## 1. Introduction

A key role of the ocean in the climate system is to store and redistribute heat, freshwater, and biogeochemical tracers such as carbon dioxide. The manner in which this occurs depends strongly on the strength of mixing processes. Although ocean models are sensitive to the strength and the spatial distribution of the parameterized mixing, our knowledge and understanding of mixing processes from observations is limited.

In the ocean interior, mesoscale eddies stir properties along neutral tangent planes (Iselin 1939). Here, a neutral tangent plane is the direction along which a water parcel can move a small distance without experiencing a vertical buoyancy force (McDougall 1987a). As a result, ocean mixing is parameterized as 1) isopycnal downgradient tracer diffusion due to stirring by mesoscale eddies by means of an isopycnal eddy diffusion coefficient *K*_{I} and 2) small-scale isotropic downgradient turbulent diffusion by means of a turbulent diffusion coefficient *D* (Redi 1982; Griffies 2004; McDougall et al. 2014). Here, mesoscale refers to length scales of the order of 50–300 km, while small scale refers to processes such as breaking internal waves and shear-induced turbulence and is of the order of millimeters to tens of meters. Mesoscale mixing in the mixed layer is alternatively defined as surface diabatic mixing (Tandon and Garrett 1997) and is parameterized as horizontal downgradient tracer diffusion due to stirring by mesoscale eddies by means of a horizontal eddy diffusion coefficient *K*_{H}. The isotropic nature of *D* is discussed in McDougall et al. (2014) but has previously been regarded to be dianeutral (Redi 1982; Griffies 2004) or vertical in the small-slope approximation.

### a. Small-scale mixing

Small-scale mixing leads to a downward flux of heat that is required to warm and subsequently upwell abyssal dense water (Ganachaud and Wunsch 2000; Sloyan and Rintoul 2000) and plays a crucial role in closing the ocean circulation (Munk 1966; Munk and Wunsch 1998; Wunsch and Ferrari 2004). Sources of small-scale mixing are (for example) wind-generated mixing by breaking of near-inertial waves near the surface and in the ocean interior (D’Asaro 1985; Alford 2001; Alford et al. 2012), dissipation of the internal tides (Nycander 2005), and generation and dissipation of lee waves (Nikurashin and Ferrari 2011). Estimates of small-scale mixing have been obtained from measurements using microstructure instruments (Polzin et al. 1997; Garabato et al. 2004), tracer release experiments (Ledwell et al. 1993), inferred from lowered ADCP and CTD profiles (Kunze et al. 2006), or calculated using moored profilers (Alford et al. 2006) or floats (Whalen et al. 2012; Meyer et al. 2015).

Microstructure observations have indicated that, in the ocean interior, *D* increases toward the ocean floor with maximum values exceeding *D* = 1 × 10^{−3} m^{2} s^{−1} in regions of rough topography. The combination of all small-scale mixing processes results in a highly spatially variable structure for *D* (Waterhouse et al. 2014), with values ranging between *O*(10^{−7}) and *O*(10^{−3}) m^{2} s^{−1}. Strong sensitivity of ocean and climate models to differences in the magnitude of the parameterized vertical mixing argues that more studies are required to further constrain these estimates (Simmons 2004; Melet et al. 2013).

### b. Mesoscale mixing

Eddy transports dominate dispersion of particles and mixing of tracers on large spatial and temporal time scales. In numerical models the eddy diffusion coefficients *K*_{H} and *K*_{I} parameterize tracer diffusion due to eddies that are not resolved. As a rule of thumb, the higher the resolution of a model, the more eddy transport is resolved. Therefore, higher-resolution models tend to need lower diffusion coefficient (Okubo 1971). In this study, we are considering the eddy diffusivity to parameterize tracer diffusion (Redi 1982), which is not the same as the quasi-Stokes streamfunction parameterization for eddy-induced advection [or Gent–McWilliams (GM) parameterization; Gent and McWilliams 1990; Gent et al. 1995; McDougall and McIntosh 2001].

Estimates of mesoscale diffusivities have been based on tracer release experiments (Ledwell et al. 1993, 1998), float dispersion (Zhurbas and Oh 2004; LaCasce et al. 2014), satellite data (Holloway 1986; Abernathey and Marshall 2013; Klocker and Abernathey 2014), or, most recently, a combination of Argo and the ECCO2 state estimate (Cole et al. 2015). The observations clearly illustrate a highly spatially variable pattern with enhanced mixing in western boundary currents and a general decay of the mesoscale diffusivities with depth, with the exception of some subsurface maximums and values ranging between 0 and *O*(10^{4}) m^{2} s^{−1}.

Allowing for spatial variation of eddies in climate models can reduce systematic drift (Ferreira et al. 2005; Danabasoglu and Marshall 2007). Differences in the *K*_{I} value used in numerical climate models can lead to differences in the simulated global surface temperature change by the end of the twenty-first century of up to 1°C (Pradal and Gnanadesikan 2014), influence the stability of the overturning circulation (Sijp et al. 2006), and affect processes due to the nonlinear equation of state, such as cabbeling and thermobaricity (McDougall 1987b; Iudicone et al. 2008; Groeskamp et al. 2016). It is therefore essential that coarse-resolution climate models use the most accurate parameterization of the spatial varying effect of eddies.

### c. Estimating mixing from an ocean climatology

In this study, we will provide a global, observation-based quantification of circulation and mixing. This is obtained by applying an inverse method to an ocean climatology, which is based on spatial and temporal averaging of observations of the ocean’s hydrography. This is possible because the spatial and temporal structure of the ocean’s hydrography is set by the processes acting upon it (e.g., tides, winds, surface freshwater and heat fluxes, and mixing). As a result the ocean’s hydrography holds information about these processes and the circulation resulting from it. The inverse model is used to extract and estimate the ocean dynamics that are embedded in a climatology.

Historically box inverse models have been used to estimate the absolute velocity from ocean hydrography (Wunsch 1978). With increasing spatial and temporal resolution of available ocean climatologies, it is now possible to obtain estimates of more complicated (higher order) processes, such as ocean mixing. However, this requires inverse methods that are specifically designed for that purpose. Some box inverse methods have estimated *D* but often contain unknowns at the boundaries that require both dynamical constraints and conservation statements, increasing the complexity of the system and sensitivity to prior estimates, leading to uncertainties in the obtained transport rates and especially *D* (Zika et al. 2010a).

In this paper, we will first describe the inverse model (section 2) and apply it, for the first time, to a global climatology (sections 3–4). As will be discussed (sections 5–7), the resulting mixing estimates have significant implications for our understanding of ocean turbulence.

## 2. The thermohaline inverse method

This section provides a conceptual explanation of the diathermohaline streamfunction (Groeskamp et al. 2014a) and subsequently the derivation of the thermohaline inverse method (THIM; Groeskamp et al. 2014b). For detailed mathematical derivations and assessment of both the diathermohaline streamfunction and the THIM, we refer the reader to Groeskamp et al. (2014a,b) and Hieronymus et al. (2014). The THIM estimates mixing and circulation in Absolute Salinity *S*_{A} and Conservative Temperature Θ coordinates. Here, Conservative Temperature is proportional to potential enthalpy (by the constant heat capacity factor ^{−1} K^{−1}), representing the heat content per unit mass of seawater (McDougall 2003; Graham and McDougall 2013). Absolute Salinity is measured on the Reference Composition Salinity Scale (Millero et al. 2008) and is an approximation to the mass fraction of dissolved material in seawater (g kg^{−1}; McDougall et al. 2012) and is the salinity variable of the IOC et al. (2010) thermodynamic description of seawater.

### a. The diathermohaline streamfunction

The Walin framework (Walin 1982) originally considered water mass transformation from one temperature to another. The Walin framework has also been applied to calculate water mass transformation as the change in a fluid parcel’s density (Marshall 1997; Marshall et al. 1999; Nurser et al. 1999; Marsh 2000; Iudicone et al. 2008; Badin and Williams 2010; Badin et al. 2013). Here, we define a water mass transformation as a change in the (*S*_{A}, Θ) values of a water parcel (Speer and Tziperman 1992; Speer 1993, 1997). Water mass transformation is then directly related to heat and salt fluxes by mixing and air–sea interactions (i.e., thermohaline forcing), as these are the only mechanisms that can change a water parcel’s *S*_{A} and Θ.

Groeskamp et al. (2014a) showed that nondivergent transformations (those that do not cause convergences or divergences of the volumes of water masses) can be represented by a streamfunction in (*S*_{A}, Θ) coordinates known as the diathermohaline streamfunction. We will use the symbol Ψ for this streamfunction. In Groeskamp et al. (2014a) and Zika et al. (2012), a subscript *S*_{A}Θ was used to distinguish it from streamfunctions in other coordinates. In Groeskamp et al. (2014a), a superscript “dia” was also used to distinguish the diathermohaline streamfunction that describes the total transformation across *S*_{A} and Θ surfaces from those that describe the flow normal to *S*_{A} and Θ surfaces (advective thermohaline streamfunction Ψ^{adv}; Döös et al. 2012; Zika et al. 2012) and the movement of those surfaces separately (local thermohaline streamfunction Ψ^{loc}). Using numerical models, Hieronymus et al. (2014) and Groeskamp et al. (2014a) provided a thorough analyses of the circulation cells of the diathermohaline streamfunction Ψ.

### b. The thermohaline inverse method

Groeskamp et al. (2014b) developed the THIM to be applied to an ocean climatology and allow for observation-based estimates of diffusion coefficients and Ψ based on estimates of air–sea fluxes and a priori estimates of the spatial distribution of mixing. The THIM exploits the fact that circulation in (*S*_{A}, Θ) coordinates requires a freshwater and/or heat flux convergence that can only be provided by boundary fluxes and/or ocean mixing.

Above we have used a conditional volume integral over the whole ocean, where *S*_{A} is between *S*_{A}, Θ) coordinates. Trends in water mass volumes can be accounted for in this method (Groeskamp et al. 2014b) but will not be considered here as we use an annual climatology of ocean temperature and salinity and surface fluxes (see section 3).

*S*

_{A}) between two isotherms (lines of constant Θ) can be represented by the change in Ψ with respect to Θ at constant

*S*

_{A}such that (Fig. 1)

^{2}s

^{−1}), and

*f*

_{s}=

*S*

_{A}(

*E*−

*P*−

*R*), if the equivalent salt flux due to evaporation

*E*, precipitation

*P*, and runoff

*R*, such that ∇

*f*

_{s}is the net salt flux divergence (g kg

^{−1}s

^{−1}) into the volume and

*L*

_{S}is a correction term (m

^{3}s

^{−1}), which accounts for the discretization in the

*S*

_{A}coordinate [see Eq. (4)]. The transport across an isotherm between two isohalines is likewise (Fig. 1)

*Q*as the boundary heat flux (W m

^{−2}) and

*ρ*

_{0}as a constant reference density (kg m

^{−3}), such that ∇

*f*

_{Θ}is the rate of cooling of a water parcel due to the divergence of the heat flux (K s

^{−1}). The quantity

*L*

_{Θ}is a correction term (m

^{3}s

^{−1}), which accounts for the discretization in the Θ coordinate.

*L*

_{S}and

*L*

_{Θ}terms are referred to as the local response term in Groeskamp et al. (2014a), defined as

*S*

_{A}and Θ but does not cause a transformation across the limits of the integration. They are a correction for the discretization that vanishes in the continuous limit where Δ

*S*

_{A}and ΔΘ tend to zero. For a continuous form of Eqs. (2) and (3), see Hieronymus et al. (2014). The

*L*

_{S}and

*L*

_{Θ}terms turn out to be small, as we use a relatively high resolution in (

*S*

_{A}, Θ) coordinates in this study.

We do not account for the explicit transport of water through the atmosphere, as this is very small relative to the transformation due to air–sea fluxes (evaporation, precipitation, and runoff are effectively represented as the addition and removal of salt).

*K*

_{H}, isopycnal

*K*

_{I}, and isotropic D. We represent each of these by tensors that vary in space (

*D*

^{inv}, respectively, and

*F*

_{S}to represent the contribution of surface fluxes, such that we can rewrite Eq. (6) as

For the purposes of our inverse method Ψ, *D*^{inv} are unknown, while the *A*, *F*, and *L* terms can be estimated from observations. In section 4, we will describe the inversion that solves for the unknowns Ψ, *D*^{inv}. In the following section 3 we describe the source of our estimates of the surface flux and *S*_{A} and Θ distributions and mixing tensor structures that allow us to calculate the *A*, *F*, and *L* terms in Eqs. (7) and (8).

## 3. Data and estimated terms

### a. Ocean climatology

The observation-based climatology for all latitudes north of 75°S is given by the CSIRO Atlas of Regional Seas 2009 (CARS; Ridgway et al. 2002; Ridgway and Dunn 2003). CARS provides in situ temperature *T*, practical salinity *S*_{P} at a resolution of 0.5° × 0.5° grid spacing, and 79 vertical levels for each month of a standard year. Here, we use monthly averaged values and apply the International Thermodynamic Equation Of Seawater—2010 (TEOS-10) software (IOC et al. 2010; McDougall and Barker 2011) to convert the CARS data to *S*_{A} and Θ. The resulting data are stabilized (such that density increases with depth) using a minimal adjustment of *S*_{A} and Θ, within the measurement error (Jackett and McDougall 1995; Barker and McDougall 2016, manuscript submitted to *J. Atmos. Oceanic Technol.*).

### b. Boundary forcing

The freshwater (*E* − *P* − *R*) and heat (*Q*) fluxes are taken from the product of Yeager and Large (2008) commonly known as CORE2 (although CORE2 specifically relates to the atmospheric state from which the fluxes are derived; Large and Yeager 2009). The Yeager and Large (2008) product has a spatial resolution of 1° × 1° grid spacing and is converted to a standard year by averaging surface fluxes for each calendar month and interpolating onto the CARS grid.

### c. Horizontal and isopycnal mixing

We separate the effect of the spatial structure of mixing by mesoscale eddies into a tensor that operates on horizontal gradients

We base our spatial variation of the magnitude of eddy diffusion on the recent estimate of Cole et al. (2015), providing the scalar field *K*^{Cole}. Cole et al. (2015), from Argo profiles between the base of the winter mixed layer to approximately 2000 m, used salinity anomalies and gradients to resolve the three-dimensional structure of isopycnal mixing on a 3° × 3° grid between 60°S and 60°N.

Near the surface, the estimate from Cole et al. (2015) is comparable to estimates obtained from satellite data (Abernathey and Marshall 2013; Klocker and Abernathey 2014), so we also use Cole’s estimates to set the structure of horizontal mixing at the sea surface.

To remove spikes in these data, we impose a maximum diffusivity of 2.5 × 10^{4} m^{2} s^{−1}. The absolute values from Cole et al. (2015) reflect the eddy mixing coefficient required to parameterize unresolved eddy fluxes of a 3° × 3° resolution climatology. We aim to diagnose our own mixing value based on our 0.5° × 0.5° resolution climatology. So for the structure tensor we divide the Cole estimates by 2.5 × 10^{4} m^{2} s^{−1} to yield a scalar function between 0 and 1 (Fig. 2b).

*z*=

*h*

_{m};

*z*decreases downward from 0 at the sea surface) to 500 m beneath it. This is achieved using the tapering function

*β*such that

*h*

_{m}is calculated using TEOS-10 software based on de Boyer Montégut et al. (2004) (Fig. 2a). To include the entire pycnocline, the thickness of the transition layer is chosen to be 500 m. This represents a simplified version of the approaches of both Ferrari et al. (2008, 2010). Our structure tensors for horizontal and isopycnal mixing are then

*S*

_{y}and

*S*

_{x}are the slopes of locally referenced potential density surfaces in the zonal and meridional directions, respectively, and

*h*

_{max}(

*x*,

*y*) is the deepest level at which a diffusivity estimate exists at a particular

*x*and

*y*location, and we have chosen a decay of 75% over 1500 m, based on the horizontal average of the original data. Mixing values

*K*

^{Cole}in the mixed layer and at high latitudes are calculated using a nearest neighbor extrapolation. These results are then interpolated onto the CARS grid (Fig. 2d).

### d. Small-scale diffusion

*D*with isotropic tracer gradients. To include the effect of increased small-scale isotropic mixing near topography, we will use a structure tensor based on Bryan and Lewis (1979). Hence,

^{−4}m

^{2}s

^{−1}. Based on an inverse calculation we will find our own estimate for the magnitude of this isotropic mixing.

### e. Estimation of terms in Eqs. (7) and (8)

We use CARS and the mixing tensors to estimate the terms in square brackets in Eqs. (7) and (8). In CARS, *S*_{A} and Θ for each month are given at each (*x*, *y*, *z*) point around which six interfaces are defined to enclose a volume. Based on the CORE2 forcing, the local sources of heat *f*_{Θ} and freshwater *f*_{S} are determined. These are then multiplied by the volume of each grid, and then all points within the defined *S*_{A} and Θ interval are summed. Likewise for the *L* terms local changes are determined between months, and these are then averaged in time.

To calculate the diffusive flux divergences due to the mixing tensors, the gradient of *S*_{A} and Θ is calculated at each of the six interfaces of each box. These are then multiplied by the area of the interface and the component of the diffusive flux normal to the interface based on the mixing tensors. The divergence is then the sum of these fluxes over all six interfaces. The volume integral is then the conditional sum of these divergences.

## 4. The inverse technique

Equations (7) and (8) describe the relationship between the transformation of water across isohalines and isotherms as a result of a convergence of salt and heat, respectively, defined at *S*_{A}, Θ) coordinates has to add up to zero. The inverse method therefore solves for the diathermohaline streamfunction Ψ defined at the corners of a grid in (*S*_{A}, Θ) coordinates (Fig. 1, black colored Ψ and ΔΨ). Hence, the transports related to each ΔΨ are calculated using Eqs. (1)–(8) for volumes in (*S*_{A}, Θ) coordinates simply shifted by either Δ*S*_{A} or ΔΘ.

Now divide the Θ and *S*_{A} range into *S*_{A} range, respectively. For each grid two unique equations can be constructed, leading to **x** is a 1 × *M* vector that contains the unknowns; *N* × *M* matrix that contains 0, ±1, and the *A* terms; and **b** is a 1 × *N* vector that contains the *F* and *L* terms. For the analyses, we have chosen grid sizes in ^{−3} and ^{−1} and Θ ∈ [−2, 31]°C, respectively. The resolution is high enough to distinguish different water masses in the ocean’s interior and provide approximately equal number of equations in both *S*_{A} and Θ directions, such that both have an equal influence on the solution.

Appendix B details exactly how the inverse model is applied, including the different column and row weighting of the *N* = 34 582 and *M* = 20 600. In section 5, we discuss the results obtained from the inverse estimate based on an inversion with one particular choice of weighting parameters, as described in appendix B. The reported standard deviation associated with the results is calculated using **C**_{p} [Eq. (B2)]. It is shown that the inverse model provides a better solution then a priori estimates. Results of a sensitivity study to changes in the prior estimates and weighting are provided and discussed in appendix C. The solution described here is within the physically realistic limits and is well constrained by the data and is only moderately sensitive to different weighting of the matrix.

## 5. Results

The surface freshwater and heat fluxes lead to water mass transformation rates in cubic meters per second per Θ bin, giving units of Sverdrups (1 Sv ≡ 10^{6} m^{3} s^{−1}) per kelvin, and Sverdrups per *S*_{A} bin, giving Sverdrups per gram per kilogram, respectively. Water mass transformation due to salt diffusion by small-scale mixing processes in (*S*_{A}, Θ) coordinates is given by

### a. Surface forcing

The distribution of the surface freshwater fluxes show three areas of freshening and one area of salinification (Fig. 3a), generally acting to broaden the salinity distribution (Zika et al. 2015). For ^{−1}) and in the North Pacific (^{−1}; Fig. 3a).

The distribution of the surface heat flux shows a general tendency for warming of already warm waters (^{−1} and ^{−1}, related to heat fluxes at high latitudes.

### b. Small-scale diffusion

As mixing acts to destroy tracer gradients, the small-scale mixing results in an opposite structure to that of surface forcing (Figs. 3b,f), especially for ^{−5} m^{2} s^{−1} (Table 1). Note that our inferred diffusivity is roughly half the value of 10^{−4} m^{2} s^{−1} used by Bryan and Lewis (1979) to multiply the vertical profile [Eq. (12)].

For the different diffusivities (column 1) we show the prior estimate *x*_{0} and related a priori error *σ*_{x} calculated using the scaling argument [Eq. (B3), column 2], the inverse estimate (column 3) with the standard deviation obtained from Eq. (B2), and the results from the sensitivity study in appendix C (column 4).

As a result of the Bryan–Lewis structure, we find that for the upper 2000 m ^{2} s^{−1}. For the depths greater than 2000 m, we find that *D* increases to ^{−5} m^{2} s^{−1}. We used 2000 m to separate interior from upper ocean, as this is the depth to which small-scale mixing is required to upwell bottom waters before isopycnal upwelling becomes relevant, reducing the required mixing to close the circulation (Toggweiler and Samuels 1998; Sloyan and Rintoul 2001). The upper-ocean estimate corresponds very well with the observed values by Waterhouse et al. (2014) and the inverse estimates of Lumpkin and Speer (2007). The deep value corresponds to that of Lumpkin and Speer (2007); it is about half of canonical 1.3 × 10^{−4} of Munk (1966) and about an order of magnitude lower than that of Waterhouse et al. (2014) and Ganachaud and Wunsch (2000).

### c. Horizontal eddy diffusion

The water mass transformation rates due to horizontal eddy diffusion in the mixed layer also generally oppose the surface forcing (Figs. 3c,g). We obtain ^{2} s^{−1}, which multiplies the structure function that corresponds to a spatial resolution of 0.5° × 0.5°. Recall that Cole et al. (2015) found profiles with a maximum 3 times this value of 25 000 m^{2} s^{−1}. Their estimates represented a lower resolution and are therefore expected to have larger values than those presented here.

### d. Isopycnal eddy diffusion

By definition the isopycnal eddy diffusion only affects the interior circulation of the ocean, reducing its spread in (*S*_{A}, Θ) coordinates significantly (Figs. 3d,h). We obtain ^{2} s^{−1}, which multiplies the structure function that corresponds to a resolution of 0.5° × 0.5°, that is, 20 times smaller than that in the mixed layer. This will lead to small values in the ocean interior, as a result of the decay with depth, as observed by Cole et al. (2015). A local inverse estimate of Zika et al. (2010b) obtained small values for *K*_{I} in the interior. Even though the Zika et al. (2010b) estimate is a local estimate, it indicates that *K*_{I} may significantly decay with depth, suggesting that our result is plausible.

### e. The streamfunctions

An estimate of the nondivergent circulation in (*S*_{A}, Θ) coordinates is given by our solution for

#### 1) Tropical cells

For *S*_{A}, Θ) coordinates related to the upper 200 m of the ocean (Fig. 5) in the (sub) tropics, the surface freshwater and heat fluxes are balanced by small-scale and mesoscale mixing processes in the mixed layer. Although strong small-scale mixing near the surface is also supported by model-based studies (Ferrari and Ferreira 2011; Hieronymus et al. 2014), we find stronger near-surface mesoscale mixing than those studies. The residual water mass transformation results in a strong residual anticlockwise circulation (Fig. 4b), which may be considered to be the observation-based equivalent of the tropical cell, identified as a shallow wind-driven (near) equatorial cell (Döös et al. 2012).

Hieronymus et al. (2014) found that the tropical cell (estimated from a numerical ocean model) is composed of an Atlantic and an Indo-Pacific circulation. We also find a slight extension into the Atlantic; however, the main difference is that the extension of this cell is at similar temperatures but is centered at ^{−1}. This cell exists only in the upper 100 m of the freshest area in the Indian Ocean (Fig. 5) and is most likely related to the circulation in the Bay of Bengal, where freshwater discharge due to the summer monsoon lead to the observed low salinity (Schott and McCreary 2001; Schott et al. 2002).

#### 2) Global cell

The large clockwise-rotating (blue) circulation cell in

#### 3) Cold cells

Both ^{−1} and ^{−1} and

The Antarctic cell covers an area in (*S*_{A}, Θ) coordinates that connects surface dynamics with the interior and is related with strong surface cooling and relatively small freshwater fluxes (Figs. 3a,e). Small-scale and mixed layer mesoscale mixing is observed on the ^{−3} contour at ^{−1} (Figs. 3b,f). Hence, this circulation cell represents the Antarctic Bottom Water (AABW) and Lower Circumpolar Deep Water (LCDW) circulation and can be identified as the observational-based equivalent of the AABW cell as found by Zika et al. (2012) and Döös et al. (2012) for an ocean model. However, the AABW we observe occupies a wider

The Arctic cell, related to the Arctic and the North Pacific (subpolar) Gyre (NPG) for depths shallower than 200 m (Figs. 4b, 5), is mainly induced by surface freshwater and heat forcing (Fig. 3) and perhaps affected by the Bering Strait flow, which is a northward flow of about 1 Sv (Roach et al. 1995). The North Pacific Gyre rotates clockwise, transporting cold and freshwater equatorward on the eastern side of the basin. Water moving to the equator and subsequently to the western side of the basin will warm and the salinity will increase (evaporation), as a result of surface heat fluxes. At the western boundary, the Kuroshio transports water northward, leading to cooling and freshening by net freshwater flux from precipitation. This NPG cell has not been identified in any of the model-based versions of

The cell observed at ^{−1} and Θ = [0, 10]°C is likely a manifestation of the Arctic circulation that may be related to ice dynamics in combination with strong seasonal warming and cooling, leading to a very shallow buoyancy-driven circulation. However, because of sparseness of data in these regions this remains speculative.

Regardless of the fact that we have not included any prior knowledge of the streamfunction structure (appendix B) the THIM provides an estimate of

## 6. Discussion

In this section, we use the inverse estimate of

### a. Dissipation of turbulent kinetic energy

Using the Osborn (1980) relationship between the isotropic diffusion of small-scale mixing processes and energy dissipation in combination with our inverse estimate of ^{−2}), *ρ* is in situ density [kg m^{−3}, calculated using IOC et al. (2010) software], and *V* is the volume (m^{3}). These bulk estimates contain errors related to the climatology, choice of

The global, depth-integrated distribution of *P* is dominated by the high values of

The value of *D*^{inv} is primarily set as a result of compensating surface fluxes, which, through the use of the Bryan and Lewis (1979) structure function, leads to strong constraints on the interior estimate of *D*. We anticipate that the interior estimate of *D* can be improved using more complex parameterizations that separate the different physical mechanisms, such as surface mixing from internal wave breaking in the deep ocean (Melet et al. 2013; Nikurashin and Ferrari 2013). This could lead to more detailed estimates of the required energy for the different small-scale mixing processes and shed light on how small-scale mixing influences the global overturning circulation by transporting bottom water to the level where isopycnal upwelling plays a significant role (Toggweiler and Samuels 1998; Sloyan and Rintoul 2001; Ferrari 2014).

### b. Uncertainty and improvements to the solution

Volume divergences in (*S*_{A}, Θ) coordinates are due to the imbalance between surface and diffusive fluxes of (equivalent) salt and heat and would be zero in a steady-state ocean. The THIM simultaneously minimizes volume divergences for each grid over the global (*S*_{A}, Θ) domain by adjusting the diffusion coefficients, while accounting for an exactly nondivergent

*T*and the transformation due to surface fluxes

*T*

^{F}, small-scale mixing

*T*

^{D}, horizontal mesoscale mixing

*T*

^{H}, and isopycnal mixing

*T*

^{I}(Fig. 7).

We want to emphasize that regardless of the models for mixing that we use, the THIM gives an exactly balanced solution for the total volume transport. This is implicit in solving for the streamfunction

For a steady-state ocean, *T* would be zero (green line, Fig. 7), which mainly requires *T*^{F} to be compensated by mixing (*T*^{D}, *T*^{H}, and *T*^{I}). We suggest that the imbalance of *T* is mainly due to the lack of variability that we allow in *D* and *K*, resulting in imbalances within some isopycnal ranges. For example, the imbalance at ^{−3} is most likely because the implemented *S*_{A}, Θ) domain. This limits the THIM to balance *T*^{F} correctly everywhere. That is, our mixing estimates are more tailored for diagnosing the transformations at the median ocean

Between ^{−3}, the total water mass transformation is dominated by surface fluxes *T*^{F}, indicating transformation to lighter density water. This result agrees with that obtained by Iudicone et al. (2008), who used a numerical ocean model with a different surface forcing. The term *T*^{I} is significant for ^{−3} and is positive and of the same order of magnitude as that of Klocker and McDougall (2010), who calculated *T*^{I} using the WOCE climatology (Gouretski and Koltermann 2004) with a constant ^{2} s^{−1}.

Groeskamp et al. (2016) estimate water mass transformation by directly applying the estimates of Cole et al. (2015) to the *World Ocean Atlas 2013* (*WOA13*; Boyer et al. 2013), without imposing a balance on the water mass transformation budget. They show similar patterns for both *T*^{H} and *T*^{I} but estimate about twice the magnitude for *T*^{H} and 5 times the magnitude of *T*^{I} as found in this study. We note, however, that *WOA13* has a 1° × 1° grid spacing, such that a direct application of Cole et al. (2015) (approximate 3° × 3° grid spacing) may therefore result in an overestimation of the transformation rates by mesoscale mixing.

Improvements in the representation of mixing in the inverse model would improve the estimates of water mass transformation in

The sensitivity of the estimates due to the aforementioned mixing choices and structure functions is important to study but is beyond the scope of this paper; here, we focus on the first application of the THIM to observations.

### c. Can isopycnal mixing be so small?

Section 5e showed that the estimate of *S*_{A}, Θ) coordinates from observations. Although there is an imbalance between surface and diffusive fluxes, over the whole (*S*_{A}, Θ) domain, the inferred diffusivities are well constrained. In addition,

Previous observationally based studies have not provided a global, full-depth estimate of *K*_{I}. However, estimates from Cole et al. (2015), between the winter mixed layer and 2000 m, are smaller than their estimate for the deep ocean. Generally, we find that *K*_{H} = *K*_{I}, while we found that *K*_{H}/*K*_{I} = *O*(20). This means that the maximum interior mesoscale mixing is about 20 times smaller than that in the mixed layer. Is this a realistic estimate?

To test if *K*_{I} is small because of the choice for *D* and one global isopycnal diffusivity *K*, without the use of structure functions (i.e., no spatial variation for *K* or *D* was included). This resulted in ^{2} s^{−1} and ^{2} s^{−1}. The smallness of *K* motivated a split of *K* into a mixed layer *K*_{H} and interior *K*_{I} component, resulting in a similar estimate for *D*, but *K*_{H} = 423 ± 4 m^{2} s^{−1} and *K*_{I} = 12 ± 1 m^{2} s^{−1}. This leads to a reasonable estimate of *D* and *K*_{H} (considering CARS’s spatial resolution), but *K*_{I} still remains small. This suggest that, with or without using *K*_{I} must consistently be much smaller than *K*_{H} in order to obtain realistic water mass transformation in the deep ocean.

Based on the discussion above, increasing the number of scaling coefficients that are solved for in the vertical will most likely result in higher values of *K*_{I} up to 1000 m^{2} s^{−1} in the deep ocean.

Finally, we have tested the sensitivity of

## 7. Conclusions

We have applied the THIM to an ocean climatology in combination with surface flux products, resulting in observationally based estimates of horizontal and isopycnal eddy diffusion coefficients *K*_{H} and *K*_{I}, respectively, the small-scale mixing coefficient *D*, and the diathermohaline streamfunction Ψ, which is representative of the global ocean circulation. The THIM provides an inverse solution that minimizes the divergence of processes based on transformations in individual *S*_{A}, Θ) coordinates.

Here, Ψ provides an observational-based quantification of the globally interconnected ocean circulation induced by water mass transformation (Broecker 1991) and 1) shows a robust tropical cell with an extension related to circulation in the Bay of Bengal; 2) suggests the existence of a cell related to the North Pacific (subpolar) Gyre and the Bering Strait outflow, not previously identified by model-based calculations of Ψ; 3) shows a wider spread of the AABW cell compared to model studies; and 4) has a general structure comparable to models. Given that Ψ can now be estimated from observations, we suggest that Ψ has emerged as a unique new diagnostic providing the ability to compare and contrast fundamental properties such as circulation, mixing, and heat and freshwater fluxes between models and observations. The Ψ would be valuable as a standard diagnostic for model intercomparison exercises (Danabasoglu et al. 2014, 2016).

Standard relationships between *D*, the stratification, and energy dissipation show that 1.34 TW of energy is used for small-scale mixing, of which 0.89 TW is used in the upper 500 m of the ocean and 0.17 TW is used below 2000 m. Hence, most mixing occurs near the surface, but small-scale mixing is important below 2000 m to close the global ocean circulation.

Changes in the magnitude or spatial variation of *K*_{I} in numerical ocean models have a significant influence on reducing systematic drift (Ferreira et al. 2005; Danabasoglu and Marshall 2007), prediction of the global temperature (Pradal and Gnanadesikan 2014), and the stability of the overturning circulation (Sijp et al. 2006). The results presented here suggest that *K*_{H}/*K*_{I} = *O*(20), meaning that the maximum interior mesoscale mixing is about 20 times smaller than that in the mixed layer. This effect is not included in most ocean models but is potentially important for the aforementioned processes. We conclude that it is very likely that isopycnal eddy mixing in the ocean interior is much smaller than currently assumed, although its vertical structure requires further study. The diffusion coefficient for the eddy-induced, quasi-Stokes (GM) advection and the Redi parameter for tracer diffusion are related, suggesting that the results obtained here may also impact the GM parameterization (Abernathey et al. 2013).

Our results provide the first observationally based, globally constrained, full-depth, isopycnal mixing estimate and reveal unexpectedly small values of the isopycnal diffusivity, compared to that commonly quoted in the literature. Little is known about the vertical structure of isopycnal mixing, while it has a large impact on the sensitivity of the climate and future climate change. Further development of THIM to more accurately estimate this structure from observations is therefore essential.

## Acknowledgments

SG gratefully acknowledges support by the NASA Grant NNX14AI46G. BMS was supported by the Australian Climate Change Science Program, jointly funded by the Department of the Environment and CSIRO. JDZ has been supported by the Natural Environment Research Council Project NE/K012932/1. We thank Daniele Iudicone, Cimarron Wortham, Ryan Abernathey, Andreas Klocker, Stephen Griffies, Jonas Nycander, and Ken Ridgway for useful discussions that have improved the manuscript. We thank Sylvia Cole for providing the mesoscale diffusivity data. We thank Paul Barker for providing preliminary hydrographic stabilization software and Terry O’Kane for providing computational facilities in early stages of this project.

## APPENDIX A

### Isopycnal Tracer Gradients

The slopes used in Eq. (10), used to calculate isopycnal *S*_{A} and Θ gradients, are given by *ρ*^{l} is the locally referenced potential density, that is, it is a potential density referred to the pressure at which one calculated the gradient. This method requires dividing by *x* and *y* direction, respectively. The values for

Tracers are defined at the a T grid at grid point *I* is the total number of longitude location in the ocean. In a similar way *j* and *J* represent latitude, *k* and *K* represent depth, and *τ* and *T* represent time. The tracer is given in

Consider only longitude and depth, such that *κ* is a fixed depth. Using linear interpolation, *i*, *j*, and *k*, linear interpolation is used to provide the values of the slopes back on the original T-grid depths

## APPENDIX B

### The Exact Setup of the Inverse Model

^{T}is the transpose,

#### a. The a priori estimates and constraints

*F*terms in Eqs. (7) and (8)] is balanced by convergence due to mixing. It follows that their maximum transformation rates should be of comparable magnitude. As such we use an a priori estimate of the diffusion coefficient such that, when multiplied by the

*A*terms in Eqs. (7) and (8), they do not exceed the water mass transformation given by the boundary forcing (the

*F*terms). For the small-scale diffusion coefficient, we use an a prior estimate of

Using the prior estimates of the diffusion coefficients, we have omitted equations that are associated with transports less than 1% of the maximum transport, as these equations do not have a signal to noise ratio that adds information to the solution. We have also omitted equations that are isolated in (*S*_{A}, Θ) coordinates (i.e., do not bound another bin with significant water mass transformation), as such equations are unconstrained.

#### b. Row weighting

To construct row weighting *S*_{A}, Θ) grid by combining *A*, *F*, and *L* terms in Eqs. (7) and (8) with the a priori mixing coefficients

*S*

_{A}, Θ) coordinates the a priori volume divergence

*S*

_{A}. An equivalent expression can be obtained for the Θ direction.

#### c. Column weighting

We will allow for a standard error of twice the prior estimate of the diffusion coefficients, leading to

The lack of information for

To obtain an approximate structure for *S*_{A}, Θ) coordinates, than the related *S*_{A}, Θ) coordinates and is thus calculated as the average of four such transformation rates. As

The results of this particular setup of the inversion are presented in section 5. Results of a sensitivity study to changes in the prior estimates and weighting are provided and discussed in appendix C. For the solution discussed, we have obtained

## APPENDIX C

### Sensitivity Study

The formal estimate of the random error of the inverse estimate is given by the square root of the diagonal of the posterior covariance matrix

Although we do not change our a priori estimate of

This sensitivity study results in 3 × 4^{3} = 192 inverse solutions that all meet the criteria of

We find that ^{2} s^{−1}, a difference of almost two orders of magnitude between the extrema. The solution for *K*_{H} ∈ [3800, 10 000] m^{2} s^{−1} is most sensitive to (in order of significance) ^{2} s^{−1} is most sensitive to

The extremes for *K*_{H} is less than a factor 3, while that of *D* is less than a factor 1.5. This shows *K*_{I} is more sensitive to changes in the prior estimates. However, we note that even when taking the largest estimate of *K*_{I} and the smallest estimate of *K*_{H}, the ratio *K*_{H}/*K*_{I} = 3800/1400 = *O*(3) and still implies significantly reduced magnitude of mesoscale mixing in the interior. Hence, the sensitivity study does not change the key result of this study.

## REFERENCES

Abernathey, R. P., and J. Marshall, 2013: Global surface eddy diffusivities derived from satellite altimetry.

,*J. Geophys. Res. Oceans***118**, 901–916, doi:10.1002/jgrc.20066.Abernathey, R. P., D. Ferreira, and A. Klocker, 2013: Diagnostics of isopycnal mixing in a circumpolar channel.

,*Ocean Modell.***72**, 1–16, doi:10.1016/j.ocemod.2013.07.004.Abernathey, R. P., I. Cerovecki, P. R. Holland, E. Newsom, M. Mazloff, and L. D. Talley, 2016: Water-mass transformation by sea ice in the upper branch of the Southern Ocean overturning.

,*Nat. Geosci.***9**, 596–601, doi:10.1038/ngeo2749.Alford, M. H., 2001: Internal swell generation: The spatial distribution of energy flux from the wind to mixed layer near-inertial motions.

,*J. Phys. Oceanogr.***31**, 2359–2368, doi:10.1175/1520-0485(2001)031<2359:ISGTSD>2.0.CO;2.Alford, M. H., M. C. Gregg, and M. A. Merrifield, 2006: Structure, propagation, and mixing of energetic baroclinic tides in Mamala Bay, Oahu, Hawaii.

,*J. Phys. Oceanogr.***36**, 997–1018, doi:10.1175/JPO2877.1.Alford, M. H., M. F. Cronin, and J. M. Klymak, 2012: Annual cycle and depth penetration of wind-generated near-inertial internal waves at Ocean Station Papa in the northeast Pacific.

,*J. Phys. Oceanogr.***42**, 889–909, doi:10.1175/JPO-D-11-092.1.Badin, G., and R. G. Williams, 2010: On the buoyancy forcing and residual circulation in the Southern Ocean: The feedback from Ekman and eddy transfer.

,*J. Phys. Oceanogr.***40**, 295–310, doi:10.1175/2009JPO4080.1.Badin, G., R. G. Williams, Z. Jing, and L. Wu, 2013: Water mass transformations in the Southern Ocean diagnosed from observations: Contrasting effects of air–sea fluxes and diapycnal mixing.

,*J. Phys. Oceanogr.***43**, 1472–1484, doi:10.1175/JPO-D-12-0216.1.Boyer, T., and Coauthors, 2013:

*World Ocean Database 2013*. NOAA Atlas NESDIS 72, 209 pp., doi:10.7289/V5NZ85MT.Broecker, W. S., 1991: The great ocean conveyor.

,*Oceanography***4**, 79–89.Bryan, K., and L. J. Lewis, 1979: A water mass model of the World Ocean.

,*J. Geophys. Res.***84**, 2503–2517, doi:10.1029/JC084iC05p02503.Cole, S. T., C. Wortham, E. Kunze, and W. B. Owens, 2015: Eddy stirring and horizontal diffusivity from Argo float observations: Geographic and depth variability.

,*Geophys. Res. Lett.***42**, 3989–3997, doi:10.1002/2015GL063827.Danabasoglu, G., and J. Marshall, 2007: Effects of vertical variations of thickness diffusivity in an ocean general circulation model.

,*Ocean Modell.***18**, 122–141, doi:10.1016/j.ocemod.2007.03.006.Danabasoglu, G., and Coauthors, 2014: North Atlantic simulations in Coordinated Ocean-Ice Reference Experiments phase II (CORE-II). Part I: Mean states.

,*Ocean Modell.***73**, 76–107, doi:10.1016/j.ocemod.2013.10.005.Danabasoglu, G., and Coauthors, 2016: North Atlantic simulations in Coordinated Ocean-Ice Reference Experiments phase II (CORE-II). Part II: Inter-annual to decadal variability.

,*Ocean Modell.***97**, 65–90, doi:10.1016/j.ocemod.2015.11.007.D’Asaro, E. A., 1985: The Energy flux from the wind to near-inertial motions in the surface mixed layer.

,*J. Phys. Oceanogr.***15**, 1043–1059, doi:10.1175/1520-0485(1985)015<1043:TEFFTW>2.0.CO;2.de Boyer Montégut, C., G. Madec, A. S. Fischer, A. Lazar, and D. Iudicone, 2004: Mixed layer depth over the global ocean: An examination of profile data and a profile-based climatology.

,*J. Geophys. Res.***109**, C12003, doi:10.1029/2004JC002378.de Lavergne, C., G. Madec, J. Le Sommer, A. J. G. Nurser, and A. C. Naveira Garabato, 2016: On the consumption of Antarctic Bottom Water in the abyssal ocean.

,*J. Phys. Oceanogr.***46**, 635–661, doi:10.1175/JPO-D-14-0201.1.Döös, K., J. Nilsson, J. Nycander, L. Brodeau, and M. Ballarotta, 2012: The World Ocean thermohaline circulation.

,*J. Phys. Oceanogr.***42**, 1445–1460, doi:10.1175/JPO-D-11-0163.1.Ferrari, R., 2014: Oceanography: What goes down must come up.

,*Nature***513**, 179–180, doi:10.1038/513179a.Ferrari, R., and D. Ferreira, 2011: What processes drive the ocean heat transport?

,*Ocean Modell.***38**, 171–186, doi:10.1016/j.ocemod.2011.02.013.Ferrari, R., J. C. McWilliams, V. M. Canuto, and M. Dubovikov, 2008: Parameterization of eddy fluxes near oceanic boundaries.

,*J. Climate***21**, 2770–2789, doi:10.1175/2007JCLI1510.1.Ferrari, R., S. M. Griffies, A. G. Nurser, and G. K. Vallis, 2010: A boundary-value problem for the parameterized mesoscale eddy transport.

,*Ocean Modell.***32**, 143–156, doi:10.1016/j.ocemod.2010.01.004.Ferreira, D., J. Marshall, and P. Heimbach, 2005: Estimating eddy stresses by fitting dynamics to observations using a residual-mean ocean circulation model and its adjoint.

,*J. Phys. Oceanogr.***35**, 1891–1910, doi:10.1175/JPO2785.1.Ganachaud, A., and C. Wunsch, 2000: Improved estimates of global ocean circulation, heat transport and mixing from hydrographic data.

,*Nature***408**, 453–457, doi:10.1038/35044048.Garabato, A. C. N., K. L. Polzin, B. A. King, K. J. Heywood, and M. Visbeck, 2004: Widespread intense turbulent mixing in the Southern Ocean.

,*Science***303**, 210–213, doi:10.1126/science.1090929.Gent, P. R., and J. C. McWilliams, 1990: Isopycnal mixing in ocean circulation models.

,*J. Phys. Oceanogr.***20**, 150–155, doi:10.1175/1520-0485(1990)020<0150:IMIOCM>2.0.CO;2.Gent, P. R., J. Willebrand, T. J. McDougall, and J. C. McWilliams, 1995: Parameterizing eddy-induced tracer transports in ocean circulation models.

,*J. Phys. Oceanogr.***25**, 463–474, doi:10.1175/1520-0485(1995)025<0463:PEITTI>2.0.CO;2.Gordon, A. L., 1986: Interocean exchange of thermocline water.

,*J. Geophys. Res.***91**, 5037–5046, doi:10.1029/JC091iC04p05037.Gouretski, V., and K. P. Koltermann, 2004: WOCE global hydrographic climatology: A technical report. Berichte des Bundesamtes für Seeschifffahrt und Hydrographie 35/2004, 52 pp.

Graham, F. S., and T. J. McDougall, 2013: Quantifying the nonconservative production of Conservative Temperature, potential temperature, and entropy.

,*J. Phys. Oceanogr.***43**, 838–862, doi:10.1175/JPO-D-11-0188.1.Griffies, S. M., 2004:

*Fundamentals of Ocean Climate Models.*Princeton University Press, 518 pp.Griffies, S. M., 2012: Elements of the Modular Ocean Model (MOM). NOAA Geophysical Fluid Dynamics Laboratory Ocean Group Tech. Rep. 7, 618 pp.

Groeskamp, S., J. D. Zika, T. J. McDougall, B. M. Sloyan, and F. Laliberté, 2014a: The representation of ocean circulation and variability in thermodynamic coordinates.

,*J. Phys. Oceanogr.***44**, 1735–1750, doi:10.1175/JPO-D-13-0213.1.Groeskamp, S., J. D. Zika, B. M. Sloyan, T. J. McDougall, and P. C. McIntosh, 2014b: A thermohaline inverse method for estimating diathermohaline circulation and mixing.

,*J. Phys. Oceanogr.***44**, 2681–2697, doi:10.1175/JPO-D-14-0039.1.Groeskamp, S., R. P. Abernathey, and A. Klocker, 2016: Water mass transformation by cabbeling and thermobaricity.

,*Geophys. Res. Lett.***43**, 10 835–10 845, doi:10.1002/2016GL070860.Hanawa, K., and L. D. Talley, 2001: Mode waters. Ocean Circulation and Climate, G. Siedler, J. Gould, and J. Church, Eds., International Geophysics Series, Vol. 103, Academic Press, 373–386.

Hieronymus, M., J. Nilsson, and J. Nycander, 2014: Water mass transformation in salinity–temperature space.

,*J. Phys. Oceanogr.***44**, 2547–2568, doi:10.1175/JPO-D-13-0257.1.Holloway, G., 1986: Estimation of oceanic eddy transports from satellite altimetry.

,*Nature***323**, 243–244, doi:10.1038/323243a0.IOC, SCOR, and IAPSO, 2010: The International Thermodynamic Equation of Seawater—2010: Calculation and use of thermodynamic properties. Intergovernmental Oceanographic Commission, Manuals and Guides 56, 220 pp. [Available online at http://www.teos-10.org/pubs/TEOS-10_Manual.pdf.]

Iselin, C. O., 1939: The influence of vertical and lateral turbulence on the characteristics of the waters at mid-depths.

,*Eos, Trans. Amer. Geophys. Union***20**, 414–417, doi:10.1029/TR020i003p00414.Iudicone, D., G. Madec, and T. J. McDougall, 2008: Water-mass transformations in a neutral density framework and the key role of light penetration.

,*J. Phys. Oceanogr.***38**, 1357–1376, doi:10.1175/2007JPO3464.1.Iudicone, D., K. B. Rodgers, I. Stendardo, O. Aumont, G. Madec, L. Bopp, O. Mangoni, and M. Ribera d’Alcala, 2011: Water masses as a unifying framework for understanding the Southern Ocean carbon cycle.

,*Biogeosciences***8**, 1031–1052, doi:10.5194/bg-8-1031-2011.Jackett, D. R., and T. J. McDougall, 1995: Minimal adjustment of hydrographic profiles to achieve static stability.

,*J. Atmos. Oceanic Technol.***12**, 381–389, doi:10.1175/1520-0426(1995)012<0381:MAOHPT>2.0.CO;2.Klocker, A., and T. J. McDougall, 2010: Quantifying the consequences of the ill-defined nature of neutral surfaces.

,*J. Phys. Oceanogr.***40**, 1866–1880, doi:10.1175/2009JPO4212.1.Klocker, A., and R. Abernathey, 2014: Global patterns of mesoscale eddy properties and diffusivities.

,*J. Phys. Oceanogr.***44**, 1030–1046, doi:10.1175/JPO-D-13-0159.1.Kunze, E., E. Firing, J. M. Hummon, T. K. Chereskin, and A. M. Thurnherr, 2006: Global abyssal mixing inferred from lowered ADCP shear and CTD strain profiles.

,*J. Phys. Oceanogr.***36**, 1553–1576, doi:10.1175/JPO2926.1.LaCasce, J. H., R. Ferrari, J. Marshall, R. Tulloch, D. Balwada, and K. Speer, 2014: Float-derived isopycnal diffusivities in the dimes experiment.

,*J. Phys. Oceanogr.***44**, 764–780, doi:10.1175/JPO-D-13-0175.1.Large, W. G., and S. G. Yeager, 2009: The global climatology of an interannually varying air–sea flux data set.

,*Climate Dyn.***33**, 341–364, doi:10.1007/s00382-008-0441-3.Ledwell, J. R., A. J. Watson, and C. S. Law, 1993: Evidence for slow mixing across the pycnocline from an open-ocean tracer-release experiment.

,*Nature***364**, 701–703, doi:10.1038/364701a0.Ledwell, J. R., A. J. Watson, and C. S. Law, 1998: Mixing of a tracer in the pycnocline.

,*J. Geophys. Res.***103**, 21 499–21 529, doi:10.1029/98JC01738.Lumpkin, R., and K. Speer, 2007: Global ocean meridional overturning.

,*J. Phys. Oceanogr.***37**, 2550–2562, doi:10.1175/JPO3130.1.Marsh, R., 2000: Recent variability of the North Atlantic thermohaline circulation inferred from surface heat and freshwater fluxes.

,*J. Climate***13**, 3239–3260, doi:10.1175/1520-0442(2000)013<3239:RVOTNA>2.0.CO;2.Marshall, D., 1997: Subduction of water masses in an eddying ocean.

,*J. Mar. Res.***55**, 201–222, doi:10.1357/0022240973224373.Marshall, J., D. Jamous, and J. Nilsson, 1999: Reconciling thermodynamic and dynamic methods of computation of water-mass transformation rates.

,*Deep-Sea Res. I***46**, 545–572, doi:10.1016/S0967-0637(98)00082-X.McDougall, T. J., 1987a: Neutral surfaces.

,*J. Phys. Oceanogr.***17**, 1950–1964, doi:10.1175/1520-0485(1987)017<1950:NS>2.0.CO;2.McDougall, T. J., 1987b: Thermobaricity, cabbeling, and water-mass conversion.

,*J. Geophys. Res.***92**, 5448–5464, doi:10.1029/JC092iC05p05448.McDougall, T. J., 2003: Potential enthalpy: A conservative oceanic variable for evaluating heat content and heat fluxes.

,*J. Phys. Oceanogr.***33**, 945–963, doi:10.1175/1520-0485(2003)033<0945:PEACOV>2.0.CO;2.McDougall, T. J., and P. C. McIntosh, 2001: The temporal-residual-mean velocity. Part II: Isopycnal interpretation and the tracer and momentum equations.

,*J. Phys. Oceanogr.***31**, 1222–1246, doi:10.1175/1520-0485(2001)031<1222:TTRMVP>2.0.CO;2.McDougall, T. J., and P. M. Barker, 2011: Getting started with TEOS-10 and the Gibbs Seawater (GSW) oceanographic toolbox. SCOR/IAPSO WG127, 28 pp. [Available online at www.teos-10.org/pubs/Getting_Started.pdf.]

McDougall, T. J., D. R. Jackett, F. J. Millero, R. Pawlowicz, and P. M. Barker, 2012: A global algorithm for estimating Absolute Salinity.

,*Ocean Sci.***8**, 1123–1134, doi:10.5194/os-8-1123-2012.McDougall, T. J., S. Groeskamp, and S. M. Griffies, 2014: On geometrical aspects of interior ocean mixing.

,*J. Phys. Oceanogr.***44**, 2164–2175, doi:10.1175/JPO-D-13-0270.1.McIntosh, P. C., and S. R. Rintoul, 1997: Do box inverse models work?

,*J. Phys. Oceanogr.***27**, 291–308, doi:10.1175/1520-0485(1997)027<0291:DBIMW>2.0.CO;2.Melet, A., R. Hallberg, S. Legg, and K. Polzin, 2013: Sensitivity of the ocean state to the vertical distribution of internal-tide-driven mixing.

,*J. Phys. Oceanogr.***43**, 602–615, doi:10.1175/JPO-D-12-055.1.Menke, W., 1984:

*Geophysical Data Analysis: Discrete Inverse Theory*. Elsevier, 260 pp.Meyer, A., B. M. Sloyan, K. L. Polzin, H. E. Phillips, and N. L. Bindoff, 2015: Mixing variability in the Southern Ocean.

,*J. Phys. Oceanogr.***45**, 966–987, doi:10.1175/JPO-D-14-0110.1.Millero, F. J., R. Feistel, D. G. Wright, and T. J. McDougall, 2008: The composition of standard seawater and the definition of the reference-composition salinity scale.

,*Deep-Sea Res. I***55**, 50–72, doi:10.1016/j.dsr.2007.10.001.Munk, W., 1966: Abyssal recipes.

,*Deep-Sea Res. Oceanogr. Abstr.***13**, 707–730, doi:10.1016/0011-7471(66)90602-4.Munk, W., and C. Wunsch, 1998: Abyssal recipes II: Energetics of tidal and wind mixing.

,*Deep-Sea Res. I***45**, 1977–2010, doi:10.1016/S0967-0637(98)00070-3.Nguyen, A. T., D. Menemenlis, and R. Kwok, 2009: Improved modeling of the Arctic halocline with a subgrid-scale brine rejection parameterization.

,*J. Geophys. Res.***114**, C11014, doi:10.1029/2008JC005121.Nikurashin, M., and R. Ferrari, 2011: Global energy conversion rate from geostrophic flows into internal lee waves in the deep ocean.

,*Geophys. Res. Lett.***38**, L08610, doi:10.1029/2011GL046576.Nikurashin, M., and R. Ferrari, 2013: Overturning circulation driven by breaking internal waves in the deep ocean.

,*Geophys. Res. Lett.***40**, 3133–3137, doi:10.1002/grl.50542.