning of the Arctic ice pack during the last few decades (Kwok and Untersteiner 2011; Laxon et al. 2013). Because this area of the Arctic Ocean is characterized by very compact winter ice cover (see Fig. 6c), much of this seasonal release of AW heat to the SML cannot vent to the atmosphere directly through leads but must contribute to thermodynamic exchanges at the sea ice base. This heat flux does not stop ice formation, otherwise there would be no convective mixing and ongoing heat release to the SML; instead, this heat flux represents a negative feedback mechanism that reduces the rate of winter ice growth, similar to the thermal barrier described by Martinson (1990) for the Southern Ocean.

We do not yet understand the actual mechanisms that cause winter cooling of the UPP. The stratification below the CHL is, in general, statically very stable, with a buoyancy frequency in the UPP of ~4 cph; therefore, we do not expect enhanced heat loss there to be due to direct penetration of surface-formed convective cells. Instead, we hypothesize that enhanced mixing below the base of the SML and intensive cooling of the UPP may be associated with shear instabilities from downward-propagating near-inertial internal gravity waves caused by storms, internal tides propagating into the EB from generation sites around its perimeter, and double diffusion. While we lack the necessary data to test whether any of these processes undergo the seasonal modulation required to explain the observed evolution of upper-ocean hydrography, we review them in the following subsections as a guide to future work.