Generated using version 3.0 of the official AMS LATEX template
Turbulence in a sheared, salt-fingering-favorable environment:1
Anisotropy and effective diffusivities2
Satoshi Kimura ∗ William Smyth
College of Oceanic and Atmospheric Sciences, Oregon State University, Oregon
3
Eric Kunze
School of Earth and Ocean Sciences, University of Victoria, Canada
4
∗Corresponding author address: College of Oceanic and Atmospheric Sciences, Oregon State UniversityOcean Admin Bldg, Corvallis, OR 97331, USA.E-mail: [email protected]
1
ABSTRACT5
Direct numerical simulations (DNS) of a shear layer with salt-fingering-favorable stratifica-6
tion have been performed for different Richardson numbers Ri and density ratios Rρ. When7
the Richardson number is infinite (unsheared case), the primary instability is square plan-8
form salt-fingering, alternating cells of rising and sinking fluid. In the presence of shear,9
salt-fingering takes the form of salt sheets, planar regions of rising and sinking fluid, aligned10
parallel to the sheared flow. After the onset of secondary instability, the flow becomes11
turbulent; however, our results indicate that the continued influence of the primary insta-12
bility biases estimates of the turbulent kinetic energy dissipation rate. Thermal and saline13
buoyancy gradients become more isotropic than the velocity gradients at dissipation scales.14
Estimates of the turbulent kinetic energy dissipation rate by assuming isotropy in the ver-15
tical direction can underestimate its true value by a factor of 2 to 3, whereas estimates of16
thermal and saline dissipation rates are relatively accurate. Γ approximated by assuming17
isotropy agrees well with observational estimates, but is larger than the true value of Γ by18
approximated a factor of 2. The transport associated with turbulent salt sheets is quanti-19
fied by thermal and saline effective diffusivities. Salt sheets are ineffective at transporting20
momentum. Thermal and saline effective diffusivities decrease with decreasing Ri, despite21
the added energy source provided by background shear. Nondimensional quantities, such as22
thermal to saline flux ratio, are close to the predictions of linear theory.23
1
1. Introduction24
Salt-fingering-favorable stratification occurs when the gravitationally unstable vertical25
gradient of salinity is stabilized by that in temperature. In such conditions, the faster diffu-26
sion of heat than salt can generate cells of rising and sinking fluid known as salt fingers, which27
take a variety of planforms such as squares, rectangles, and sheets (Schmitt 1994b; Proctor28
and Holyer 1986). Salt-fingering-favorable stratification is found in much of the tropical29
and subtropical pycnocline (You 2002). The most striking signatures are in thermohaline30
staircases, which are stacked layers of different water-types separated by sharp thermal and31
saline gradients. These are found at several locations, such as in the subtropical confluence32
east of Barbados, under the Mediterranean and Red Sea salt tongues and in the Tyrrhenian33
Sea (Tait and Howe 1968; Lambert and Sturges 1977; Schmitt 1994a).34
Salt-fingering is subjected to vertically varying horizontal currents (background shear)35
(Gregg and Sanford 1987; St. Laurent and Schmitt 1999). Zhang et al. (1999) found a36
significant relationship between salt fingering and shear in a model of the North Atlantic.37
A theory for the possible relationship between salt fingering and shear has been introduced38
by Schmitt (1990). In the presence of shear, salt-fingering instability is supplanted by salt-39
sheet instability, alternating planar regions of rising and sinking fluid, aligned parallel to the40
sheared flow (Linden 1974). While theories for the initial growth of salt-fingering and salt-41
sheet instabilities are well-established (Stern 1960; Linden 1974; Schmitt 1979; Kunze 2003;42
Smyth and Kimura 2007), the vertical fluxes of momentum, heat, and salt in the non-linear43
regime are not well understood. We call this non-linear regime turbulent salt-fingering.44
Dissipation rates of property variances in double-diffusive turbulence can be measured45
in the ocean by microstructure profilers, but the interpretation of these results has relied46
on Kolmogorov’s hypothesis of isotropic turbulence (Gregg and Sanford 1987; Lueck 1987;47
Hamilton et al. 1989; St. Laurent and Schmitt 1999; Inoue et al. 2008). Kolmogorov (1941)48
2
proposed that small-scale turbulence statistics are universal in the limit of high Reynolds49
number. According to this hypothesis, anisotropy of the energy-containing scales is lost in50
the turbulent energy cascade, so that the small scales, where energy is finally dissipated,51
are statistically isotropic. The assumption of small-scale isotropy greatly simplifies both the52
theory and modeling of turbulence, as well as interpretation of microstructure measurements.53
Because the Reynolds number in oceanic salt-fingering systems is O(10) (McDougall and54
Taylor 1984), the validity of this assumption is questionable. We will examine both mixing55
rates and the isotropy of the dissipation range in turbulent salt-fingering by means of direct56
numerical simulations of diffusively unstable shear layers.57
Based on observations of shear-driven turbulence in flow over a sill, Gargett et al. (1984)58
concluded that the isotropy assumption was accurate as long as the separation between59
Ozmidov and Kolmogorov length scale was sufficiently large. Itsweire et al. (1993) tested60
isotropic approximations on flows with uniform shear and stratification for Richardson num-61
ber using DNS, and found that dissipation rates could be underestimated by factors of 2 to62
4 at low Reynolds number. Smyth and Moum (2000) extended the analysis to a localized63
shear layer and found similar results.64
Estimations of dissipation rates, combined with the Osborn and Cox (1972) diffusivity65
model, can furnish estimates of the effective diffusivities of heat, salt, and momentum. Ef-66
fective diffusivities are used to parameterize turbulent fluxes in order to model large-scale67
phenomena affected by mixing processes. These applications range from fine-scale thermo-68
haline intrusions (e.g. Toole and Georgi 1981; Walsh and Ruddick 1995; Smyth and Ruddick69
2010) to basin-scale circulations (e.g. Zhang et al. 1999; Merryfield et al. 1999).70
Effective diffusivities can be calculated directly from DNS. Thermal and saline effective71
diffusivities for 2D salt-fingering have been computed in previous studies (Merryfield and72
Grinder 2000; Stern et al. 2001; Yoshida and Nagashima 2003; Shen 1995). The effective73
diffusivities for 3D sheared, salt-fingering were first calculated by Kimura and Smyth (2007)74
3
for a single initial state.75
In this paper, we extend Kimura and Smyth (2007) to cover a range of oceanographically76
relevant initial background states. Although oceanic salt fingering does not always lead to77
staircases, that is the regime we focus on here as it is most amenable to direct simulation.78
Our objective is twofold. First, we explore the isotropy assumption for turbulent salt fingers79
and salt sheets. Second, we compute effective diffusivities to identify effects of anisotropy80
that could affect the accuracy of in situ estimates.81
In section 2, we describe our DNS model and initial conditions. An overview of salt sheet82
evolution is given in section 3. Section 4 discusses the nature of anisotropy at dissipation83
scales and its implications for interpreting profiler measurements. In section 5, we discuss84
the effective diffusivities of momentum, heat, and salt, and suggest a parameterization in85
terms of ambient property gradients. Our conclusions are summarized in section 6.86
2. Methodology87
We employ the three-dimensional incompressible Navier-Stokes equations with the Boussi-88
nesq approximation. The evolution equations for the velocity field, ~u(x, y, z, t) = {u, v, w},89
in a nonrotating, Cartesian coordinate system, {x, y, z}, are90
[
D
Dt− ν∇2
]
~u = −∇π + bk̂ + ν∇2~u
∇ · ~u = 0. (1)
D/Dt = ∂/∂t+ ~u · ~∇ and ν are the material derivative and kinematic viscosity, respectively.91
The variable π represents the reduced pressure (pressure scaled by the uniform density ρ0).92
The total buoyancy is defined as b = −g(ρ−ρ0)/ρ0, where g is the acceleration due to gravity.93
Buoyancy acts in the vertical direction, as indicated by the vertical unit vector k̂. We assume94
4
that equation of state is linear, and therefore the total buoyancy is the sum of thermal and95
saline buoyancy components (bT and bS), each governed by an advection-diffusion equation:96
b = bT + bS;
DbT
Dt= κT∇2bT ; (2)
DbS
Dt= κS∇2bS. (3)
Molecular diffusivities of heat and salt are denoted by κT and κS, respectively.97
Periodicity intervals in streamwise (x) and spanwise (y) directions are Lx and Ly. The98
variable Lz represents the vertical domain length. Upper and lower boundaries, located at99
z = −Lz/2 and z = Lz/2, are impermeable (w = 0), stress-free (∂u/∂z = ∂v/∂z = 0), and100
insulating with respect to both heat and salt (∂bT /∂z = ∂bS/∂z = 0).101
To represent mixing in the high-gradient “interface” of a thermohaline staircase, we102
initialize the model with a stratified shear layer in which shear and stratification are concen-103
trated at the center of the vertical domain with a half-layer thickness of h:104
u
∆u=
bT
∆BT
=bS
∆BS
= tanh(z
h
)
.
The constants ∆u, ∆BT , and ∆BS represent the change in streamwise velocity, thermal buoy-105
ancy, and saline buoyancy across the half-layer thickness of h. For computational economy,106
we set h = 0.3 m. This is at the low end of the range of observed layer thickness (e.g. Gregg107
and Sanford 1987; Kunze 1994). The change in the total buoyancy is ∆B = ∆BT + ∆BS.108
In all the DNS experiments, the initial buoyancy frequency (√
∆B/h) is fixed at 1.5× 10−2109
rad s−1, a value typical of the thermohaline staircase east of Barbados (Gregg and Sanford110
1987).111
These constants can be combined with the fluid parameters ν, κT , and κS to form non-112
5
dimensional parameters, which characterize the flow at t = 0:113
Ri =∆Bh
∆u2;
Rρ = −∆BT
∆BS
;
Pr =ν
κT;
τ =κS
κT.
We have done 7 experiments with different Ri and Rρ (table 1). The bulk Richardson114
number, Ri, measures the relative importance of stratification and shear. If Ri < 0.25, the115
initial flow is subjected to shear instabilities (Miles 1961; Howard 1961; Hazel 1972). Here,116
we chose high enough Ri to ensure that shear instabilities do not disrupt the growth of117
salt-fingering modes. A typical range of Ri in a sheared, thermohaline staircase is ∼ 3− 100118
(Gregg and Sanford 1987; Kunze 1994). The Reynolds number based on the half-layer119
thickness and the half change in streamwise velocity is given by Re = h∆u/ν. Since ∆B120
and h are kept constant in our simulations, Re and Ri are not independent: Re = 1354Ri−1/2121
with Re = 0 and Ri = ∞ representing the unsheared case.122
The density ratio, Rρ, quantifies the stabilizing effect of thermal to destabilizing effect123
of saline buoyancy components; salt-fingering grows more rapidly as Rρ approaches unity.124
We varied Rρ between 1.2 and 2, which covers the range of observational data available for125
comparison (St. Laurent and Schmitt 1999; Inoue et al. 2008). The Prandtl number, Pr,126
and the diffusivity ratio, τ , represent ratios of the molecular diffusivities of momentum, heat,127
and salt. The Prandtl number was set to 7, which is a typical value for salt water. The128
diffusivity ratio in the ocean is 0.01, i.e., the heat diffuses two orders of magnitude faster than129
salt. The vast difference in diffusivity requires DNS to resolve a wide range of spatial scales,130
making it computationally expensive. In previous DNS of salt water, τ has been artificially131
increased to reduce required computational expenses (e.g. Stern et al. 2001; Gargett et al.132
6
2003; Smyth et al. 2005). Kimura and Smyth (2007) conducted the first 3D simulation with133
τ = 0.01 and found that increasing τ from 0.01 to 0.04 reduced thermal and saline effective134
diffusivities by one half. In the cases presented here, we set τ to 0.04 to allow an extended135
exploration of the Ri and Rρ dependence.136
The fastest-growing salt-sheet wavelength predicted by linear stability analysis, is λfg =137
2π(νκT 2h/∆B)1/4 (Stern 1975; Schmitt 1979). Our value matches the observed value, λ =138
0.032 m (Gregg and Sanford 1987; Kunze 2003). We accommodate four wavelengths of139
the fastest-growing primary instability in the spanwise direction, Ly = 4λfg. The vertical140
domain length, Lz, was chosen such that vertically propagating plumes reach statistical141
equilibrium. We found that Lz equal to six times h was sufficient. Lx was chosen to be large142
enough to accommodate subsequent secondary instabilities. After sensitivity tests, we chose143
Lx = 28λfg.144
The primary instability was seeded by adding an initial disturbance proportional to the145
fastest-growing mode of linear theory, computed numerically as described in Smyth and146
Kimura (2007). We seed square salt-fingering for Ri = ∞ and sheets in the presence of147
shear. The vertical displacement amplitude is set to 0.02h, and a random noise was added148
to the initial velocity field with an amplitude of 1× 10−2hσL to seed secondary instabilities.149
The variable, σL indicates the growth rate of the fastest growing linear normal mode.150
The numerical code used to solve (1) - (3) is described by Winters et al. (2004). The151
code uses Fourier pseudospectral discretization in all three directions, and time integration152
using a third-order Adams-Bashforth operator. A time step is determined by a Courant-153
Friedrichs-Lewy (CFL) stability condition. The CFL number is maintained below 0.21 for154
DNS experiments presented here. The code was modified by Smyth et al. (2005) to accom-155
modate a second active scalar, which is resolved on a fine grid with spacing equal to one half156
the spacing used to resolve the other fields. The fine grid is used to resolve salinity. The157
fine grid spacing is equal to 0.15λfg
√τ in all three directions, as suggested by Stern et al.158
7
(2001).159
3. Flow overview160
Figure 1 shows the salinity buoyancy field for the case Ri = 6, Rρ = 1.6 at selected times.161
The time is scaled by the linear normal growth rate of salt sheets, σL, described by Smyth162
and Kimura (2007). Figure 1a shows the salt-sheet instability; the planar regions of vertical163
motions oriented parallel to the background shear flow. Rising and sinking fluid are shown164
in green and blue, respectively. The computational domain accommodates four salt sheets.165
When the salt sheets reach the edges of the interface (indicated by purple and red), they166
start to undulate in the spanwise direction (figure 1b). At the same time, the salt sheets167
develop streamwise dependence at the edges and center of the interface. At the edges of the168
interface, the streamwise dependence appears as ripples. At the center, tilted laminae are169
evident, reminiscent of those seen in shadowgraph images of optical microstructure (Kunze170
1987, 1990; St. Laurent and Schmitt 1999). After the salt sheets are disrupted, the flow171
evolves into turbulent salt-fingering (figure 1c).172
4. Dissipation rates and isotropy173
We will explore the geometry of scalar (thermal and saline buoyancy) and velocity gra-174
dient fields. The isotropy of scalar fields is diagnosed in terms of the thermal and saline175
buoyancy variance dissipation rates:176
χT = 2κT
⟨
|∇b′T |2⟩
; χS = 2κS
⟨
|∇b′S|2⟩
. (4)
8
Variables, b′T and b′S are thermal and saline buoyancy perturbations:177
b′T (x, y, z, t) = bT (x, y, z, t) − BT (z, t);
b′S(x, y, z, t) = bS(x, y, z, t) − BS(z, t),
where BT and BS are horizontally averaged bT and bS, respectively. The angle brackets178
indicate averaging over volume between −2h < z < 2h, where turbulent salt-fingering is179
most active.180
Turbulent kinetic energy dissipation rate, ǫ, is used to quantify the anisotropy of velocity181
fields:182
ǫ =ν
2
⟨
(
∂u′i
∂xj+
∂u′j
∂xi
)2⟩
.
The primes indicate the perturbation velocity fields:183
~u′(x, y, z, t) = ~u(x, y, z, t) − {U(z, t), 0, 0}, (5)
where U(z, t) is the streamwise velocity averaged over horizontal directions.184
In shear-driven turbulence, the degree of isotropy is predicted by the buoyancy Reynolds185
number, Reb = ǫ/νN2, where N2 = 〈BT ,z + BS,z〉. As Reb increases, the dissipation range186
separates from the energy-containing scales and become increasingly isotropic. Gargett et al.187
(1984) concluded from observations of shear-driven turbulence that ǫ can be accurately188
estimated from a single term when Reb > 2 × 102. Itsweire et al. (1993) found that the189
assumption of isotropy is invalid for Reb < 102 using DNS of an uniform stratified-shear190
layer. A similar result holds in DNS of localized stratified-shear layer (Smyth and Moum191
2000).192
In fingering-favorable stratification with turbulence, laboratory experiments showed that193
the finger structures have Reb ∼ O(10) for Rρ < 2 (McDougall and Taylor 1984). Oceanic194
9
observations of a thermohaline staircase also suggest that Reb ∼ O(10) (St. Laurent and195
Schmitt 1999; Inoue et al. 2008). Reb from our simulation increases until salt sheets start to196
break up (σLt ≈ 5) and reaches a maximum value of Reb = 10.8 at σLt = 8 (figure 2). After197
σLt > 8, the flow becomes quasi-steady with Reb ≈ 10.198
In our DNS, Reb is comparable to that observed Reb; therefore, we may use our results199
to quantify the bias inherent in observational estimates of dissipation rates. Although the200
criterion Reb ∼ 200 found in shear-driven turbulence does not apply directly to turbulent201
salt-fingering, the relatively small value found in both DNS and observations suggests that202
the dissipation scales could be significantly anisotropic.203
a. Expressions of χS and χT204
For isotropic turbulence, the saline and thermal variance dissipation rates can be ex-205
pressed in three different forms as206
χSi = 6κS
⟨
(
∂b′S∂xi
)2⟩
and (6)
χT i = 6κT
⟨
(
∂b′T∂xi
)2⟩
(7)
where i ranges from 1 to 3 and is not summed over. Each of three different forms takes207
exactly the same value for isotropic turbulence (e.g. χS = χS1 = χS2 = χS3). Despite208
Reb ∼ O(10) in salt-fingering system, observations and numerical simulations support the209
thermal buoyancy field being nearly isotropic. Two-dimensional numerical simulations of210
Shen (1995) showed that the thermal spectral variance is distributed approximately equally211
in both vertical and horizontal wavenumbers. Lueck (1987) found that the magnitude of the212
vertical thermal buoyancy gradient was similar to the horizontal gradient in a thermocline213
staircase east of Barbados. Since the measurements were taken at sites with Ri ∼ 10 and214
10
never less than unity, Lueck (1987) argued that the isotropic structure is not likely the result215
of shear-driven turbulence. Shadowgraph images of salt-fingering showed coherent tilted216
laminae (Kunze 1987). Because shadowgraph images tend to emphasize the dissipative217
scales of the salinity field, the shadowgraph images are thought to reflect anisotropic salinity218
structures (Kunze 1987, 1990; St. Laurent and Schmitt 1999).219
Signatures of salt-sheet anisotropy decrease with time in both saline and thermal buoy-220
ancy fields (figure 3a, b). In the linear regime (0 < σLt < 3), all salinity and thermal221
buoyancy variance dissipation rates come from spanwise derivatives (indicated by solid lines222
in figure 3a, b), consistent with the motion of salt sheets. With the onset of secondary223
instability, the contribution from spanwise derivative decreases, while contributions from224
streamwise and vertical derivatives increase. These ratios become quasi-steady in the tur-225
bulent regime (σLt > 8), by which time these fields are nearly isotropic. The slight (∼ 10%)226
anisotropy that remains in the turbulent state is consistent with that of the initial instability,227
with the contribution from the spanwise derivative being the largest. Even after transition,228
salt-fingering-favorable stratification is maintained, and salt-sheet instability continues to229
influence the anisotropy at small scales.230
We turn now to the effects of Ri and Rρ on anisotropy in the turbulent regime, focusing on231
the thermal buoyancy as it is easiest to measure. Time averages of the ratios χT i over σLt > 8232
succinctly represent the Ri and Rρ dependences of anisotropy in the turbulent state (figure 4).233
As in figure 3, anisotropy characteristics of the linear instability are evident in the turbulent234
regime (figure 4a). In the unsheared case Ri = ∞, the contributions from the horizontal235
derivatives are approximately equal (χT 1 ≈ χT 2), indicating that the thermal buoyancy236
gradient is statistically axisymmetric about the vertical direction. As Ri decreases (i.e., as237
background shear is increased from zero), the contribution from χT 2increases, while the238
contribution from χT 1 decreases. The contribution from χT 3 to χT increases with decreasing239
Ri, resembling the characteristics of shear-driven turbulence (Itsweire et al. 1993; Smyth and240
11
Moum 2000).241
As Rρ increases, the geometric characteristics of salt sheets become more dominant (figure242
4b), i.e., the contribution from χT 2increases with increasing Rρ, while the contribution from243
χT 1 decreases. Salt sheets become more horizontally isotropic (χT 1 ≈ χT 2) with decreasing244
Rρ (figure 4b), as observed in laboratory experiments of salt-fingering by Taylor (1992).245
b. Expressions for ǫ246
In isotropic turbulence, ǫ can be represented by any one of the nine expressions:247
ǫ =15ν
2 − δi,j
⟨
(
∂u′i
∂xj
)2⟩
, (8)
with no summation over i and j (Taylor 1935). The variable, δi,j, represents the Kronecker248
delta function. In general, these expressions are unequal, and their differences reflect the249
degree of anisotropy of the velocity gradients at dissipation scales.250
We look first at expressions involving the three derivatives of the vertical velocity com-251
ponents, as these will prove to dominate. In the unsheared case (figure 5a), the contribution252
from w′,y
2 and w′,x
2 are the same, reflecting the geometry of salt-fingering instability. In the253
sheared case (figure 5b), the contributions from w′,y
2 to ǫ exceeds these of w′,x
2 and w′,z
2,254
reflecting the geometry of salt sheet instability. As the flow becomes turbulent, all the ratios255
reach steady values between 1.5 and 3 at σLt ≈ 8 (figures 5a and 5b); however, none of the256
ratios converge to unity. Anisotropy of salt-fingering instability continues to influence the257
geometry of the dissipation scales in the turbulent state (figure 5a). In contrast, anisotropy258
of salt-sheet instability is disrupted by the increase in contribution from w′,x in the turbulent259
state (figure 5b).260
To quantify Ri and Rρ dependence, the nine approximations in (8) were averaged over261
σLt > 8 and normalized by the true value of ǫ in the turbulent regime (figure 6). As already262
12
seen (figure 5), approximations involving the vertical velocity overestimate ǫ, whereas those263
involving horizontal velocities underestimate ǫ. This is consistent with the dominance of ver-264
tical motions in turbulent salt fingers. In the absence of background shear, the contribution265
from w′,x
2 and w′,y
2 are the largest, which is consistent with the axisymmetry of salt-fingering266
about the vertical direction. In the sheared cases, the contribution from w′,y
2 is the largest,267
consistent with salt-sheet geometry, but unlike in the linear regime, the contribution from268
w′,x
2 is significant.269
In both sheared and unsheared cases, the second largest contribution comes from w′,z
2.270
This vertical divergence acts to squeeze the fluid vertically at the tips of rising and sinking271
plumes. This vertical compression is compensated by normal strains in horizontal direc-272
tions(
〈w′,z
2〉 ≈ 〈u′,x
2〉, 〈v′,y
2〉)
. Because of the difference in geometry between sheared and273
unsheared cases, the contributions from u′,x
2 and v′,y
2 to balance w′,z
2 are different. In the274
unsheared case, the vertically squeezed fluid at the edges of plumes is displaced equally in the275
streamwise and spanwise directions(
〈u′,x
2〉 ≈ 〈v′,y
2〉)
, resembling salt-fingering. In contrast,276
the geometry of the sheared case displaces more fluid in spanwise direction(
〈u′,x
2〉 < 〈v′,y
2〉)
.277
The approximations using horizontal shear strain rates also show the influence of the278
linear instabilities. The balances shown in the unsheared case, 〈u′,y
2〉 ≈ 〈v′,x
2〉, 〈u′,x
2〉 ≈ 〈v′,y
2〉,279
and 〈u′,z
2〉 ≈ 〈v′,z
2〉, indicate axisymmetry about the vertical. As Ri decreases, contributions280
from u′,z
2 and v′,z
2 increase. This indicates that the flow is developing characteristics of281
shear-driven turbulence as described by Itsweire et al. (1993) and Smyth and Moum (2000).282
c. Impact of anisotropy on ǫ and χT from vertical profilers283
Observational estimates of dissipation rates are often based on data from vertical profilers.284
These profilers measure the vertical change of horizontal velocities and temperature for which285
ǫ and χT are estimated by286
13
ǫz =15ν
4
(⟨
(
∂u′
∂z
)2⟩
+
⟨
(
∂v′
∂z
)2⟩)
; (9)
287
χzT = 6κT
⟨
(
∂b′T∂z
)2⟩
. (10)
Though Reb associated with fingering is ∼ O(10), these isotropic approximations have288
been justified by observations (Lueck 1987) and numerical simulation (Shen 1995), both289
showing nearly isotropic thermal buoyancy field in the turbulent regime. But, isotropy in290
temperature gradient does not guarantee isotropy in velocity gradients. For isotropic flows,291
these approximations are exact, i.e. ǫ = ǫz and χT = χzT . Estimates of the salinity vari-292
ance dissipation rate are relatively scarce (e.g. Nash and Moum 1998), because conductivity293
sensors need to resolve finer scales than temperature. This justification is not consistent for294
the early stage of the flow when salt sheets or salt fingers are active. The geometry of salt295
fingers and salt sheets led theoretical models to utilize the “tall fingers” (TF) approximation296
in an unbounded salt fingers and salt sheets (Stern 1975; Kunze 1987; Smyth and Kimura297
2007). Since salt fingers and salt sheets are tall and narrow, the TF approximation assumes298
that the vertical derivative is negligible relative to horizontal derivatives (ǫz = χzT = 0).299
The ǫz gives a poor estimate of ǫ, even in the turbulent regime (figures 7a and 7b). As300
the flow evolves, ǫz/ǫ increases from near zero but does not converge to unity. Instead, the301
ratio becomes quasi-steady, ranging between 0.32 and 0.52 for σLt > 8 (figure 7a and 7b). In302
contrast, the value of χzT /χT becomes quasi-steady between 0.8 and 1.2 (figures 7c and 7c).303
These results suggest that, in the presence of turbulent salt fingers, χzT is an appropriate304
approximation, but ǫz underestimates ǫ by a factor of 2 to 3.305
14
5. Turbulent fluxes and diffusivities306
Of primary interest to the oceanographic community is understanding the turbulent fluxes307
associated with turbulent salt fingers. Measurements of ǫ and χT allow indirect estimates308
of turbulent fluxes via the dissipation ratio (also called “mixing efficiency”; e.g. McEwan309
1983):310
Γ =N2χT
2ǫ⟨
BT,z
2⟩ . (11)
In turbulent mixing, mechanical energy that goes into mixing can be expended by raising311
the mass of fluid or kinetic energy dissipation by molecular viscosity. Γ approximates the312
fraction of the turbulent kinetic energy that is irreversibly converted to potential energy due313
to mixing. In shear-driven turbulence, Γ can be used to estimate the effective diffusivity of314
heat and salt as KT = KS = Γ ǫN2 (Osborn 1980).315
The effective diffusivities of heat, salt, and momentum are defined via the standard flux-316
gradient relations:317
KT = −〈w′b′T 〉⟨
∂BT
∂z
⟩ ; KS = −〈w′b′S〉⟨
∂BS
∂z
⟩ ; KU = −〈u′w′〉⟨
∂U∂z
⟩ . (12)
Relationships between the thermal and saline buoyancy and momentum fluxes can be ex-318
pressed using the heat-salt flux ratio and the Schmidt number:319
γs = −〈w′b′T 〉〈w′b′S〉
; Sc =KU
KS
.
In salt-fingering and salt-sheet instabilities, an unstable distribution of mean saline buoyancy320
drives salt and heat fluxes downward. This implies that thermal buoyancy flux is working321
against the gravity, w′b′T < 0. Schmidt number quantifies the relative importance of effective322
diffusivity of momentum to that of salt. Next, we will quantify the Ri and Rρ dependences323
of Γ, the effective diffusivities, γs, and Sc.324
15
a. Estimation of Γ325
In homogeneous, stationary isotropic turbulence,326
Γz =N2χz
T
2ǫz⟨
BT,z
2⟩ . (13)
In shear-driven turbulence, measurements indicate Γz ≈ 0.2 (Osborn 1980; Oakey 1982;327
Moum 1996). This value is accurate for Reb > 2×102 (Gargett et al. 1984). In salt-fingering328
systems, fastest-growing finger theory suggests 0.25 < Γz < 0.5, while measurements suggest329
that Γ can have higher values: 0.4 < Γz < 2 (St. Laurent and Schmitt 1999; Inoue et al.330
2008). At steady state, balances of turbulent kinetic energy and scalar variance imply331
Γ =Rρ − 1
Rρ
γs
1 − γs(14)
in the absence of shear (Hamilton et al. 1989; McDougall and Ruddick 1992). For fastest-332
growing fingers, Γ is a function of Rρ by substituting γs = (Rρ)1/2[(Rρ)
1/2 − (Rρ − 1)1/2]333
(Stern 1975; Kunze 1987). In the presence of shear, for the fastest-growing salt sheets,334
Γ =Rρ − 1
Rρ
PrRi
PrRi + 1(15)
(Smyth and Kimura 2007). In the present DNS, Γ lies below 0.6 (figure 8a). In the linear335
regime (σLt < 2), Γ is consistent with (15). Γ decays slowly as the flow becomes unstable336
to secondary instability (σLt ∼ 3) then becomes quasi-steady between 0.3 and 0.4 in the337
turbulent regime.338
Γz is generally larger than Γ (figure 8b), consistent with ǫz > ǫ. For Ri > 6, Γz can be339
30 times larger than Γ in the pre-turbulent state (0 < σLt < 3). A local peak of Γz/Γ is340
formed during the secondary instability (σLt ∼ 5). As the flow becomes turbulent, the ratio341
16
decreases and approaches a quasi-steady value ∼ 2, i.e., Γz overestimates Γ by a factor of 2342
in the turbulent regime.343
Since both Γ and Γz become quasi-steady after σLt = 8, each of Γ and Γz is averaged344
over σLt > 8 to quantify Ri and Rρ dependences in the turbulent regime. In all cases,345
Γz coincides approximately with observations (figure 9). Both observed and simulated Γz346
increase with increasing Ri. Similarly, Γ increases with increasing Ri. The Rρ dependence of347
the observed Γz and the DNS result have different trends (figure 10). Observed Γz decreases348
with increasing Rρ, while both Γz and Γ from DNS increase with increasing Rρ. The linear349
result (15) predicts Γ from DNS remarkably well (figures 9 and 10).350
Our results cast doubt on the isotropy assumption used to infer Γ from microstructure351
profiles. This has implications for recent attempts to extend the k − ǫ formalism to include352
double diffusion (Canuto et al. 2008), where observed Γ estimates were used to validate353
model parameters. Also, Inoue et al. (2008) used Γz from St. Laurent and Schmitt (1999) to354
suggest that the fingers must have larger wavelengths than the fastest-growing mode (their355
figure 5). In contrast, we find Γ consistent with linear fastest-growing fingers (figures 9 and356
10).357
b. Effective diffusivities358
Effective diffusivities are often used in large-scale models in order to represent small-scale359
physics (e.g. Bryan 1987; Gargett and Holloway 1992; Walsh and Ruddick 1995). Gargett360
and Holloway (1992) found that steady-state solutions of low-resolution general circulation361
models (GCMs) were sensitive to the ratio of KT to KS. More recent modeling studies have362
found that regional circulations are significantly altered when parameterizations of KT and363
KS due to fingering are introduced (Merryfield et al. 1999; Zhang et al. 1999).364
In our DNS, KS increases exponentially until disruption of the primary instability at365
17
σLt ≈ 5 (figure 11). After a period of slow decline, KS approaches a quasi-steady state at366
σLt ≈ 8. KS is reduced in the presence of the background shear (figure 11a). For Ri = 0.5,367
KS is smaller by a factor of 4 compared to Ri = ∞ case. This finding was unexpected,368
and has important implications for turbulence modeling, where it is often assumed that,369
since background shear represents an energy source for mixing, its introduction will increase370
effective diffusivities (e.g. Large et al. 1994; Zhurbas and Oh 2001; Smyth and Ruddick 2010).371
We also find that KS increases with decreasing Rρ (figure 11b). This Rρ dependence372
had been reported by the previous DNS of two-dimensional salt-fingering (Merryfield and373
Grinder 2000; Stern et al. 2001). The flux ratio γs remains within ∼ 10% of 0.6 (figure 12a),374
which agrees with its linear value. In the the turbulent regime, γs is about 8% higher than375
its linear value in the unsheared case, but fluctuates within 3% of its linear value in the376
sheared cases.377
Ruddick (1985) suggested that individual salt sheets rapidly lose their momentum via378
lateral diffusion and therefore Sc < 1. A laboratory experiment to confirm this hypothesis379
was inconclusive (Ruddick et al. 1989). Linear stability analysis of salt sheets (Smyth and380
Kimura 2007), and subsequent DNS (Kimura and Smyth 2007) for a single initial case381
confirmed that Sc < 1 for Ri = 2. Sc < 1 holds in Ri ranging from 0.5 to 20 (figure 12b).382
Effective diffusivity of the momentum is an order of magnitude smaller than that of salt,383
confirming that salt sheets are inefficient in transporting momentum over the observed range384
of Ri and Rρ.385
In two-dimensional unsheared, turbulent salt fingers, Stern et al. (2001) showed that386
decrease in τ from 0.04 to 0.01 increased the heat flux by 15%. In 3D sheared, double-387
diffusive turbulence, Kimura and Smyth (2007) found that the decrease in τ from 0.04 to388
0.01 increases the KT and KS by a factor of 2. The effective diffusivities of heat and salt for389
the oceanic water may therefore be 2 times larger than our DNS results presented here.390
Both KT and KS decrease with increasing Rρ (figures 13 b,d), consistent with DNS results391
18
from Merryfield and Grinder (2000), despite the difference in Ri, τ , and computational392
domain dimensions. Estimates for three-dimensional salt-fingering by Stern et al. (2001)393
are twice as large; this could be due the presence of the shear and difference in τ in our394
simulations.395
Parameterizations of these diffusivities are essential in modeling large-scale flows where396
double-diffusive turbulence controls small-scale fluxes of heat and salt, such as thermohaline397
interleaving. Walsh and Ruddick (2000) employed a parameterization of KS for pure salt-398
fingering: KS = KS0R−nρ . Smyth (2008) fitted this simple model to DNS results from Stern399
et al. (2001) and obtained n = 2 and KS0 = 10−4. To add dependence on Ri, we compute a400
least squares fit of KT and KS, averaged over σLt > 8401
KT (Rρ, Ri) = 3.07 × 10−5R−4.0ρ Ri0.17 (16)
KS(Rρ, Ri) = 4.38 × 10−5R−2.7ρ Ri0.17. (17)
This Ri dependence differs significantly from the prediction of Kunze (1994), KS ∝ Ri1.0,402
but is closer to that of Inoue et al. (2008), KS ∝ Ri0.403
c. Observational estimates of effective diffusivities404
In interpretation of observational data, the Osborn and Cox (1972) diffusivity model in405
conjunction with the isotropy assumption gives406
Kχz
T =χz
T
2⟨
(
∂BT
∂z
)2⟩ ; Kχz
S =χz
S
2⟨
(
∂BS
∂z
)2⟩ , (18)
based on property variance dissipations, where χzS = 6κS
⟨
(
∂b′S∂z
)2⟩
. These estimates are407
exact for stationary, homogeneous, isotropic turbulence. Here, we compare these estimates408
19
with the true diffusivities KT and KS, computed directly using fluxes (12).409
The isotropic Osborn-Cox model (18) captures the Rρ dependence of KT and KS (figure410
14a and 14b). Both Kχz
T and Kχz
S decrease with increasing Rρ, consistent with the trends411
of KT and KS from DNS and observations by St. Laurent and Schmitt (1999). However,412
both Kχz
T and Kχz
S are smaller than KT and KS for the range of Rρ discussed here. The413
Osborn-Cox diffusivity model (1972) can therefore underestimate KT and KS by up to a414
factor of 3 for salt fingers at steady state.415
6. Conclusions416
We have simulated sheared, double-diffusive turbulence in a thermohaline staircase, using417
DNS. The following caveats should be noted:418
i. the ratio of molecular diffusivity of salt to heat is 4 times larger than the real ocean;419
ii. the interface thickness is at least 3 times smaller than observed thickness of thermoha-420
line staircases. In much thicker layers and continuous stratification, salt-finger flux laws421
change because the fingers no longer extend across the interface from one homogeneous422
layer to the other (Kunze 1987);423
iii. the equation of state is linear.424
Our main findings are:425
i. Velocity gradients are not isotropic, but scalar gradients are nearly isotropic. The426
approximations of ǫ based on vertical shears, which are often used in interpretations427
of microstructure data, underestimate its value by a factor of 2 to 3. This suggests428
that the rate of dissipation by sheared, double-diffusive turbulence can be 2 to 3 times429
larger than measured inferences.430
20
ii. The geometry of the primary instability is often (though not always) reflected in the431
anisotropy of the turbulent flow.432
iii. The isotropy assumption can lead to overestimation of the dissipation ratio Γ by a433
factor of 2 to 3. This will impact estimates of correlation time scales in the k−ǫ model434
of double-diffusive turbulence (e.g. Canuto et al. 2008). It also caused Inoue et al.435
(2008) to incorrectly ascribe the signals in St. Laurent and Schmitt (1999) to fingers436
with wavelengths larger than those of the fastest-growing mode.437
iv. A decrease in Ri or an increase in Rρ reduces the effective diffusivities of heat and438
salt. Empirical fits give KT (Rρ, Ri) = 3.07×10−5R−4.0ρ Ri0.17 and KS(Rρ, Ri) = 4.38×439
10−5R−2.7ρ Ri0.17. Kimura and Smyth (2007) showed that decrease in τ from 0.04 to 0.01440
increase KT and KS by a factor of 2. The dependence on Ri differs significantly from441
the theoretical prediction of Kunze (1994).442
v. Effective diffusivities estimated by the Osborn-Cox model (1972) produced consistent443
Rρ dependence but are up to a factor of three too small.444
vi. Values of nondimensional mixing efficiency Γ, flux ratio γ, and Schmidt number Sc in445
the turbulent regime are predicted by linear theory with remarkable accuracy.446
Acknowledgments.447
Computer time was provided by the National Center for Atmospheric Research (NCAR).448
The graphics are produced using VAPOR with help from VisLab at NCAR. The work was449
supported by the National Science Foundation under Grant No. 0453140.450
21
451
REFERENCES452
Bryan, F., 1987: On the parameter sensitivity of primitive equation ocean general circulation.453
J. Phys. Oceanogr., 17, 970–985.454
Canuto, V., Y. Cheng, and A. Howard, 2008: A new model for double diffusion + turbulence455
using a Reynolds stress model. Geophys. Res. Lett., 35, L02 613.456
Gargett, A., T. Osborn, and P. Nasmyth, 1984: Local isotropy and the decay of turbulence457
in a stratified fluid. J. Fluid Mech., 144, 231–280.458
Gargett, A. E. and G. Holloway, 1992: Sensitivity of the GFDL ocean model to different459
diffusivities of heat and salt. J. Phys. Oceanogr., 22, 1158–1177.460
Gargett, A. E., W. Merryfield, and G. Holloway, 2003: Direct numerical simulation of differ-461
ential scalar diffusion in three-dimensional stratified turbulence. J. Phys. Oceanogr., 33,462
1758–1782.463
Gregg, M. and T. Sanford, 1987: Shear and turbulence in thermohaline staircases. Deep Sea464
Res., 34, 1689–1696.465
Hamilton, J., M. Lewis, and B. Ruddick, 1989: Vertical fluxes of nitrate associated with salt466
fingers in the world’s oceans. J. Geophys. Res., 94, 2137–2145.467
Hazel, P., 1972: Numerical studies of the stability of inviscid parallel shear flows. J. Fluid468
Mech., 51, 39–62.469
Howard, L., 1961: Note on a paper of John W. Miles. J. J. Fluid Mech., 10, 509–512.470
22
Inoue, R., E. Kunze, L. S. Laurent, R. Schmitt, and J. Toole, 2008: Evaluating salt-fingering471
theories. J. Mar. Res., 66, 413–440.472
Itsweire, E., J. Kosefe, D. Briggs, and J. Ferziger, 1993: Turbulence in stratified shear flows:473
implications for interpreting shear-induced mixing in the ocean. J. Phys. Oceanogr., 23,474
1508–1522.475
Kimura, S. and W. Smyth, 2007: Direct numerical simulation of salt sheets and turbulence476
in a double-diffusive shear layer. Geophys. Res. Lett., 34, L21 610.477
Kolmogorov, A., 1941: The local structure of turbulence in incompressible viscous fluid for478
very large reynolds number. Dokl. Akad. Nauk SSSR, 30 (4), 301–305.479
Kunze, E., 1987: Limits on growing finite-length salt fingers: A Richardson number con-480
straint. J. Mar. Res., 45, 533–556.481
Kunze, E., 1990: The evolution of salt fingers in inertial wave shear. J Mar. Res., 48,482
471–504.483
Kunze, E., 1994: A proposed flux constraint for salt fingers in shear. J Mar. Res., 52,484
999–1016.485
Kunze, E., 2003: A review of oceanic salt fingering theory. Prog. Oceanogr., 56, 399–417.486
Lambert, R. B. and W. Sturges, 1977: A thermohaline staircase and vertical mixing in the487
thermocline. Deep-Sea Res., 24, 211–222.488
Large, W. G., J. C. McWilliams, and S. C. Doney, 1994: Oceanic vertical mixing: A review489
and a model with a nonlocal boundary layer parameterization. Rev. Geophys., 32, 363–403.490
Linden, P., 1974: Salt fingers in a steady shear flow. Geophys. Fluid Dyn., 6, 1–27.491
23
Lueck, R., 1987: Microstructure measurements in a thermohaline staircase. Deep-Sea Res.,492
34, 1677–1688.493
McDougall, T. J. and B. R. Ruddick, 1992: The use of ocean microstructure to quantify494
both turbulent mixing and salt fingering. Deep-Sea Res., 39, 1931–1952.495
McDougall, T. J. and J. R. Taylor, 1984: Flux measurements across a finger interface at low496
values of the stability ratio. J. Mar. Res., 42, 1–14.497
McEwan, A., 1983: Internal mixing in stratified fluids. J. Fluid Mech., 128, 59–80.498
Merryfield, W. J. and M. Grinder, 2000: Salt fingering fluxes from numerical simulations.499
unpublished.500
Merryfield, W. J., G. Holloway, and A. E. Gargett, 1999: A global ocean model with double-501
diffusive mixing. J. Phys. Oceanogr., 29, 1124–1142.502
Miles, J., 1961: On the stability of heterogeneous shear ows. J. Fluid Mech., 10, 496–508.503
Moum, J. N., 1996: Efficiency of mixing in the main thermocline. J. Geophys. Res., 11 (C5),504
12 057–12 069.505
Nash, J. D. and J. N. Moum, 1998: Estimating salinity variance dissipation rate from con-506
ductivity microstructure measurements. J. Atmos. Oceanic Technol., 16, 263–274.507
Oakey, N. S., 1982: Determination of the rate of dissipation of turbulent energy from simul-508
taneous temperature and velocity shear microstructure measurements. J. Phys. Oceanogr.,509
12, 256–271.510
Osborn, T. R., 1980: Estimates of the local rate of vertical diffusion from dissipation mea-511
surements. J. Phys. Oceanogr., 10, 83–89.512
24
Proctor, M. R. E. and J. Y. Holyer, 1986: Planform selection in salt fingers. J. Fluid Mech.,513
168, 241–253.514
Ruddick, B. R., 1985: Momentum transport in thermohaline staircases. J. Geophys. Res.,515
90, 895–902.516
Ruddick, B. R., R. W. Griffiths, and G. Symonds, 1989: Frictional stress at a sheared517
double-diffusive interface. J. Geophys. Res., 94, 18 161–18 173.518
Schmitt, R. W., 1979: The growth rate of super-critical salt fingers. Deep-Sea Res., 26A,519
23–40.520
Schmitt, R. W., 1990: On the density ratio balance in the central water. jpo, 20, 900–906.521
Schmitt, R. W., 1994a: Double diffusion in oceanography. Ann. Rev Fluid Mech., 26, 255–522
285.523
Schmitt, R. W., 1994b: Triangular and asymmetric salt fingers. J. Phys. Oceanogr, 24,524
855–860.525
Shen, C., 1995: Equilibrium salt-fingering convection. Phys. Fluids, 7, 706–717.526
Smyth, W. D., 2008: Instabilities of a baroclinic, double diffusive frontal zone. J. Phys.527
Oceanogr., 38 (4), 840–861.528
Smyth, W. D. and S. Kimura, 2007: Instability and diapycnal momentum transport in a529
double-diffusive stratified shear layer. J. Phys. Oceanogr., 37, 1551–1565.530
Smyth, W. D. and J. N. Moum, 2000: Anisotropy of turbulence in stably stratified mixing531
layers. Phys. Fluids, 12, 1327–1342.532
Smyth, W. D., J. Nash, and J. Moum, 2005: Differential diffusion in breaking Kelvin-533
Helmholtz billows. J. Phys. Oceanogr., 35, 1044–1022.534
25
Smyth, W. D. and B. Ruddick, 2010: Effects of ambient turbulence on interleaving at a535
baroclinic front. J. Phys. Oceanogr., 40, 685–712.536
St. Laurent, L. and R. Schmitt, 1999: The contribution of salt fingers to vertical mixing in537
the North Atlantic tracer-release experiment. J. Phys. Oceanogr., 29, 1404–1424.538
Stern, M., 1960: The ‘salt fountain’ and thermohaline convection. Tellus, 12, 172–175.539
Stern, M., 1975: Ocean Circulation. Physics. Academic Press, 246 pp.540
Stern, M., T. Radko, and J. Simeonov, 2001: Salt fingers in an unbounded thermocline. J.541
Mar. Res., 59, 355–390.542
Tait, R. I. and M. R. Howe, 1968: Some observations of thermohaline stratification in the543
deep ocean. Deep-Sea Res., 15, 275–280.544
Taylor, G. I., 1935: Statistical theory of turbulence. ii. Proceedings of the Royal Society of545
London. Series A, Mathematical and Physical Sciences, 151, No. 873, 444–454.546
Taylor, J. R., 1992: Anisotropy of salt fingers. J. Phys. Oceanogr., 23, 554–565.547
Toole, J. and D. Georgi, 1981: On the dynamics of double diffusively driven intrusions. Prog.548
Oceanogr., 10, 123–145.549
Walsh, D. and B. Ruddick, 1995: Double diffusive interleaving: the effect of nonconstant550
diffusivities. J. Phys. Oceanogr., 25, 348–358.551
Walsh, D. and B. Ruddick, 2000: Double-diffusive interleaving in the presence of turbu-552
lencethe effect of a nonconstant flux ratio. J. Phys. Oceanogr., 30, 2231–2245.553
Winters, K., J. MacKinnon, and B. Mills, 2004: A spectral model for process studies of554
rotating, density-stratifed flows. J. Atmos. Oceanic Technol., 21, (1)69–94.555
26
Yoshida, J. and H. Nagashima, 2003: Numerical experiments on salt-finger convection. Prog.556
Oceanogr., 56, 435–459.557
You, Y., 2002: A global ocean climatological atlas of the Turner angle: implications for558
double diffusion and watermass structure. Deep Sea Res., 49, 2075–2093.559
Zhang, J., R. Schmitt, and R. Huang, 1999: Sensitivity of the GFDL modular ocean model560
to parameterization of double-diffusive processes. J. Phys. Oceanogr, 28, 589–605.561
Zhurbas, V. and I. S. Oh, 2001: Can turbulence suppress double-diffusively driven interleav-562
ing completely? J. Phys. Oceanogr., 31, 2251–2254.563
27
List of Tables564
1 Relevant parameters used in our DNS experiments. The wave number of the565
fastest growing salt-fingering instability is determined by the magnitude of566
wave number, k2 + l2, where k and l represent the streamwise and spanwise567
wave numbers. In the case of salt sheets (all cases except DNS5), there is568
not streamwise dependence (k = 0), where the salt-fingering case (DNS5) has569
k = l. In our DNS experiments, k2 + l2 is kept constant. 29570
28
DNS1 DNS2 DNS3 DNS4 DNS5 DNS6 DNS7Ri 0.5 2 6 20 ∞ 6 6Rρ 1.6 1.6 1.6 1.6 1.6 1.2 2.0Pr 7 7 7 7 7 7 7Re 1914 957 553 302 0 553 553τ 0.04 0.04 0.04 0.04 0.04 0.04 0.04λfg[m] 0.031 0.031 0.031 0.031 0.044 0.031 0.032Lx[m] 0.9 0.9 0.9 0.9 1.2 0.9 0.9Ly[m] 0.12 0.12 0.12 0.12 0.18 0.12 0.12Lz [m] 1.8 1.8 1.8 1.8 1.8 1.8 1.8nx 1024 1024 1024 1024 1024 1024 1024ny 144 144 144 144 144 144 144nz 2048 2048 2048 2048 1538 2048 2048
Table 1. Relevant parameters used in our DNS experiments. The wave number of thefastest growing salt-fingering instability is determined by the magnitude of wave number,k2 + l2, where k and l represent the streamwise and spanwise wave numbers. In the case ofsalt sheets (all cases except DNS5), there is not streamwise dependence (k = 0), where thesalt-fingering case (DNS5) has k = l. In our DNS experiments, k2 + l2 is kept constant.
29
List of Figures571
1 Evolution of salinity buoyancy field for Ri = 6, Rρ = 1.6 in an interface572
sandwiched between two homogeneous layers with respect to the scaled time573
σLt. The interface occupies one third of the domain height. Homogenous574
regions above and below the interface are rendered transparent. Inside the575
interface, the lowest (−7.15 × 10−5m2s−1) and highest (7.15 × 10−5m2s−1)576
salinity buoyancy are indicated by purple and red, respectively. 33577
2 Evolution of volumed-averaged Reb for Ri = 6, Rρ = 1.6. 34578
3 (a) Saline (a) and thermal (b) buoyancy variance dissipation rates computed579
from individual derivatives and represented as fractions of the true value for580
the case, Ri = 6, Rρ = 1.6. Thin solid horizontal lines in (a) and (b) indicate581
unity (the ratio for isotropic turbulence). 35582
4 Approximate thermal variance dissipation rate from derivatives of squared583
perturbations as a fraction of its true values for different Ri and Rρ. Each of584
the three different forms in (6) and (7) is normalized by its true value, χS or585
χT , to quantify the degree of anisotropy. Each ratio is averaged for σLt > 8586
to represent the geometry in the turbulent state. Ratios below unity are blue,587
above unity red, and equal unity black. 36588
5 Evolution of7.5ν〈w′
,x2〉
ǫ,
7.5ν〈w′
,y2〉
ǫ, and
15ν〈w′
,z2〉
ǫfor (a) unsheared case (Ri =589
∞, Rρ = 1.6) and (b) sheared case (Ri = 6, Rρ = 1.6) . Thin solid horizontal590
lines in (a) and (b) indicate the ratio for isotropic turbulence. 37591
30
6 Approximations of ǫ from each of the squared perturbation velocity derivatives592
as a fraction of its value, ǫ for different Ri and Rρ. Each ratio is averaged for593
σLt > 8 to represent the geometry in the turbulent state. Ratios below unity594
are blue, above unity red, and equal unity black. 38595
7 Approximations of ǫ as a fraction of its true value with respect to σLt for (a)596
different Richardson number Ri and (b) different density ratio Rρ. Approxi-597
mations of χT as a fraction of its true value with respect to σLt for (c) different598
Richardson number Ri and (d) different density ratio Rρ. A solid horizontal599
line on each panel indicates the ratio for isotropic turbulence. 39600
8 Evolution of (a) Γ and (b) Γz normalized by its true value Γ for different Ri.601
These ratios are unity for isotropic turbulence as indicated by a thin solid line. 40602
9 Γ and Γz for different Ri compared to observations from Inoue et al (2008).603
Vertical bars denote 95% confidence limits (Inoue et al. 2008). Mean Rρ604
is nearly constant around 1.65 in Inoue et al. (2008). Smyth and Kimura605
(2007) calculated Γ using linear stability analysis. Here we showed their Γ for606
Rρ = 1.6. 41607
10 Γ and Γz for different Rρ compared to observations from Inoue et al. (2008).608
Vertical bars denote 95% confidence limits (Inoue et al. 2008). Mean Ri609
ranges between 0.4 and 10 for Inoue et al. (2008). Smyth and Kimura (2007)610
calculated Γ using linear stability analysis. Here we plot their Γ for Ri = 6. 42611
31
11 Effective diffusivity of salt, KS with respect to scaled time for (a) different Ri612
keeping Rρ = 1.6 and (b) different Rρ keeping Ri = 6. 43613
12 Evolution of (a) flux ratio, γs, and Schmidt number, Sc, with respect to scaled614
time for different Ri with keeping Rρ = 1.6. 44615
13 Effective diffusivity of heat with respect to Ri (a) and Rρ (b), and of salt with616
respect to Ri (c) and Rρ (d). Circles in (a) and (b) indicate diffusivities from617
three-dimensional DNS with molecular diffusivity ratio τ = 0.01 (Kimura and618
Smyth 2007); DNS results presented here have τ = 0.04. Downward triangles619
in (b) and (d) indicate two- dimensional DNS results with Ri = ∞ and620
τ = 0.01 (Merryfield and Grinder 2000). Squares in (b) and (d) are effective621
diffusivities for Ri = ∞ (Stern et al. 2001) estimated from the ratio of 2- to 3-622
D fluxes using accessible values of τ , then multiplying the ratio onto directly623
computed 2-D fluxes with τ = 0.01. 45624
14 Comparisons of effective diffusivities of heat (a) and salt (b) with respect625
to density ratio Rρ. Because of small scale structures pertained in salinity,626
estimation of χS from observations is difficult. Thus, KS in (b) is estimated627
as KobsS = Rρ
γsKχz
T in the interpretations of observations. 46628
32
(a) σLt = 3.1 (b) σLt = 5 (c) σLt = 8.7
Lx = 0.9m
Ly = 0.12m
Lz = 1.8m
Fig. 1. Evolution of salinity buoyancy field for Ri = 6, Rρ = 1.6 in an interface sandwichedbetween two homogeneous layers with respect to the scaled time σLt. The interface occupiesone third of the domain height. Homogenous regions above and below the interface arerendered transparent. Inside the interface, the lowest (−7.15 × 10−5m2s−1) and highest(7.15 × 10−5m2s−1) salinity buoyancy are indicated by purple and red, respectively.
33
0 1 2 3 4 5 6 7 8 9 10 110
2
4
6
8
10
12
σLt
Re b
Fig. 2. Evolution of volumed-averaged Reb for Ri = 6, Rρ = 1.6.
34
0
0.5
1
1.5
2
2.5
3
3.5
0 1 2 3 4 5 6 7 8 9 100
0.5
1
1.5
2
2.5
3
3.5
χS1/χS
χS2/χS
χS3/χS
χT 1/χT
χT 2/χT
χT 3/χT
(a)
(b)
σLt
Fig. 3. (a) Saline (a) and thermal (b) buoyancy variance dissipation rates computed fromindividual derivatives and represented as fractions of the true value for the case, Ri = 6,Rρ = 1.6. Thin solid horizontal lines in (a) and (b) indicate unity (the ratio for isotropicturbulence).
35
0.54 0.77 0.92 1 1.1
0.96
0.78
1.2 1.2 1.2 1.2 1.1
1
1.3
1.3 1.1 0.9 0.81 0.82
1
0.88
1.2
1.6
2
0.50.50.5 222 666 202020 ∞∞∞
Rρ
RiRiRi
(a)6κT 〈b′
T2,x〉
χT(b)
6κT 〈b′T2,y〉
χT(c)
6κT 〈b′T2,z〉
χT
Fig. 4. Approximate thermal variance dissipation rate from derivatives of squared pertur-bations as a fraction of its true values for different Ri and Rρ. Each of the three differentforms in (6) and (7) is normalized by its true value, χS or χT , to quantify the degree ofanisotropy. Each ratio is averaged for σLt > 8 to represent the geometry in the turbulentstate. Ratios below unity are blue, above unity red, and equal unity black.
36
10−2
10−1
100
101
0 1 2 3 4 5 6 7 8 9 10 1110
−2
10−1
100
101
7.5ν〈w′
,x2〉
ǫ
7.5ν〈w′
,y2〉
ǫ
15ν〈w′
,z2〉
ǫ
(a)
(b)
σLt
Fig. 5. Evolution of7.5ν〈w′
,x2〉
ǫ,
7.5ν〈w′
,y2〉
ǫ, and
15ν〈w′
,z2〉
ǫfor (a) unsheared case (Ri = ∞, Rρ =
1.6) and (b) sheared case (Ri = 6, Rρ = 1.6) . Thin solid horizontal lines in (a) and (b)indicate the ratio for isotropic turbulence.
37
0.54 0.62 0.62 0.61 0.61
0.67
0.6
0.5 0.24 0.16 0.14 0.18
0.23
0.15
0.48 0.41 0.35 0.33 0.34
0.41
0.36
0.2 0.18 0.15 0.14 0.17
0.21
0.14
1 0.84 0.7 0.63 0.62
0.69
0.77
0.56 0.43 0.35 0.32 0.34
0.41
0.36
1.4 1.8 2.2 2.4 2.5
2.3
1.9
2.7 2.7 2.8 2.7 2.5
2.4
3
1.7 1.7 1.6 1.6 1.5
1.6
1.7
1.2
1.2
1.2
1.6
1.6
1.6
2
2
2
0.50.50.5 222 666 202020 ∞∞∞
Rρ
Rρ
Rρ
RiRiRi
(a)15ν〈u′
,x2〉
ǫ (b)7.5ν〈u′
,y2〉
ǫ(c)
7.5ν〈u′
,z2〉
ǫ
(d)7.5ν〈v′,x
2〉
ǫ (e)15ν〈v′,y
2〉
ǫ(f)
7.5ν〈v′,z2〉
ǫ
(g)7.5ν〈w′
,x2〉
ǫ (h)7.5ν〈w′
,y2〉
ǫ(i)
15ν〈w′
,z2〉
ǫ
Fig. 6. Approximations of ǫ from each of the squared perturbation velocity derivatives as afraction of its value, ǫ for different Ri and Rρ. Each ratio is averaged for σLt > 8 to representthe geometry in the turbulent state. Ratios below unity are blue, above unity red, and equalunity black.
38
10−3
10−2
10−1
100
101
0 2 4 6 8 1010
−3
10−2
10−1
100
101
0 2 4 6 8 10
(a) ǫz
ǫ for different Ri
Ri = 0.5, Rρ = 1.6Ri = 20, Rρ = 1.6
Ri = 0.5, Rρ = 1.6Ri = 20, Rρ = 1.6
(b) ǫz
ǫ for different Rρ(b) ǫz
ǫ for different Rρ
Ri = 6, Rρ = 1.2Ri = 6, Rρ = 2
Ri = 6, Rρ = 1.2Ri = 6, Rρ = 2
(c)χz
T
χTfor different Ri(c)
χzT
χTfor different Ri
σLt
(d)χz
T
χTfor different Rρ
σLt
Fig. 7. Approximations of ǫ as a fraction of its true value with respect to σLt for (a)different Richardson number Ri and (b) different density ratio Rρ. Approximations of χT asa fraction of its true value with respect to σLt for (c) different Richardson number Ri and(d) different density ratio Rρ. A solid horizontal line on each panel indicates the ratio forisotropic turbulence.
39
0
0.1
0.2
0.3
0.4
0.5
0.6
0 1 2 3 4 5 6 7 8 9 10 110
2
4
6
8
Γ, Ri = 0.5, Rρ = 1.6
Γ, Ri = 6, Rρ = 1.6
Γ, Ri = ∞, Rρ = 1.6
Γz/Γ, Ri = 0.5, Rρ = 1.6
Γz/Γ, Ri = 6, Rρ = 1.6
Γz/Γ, Ri = ∞, Rρ = 1.6
ΓΓ
z/Γ
σLt
(a)
(b)
Fig. 8. Evolution of (a) Γ and (b) Γz normalized by its true value Γ for different Ri. Theseratios are unity for isotropic turbulence as indicated by a thin solid line.
40
100
101
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Γ from DNS Rρ = 1.6Γz from DNS Rρ = 1.6Γz from Inoue et al. (2008)Γ from Smyth and Kimura (2007)
Γ
Ri
Fig. 9. Γ and Γz for different Ri compared to observations from Inoue et al (2008). Verticalbars denote 95% confidence limits (Inoue et al. 2008). Mean Rρ is nearly constant around1.65 in Inoue et al. (2008). Smyth and Kimura (2007) calculated Γ using linear stabilityanalysis. Here we showed their Γ for Rρ = 1.6.
41
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.10
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Γ from DNS Ri = 6Γz from DNS Ri = 6Γz from Inoue et al. (2008)Γ from Smyth and Kimura (2007)
Γ
Rρ
Fig. 10. Γ and Γz for different Rρ compared to observations from Inoue et al. (2008).Vertical bars denote 95% confidence limits (Inoue et al. 2008). Mean Ri ranges between 0.4and 10 for Inoue et al. (2008). Smyth and Kimura (2007) calculated Γ using linear stabilityanalysis. Here we plot their Γ for Ri = 6.
42
10−7
10−6
10−5
10−4
0 1 2 3 4 5 6 7 8 9 10 1110
−7
10−6
10−5
10−4
Ri = 0.5, Rρ = 1.6
Ri = 20, Rρ = 1.6
Ri = ∞, Rρ = 1.6
Ri = 6, Rρ = 1.2
Ri = 6, Rρ = 1.6
Ri = 6, Rρ = 2
σLt
KS
[m
2s−
1]
KS
[m
2s−
1]
(b)
(a)
Fig. 11. Effective diffusivity of salt, KS with respect to scaled time for (a) different Rikeeping Rρ = 1.6 and (b) different Rρ keeping Ri = 6.
43
0.5
0.55
0.6
0.65
0.7
0 1 2 3 4 5 6 7 8 9 10 110
0.05
0.1
Ri = 0.5, Rρ = 1.6Ri = 20, Rρ = 1.6Ri = ∞, Rρ = 1.6
Ri = 0.5, Rρ = 1.6Ri = 2, Rρ = 1.6Ri = 20, Rρ = 1.6
σLt
γs
Sc
(b)
(a)
Fig. 12. Evolution of (a) flux ratio, γs, and Schmidt number, Sc, with respect to scaledtime for different Ri with keeping Rρ = 1.6.
44
10−6
10−5
100
101
10−6
10−5
1.2 1.4 1.6 1.8 2
KT[m
2s−
1]
(a)
KT (Rρ = 1.6, Ri)
DNS
Kimura and Smyth (2007), τ = 0.01
(b)
KT (Rρ, Ri = 6)
DNS
DNS, Ri = ∞
Merryfield and Grinder (2000), τ = 0.01
Estimate of Stern et al. (2001), τ = 0.01
KS[m
2s−
1]
(c)
KS(Rρ = 1.6, Ri)
DNS
Kimura and Smyth (2007), τ = 0.01
(d)
KS(Rρ, Ri = 6)
DNS
DNS, Ri = ∞
Merryfield and Grinder (2000), τ = 0.01
Estimate of Stern et al. (2001),τ = 0.01
Ri Rρ
Fig. 13. Effective diffusivity of heat with respect to Ri (a) and Rρ (b), and of salt with re-spect to Ri (c) and Rρ (d). Circles in (a) and (b) indicate diffusivities from three-dimensionalDNS with molecular diffusivity ratio τ = 0.01 (Kimura and Smyth 2007); DNS results pre-sented here have τ = 0.04. Downward triangles in (b) and (d) indicate two- dimensionalDNS results with Ri = ∞ and τ = 0.01 (Merryfield and Grinder 2000). Squares in (b) and(d) are effective diffusivities for Ri = ∞ (Stern et al. 2001) estimated from the ratio of 2- to3- D fluxes using accessible values of τ , then multiplying the ratio onto directly computed2-D fluxes with τ = 0.01.
45
10−6
10−5
10−4
1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.110
−6
10−5
10−4
(a)
(b)
Rρ
KT[m
2s−
1]
KS[m
2s−
1]
KT , Ri = 6
Kχz
T , Ri = 6
Kχz
T from St. Laurent and Schmitt (1999)
KS , Ri = 6
Kχz
S , Ri = 6
KobsS from St. Laurent and Schmitt (1999)
Fig. 14. Comparisons of effective diffusivities of heat (a) and salt (b) with respect todensity ratio Rρ. Because of small scale structures pertained in salinity, estimation of χS from
observations is difficult. Thus, KS in (b) is estimated as KobsS = Rρ
γsKχz
T in the interpretationsof observations.
46
Top Related