## 1. Introduction

Experiments in the laboratory (Gibson and Schwarz 1963; Sreenivasan 1996) and in naturally occurring flows (Dillon and Caldwell 1980; Nash and Moum 1999) indicate that there is something akin to a universal form in scalar gradient spectra at high wavenumbers, although the exact form may be open to question. Batchelor (1959) and Kraichnan (1968) have offered versions that are equally plausible, while subtleties remain in defining the precise values of spectral constants and in the degree of isotropy required for the maintenance of a universal spectral form (Gargett 1985; Smyth 1999, e.g.). Despite these issues, dynamically consistent estimates of the temperature variance dissipation rate *χ _{T}* appear to be derived from scaling spectral subranges (Luketina and Imberger 2001; Nash and Moum 2002).

Recently, Moum and Nash (2009) demonstrated a novel manner of obtaining estimates of *χ _{T}* from well-resolved temperature gradient measurements on an equatorial mooring. These were obtained by scaling temperature gradient spectra in the high-wavenumber subranges: the viscous-convective and viscous-diffusive subranges shown in Fig. 1. From an analysis of temperature gradient spectra obtained from a towed body (Marlin) at lower wavenumbers, Klymak and Moum (2007) have demonstrated the feasibility of obtaining reliable estimates of turbulence quantities in the inertial-convective subrange. In this analysis, estimates of the turbulent kinetic energy dissipation rate

*ε*were compared to more direct measurements using shear probes. The towed body measurements offer a unique perspective of the turbulence in the sense that they sample the turbulence horizontally, which is the dimension we expect to have the greatest physical dimension in a vertically stratified fluid. In contrast, vertically sampled measurements as obtained from more commonly used vertical profilers sample the small dimension of the turbulence. In the vertical, it is difficult to obtain records that include the inertial-convective subrange. However, the towed records of Klymak and Moum (2007) appear to demonstrate inertial-convective subranges to surprisingly large horizontal wavenumbers (>100 m). This suggests that we may be able to use inertial-convective subrange scaling to estimate turbulence quantities from point measurements of turbulence as well, at least in the case where the fluid is rapidly advected past the point. The equatorial

*χ*pod measurements described by Moum and Nash (2009) meet this requirement.

The motivation for testing the computation of *χ _{T}* by scaling in the inertial-convective subrange is to determine how well estimates can be made using sensors with a reduced frequency range and sampling rate, in turn requiring smaller data storage capacity and potentially smaller battery capacity than we have used to date for our

*χ*pods. If the inertial-convective estimates are of sufficient quality, then we might expect that a broader range of sensors and instrumentation may be used for making direct turbulence measurements, at least in moored configurations.

The initial evaluation of temperature gradient spectra from *χ*pods that included a motion package (three linear accelerometers) indicated that, while vortex-induced vibrations caused by the interaction of the mean current with the mooring cable had no detectable effect on measured temperature gradient spectra, wave-induced motions of the cable coupled to a surface float were significant. These motions resulted in ±1-m cable displacements at ±0.5 m s^{−1} cable velocities in the surface gravity wave band (roughly 2–20-s periods). The consequent contamination of the temperature gradient spectra in the surface wave band was shown to be due to the vertical motion of the sensor through the vertical temperature gradient [quantified as *T _{t}* =

*w*, where

_{c}T_{z}*T*is the measured temperature gradient at a point (this is measured as an analog time derivative),

_{t}*w*is the vertical velocity at that point on the cable, and

_{c}*T*the local vertical temperature gradient]. Fortunately, viscous-convective and viscous-diffusive subranges were unaffected and reasonable estimates of

_{z}*χ*were obtained. However, the surface wave–induced signal directly affects the inertial-convective subrange. Hence, this contamination must be removed prior to spectral scaling.

_{T}In this paper, we first demonstrate that the contaminating wave signal can be removed spectrally through the computation of the coherence spectrum between measured and surface wave–induced signals that are observed directly and independently. To evaluate the combined effectiveness of the removal scheme and the viability of the inertial-convective subrange estimate, we compare *χ _{T}* estimates made from high-wavenumber subranges to those estimates from the inertial-convective subrange. We then incrementally degrade the measurement, first by resampling the temperature gradient and accelerations at 2 Hz (just sufficiently enough to resolve the surface waves) and then by eliminating the accelerations from the computations and using a pressure signal to infer vertical motions only. An evaluation of

*χ*estimates from the degraded measurements is made.

_{T}This paper begins with a brief review of the measurements (section 2), followed by a description of the method used to estimate *χ _{T}*,

*ε*, and

*K*from scalar gradient spectra (section 3), including the particular problem of removing the part of the temperature gradient signal solely resulting from vertical pumping through a temperature gradient by surface waves (section 3c). We compare the results of inertial-convective subrange estimates to those estimated from high-wavenumber scaling of the spectra in section 4, and then address the issue of how well we might quantify the turbulence with less well-resolved measurements by suitably degrading the signal (section 5). We close with a set of conclusions (section 6).

_{T}## 2. Measurements

In September 2005, three *χ*pods were deployed on National Oceanic and Atmospheric Administration’s (NOAA’s) Tropical Atmosphere Ocean (TAO) mooring at 0°, 140°W at depths of 29, 49, and 84 m. Each unit was equipped with two fast thermistors, a pressure sensor, a compass, and three orthogonally positioned linear accelerometers. Temperature (*T*) was sampled at 10 Hz and its time derivative (*T _{t}*) at 120 Hz. The accelerometers reveal that the

*χ*pods experience a convoluted motion, primarily resulting from surface wave pumping. The mooring’s surface float transmits large vertical and horizontal motions induced by surface gravity waves to the mooring cable. Horizontal motions are attenuated with increasing depth, while the vertical motions in the surface gravity wave band (0.05–1 Hz) are constant with depth to at least 84 m.

A complete description of *χ*pod and of this particular deployment is given by Moum and Nash (2009). Corresponding background current speed (*U _{s}*) in the earth’s coordinate system was calculated from current components (

*U*=

_{s}*u*) past a sensor requires the measurement of the current components (

*U*,

^{B}*V*,

^{B}*W*) in the

^{B}*χ*pod coordinate system transformed from the earth’s coordinate system as well as the

*χ*pod motion itself (

*u*,

_{c}*υ*,

_{c}*w*), computed through time integration of accelerometer measurements. Thus,

_{c}*u*= [(

*u*−

_{c}*U*)

^{B}^{2}+ (

*υ*−

_{c}*V*)

^{B}^{2}+ (

*w*−

_{c}*W*)

^{B}^{2}]

^{1/2}. The transformations to correctly obtain

*u*and a complete analysis of the motion of

*χ*pod on the TAO mooring using a full inertial navigation system is described by A. Perlin and J. N. Moum (2010, unpublished manuscript).

Using the horizontal component of the temperature gradient *T _{x}* computed from the time derivative

*T*via Taylor’s frozen flow hypothesis

_{t}*T*=

_{x}*T*/

_{t}*u*, Moum and Nash (2009) computed

*χ*from the viscous-convective and viscous-diffusive subranges of the temperature gradient spectra on 1-s intervals. These computations form the basis for comparison to the inertial-convective estimates made here.

_{T}The local vertical gradient of temperature *T _{z}* was derived by fitting temperature to depth over many surface wave periods. Two-minute-averaged fits were used to compute

*T*in this paper, the same as in the analysis of Moum and Nash (2009).

_{z}## 3. Methodology

### a. Definitions

*χ*from either viscous-convective and viscous-diffusive subranges (Moum and Nash 2009) or from inertial-convective subranges (herein), we use an iterative procedure described by Moum and Nash (2009) that requires an estimate of the turbulent kinetic energy dissipation rate of

_{T}*ε*;

*χ*is defined as where

_{T}*D*is the thermal diffusivity and Φ

_{T}_{Tx}(

*f*) is the frequency (

*f*) spectrum of

*T*.

_{x}*K*) can be estimated as

_{T}*ε*leads to another estimate of eddy diffusivity for density (

*K*), where

_{ρ}*N*

^{2}is the local squared buoyancy frequency, and we assume Γ = 0.2 is a constant mixing efficiency (Osborn 1980). If

*K*=

_{T}*K*, then Here, we use

_{ρ}*ε*to indicate that the TKE dissipation rate estimated from thermal variance dissipation rate

_{χ}*χ*.

_{T}To avoid the wave-induced contamination between 0.05 and 1 Hz, Moum and Nash (2009) estimated *χ _{T}* from windowed and detrended 1-s records. Such short records permitted the evaluation of temperature gradient spectra in viscous-convective and viscous-diffusive subranges. We use high-wavenumber estimates of

*χ*,

_{T}*ε*, and

_{χ}*K*in the present analysis and refer to them here as

_{T}*χ*,

_{T}^{o}*ε*and

_{χ}^{o}*K*(Table 1).

_{T}^{o}### b. Inertial-convective subrange

_{Tx}in the inertial-convective subrange (Batchelor 1959) is defined as where

*C*≈ 0.4 (Sreenivasan 1996) is the Obukhov–Corrsin constant.

_{T}*k*= 2

*πf*/

*u*, the equivalent frequency spectrum is Using Eq. (4), Eq. (6) can be rewritten in terms of

*χ*as or in terms of

_{T}*ε*as

_{χ}*f*< 0.5 Hz is dominated by inertial-convective motions. We fit observed spectra to Eq. (7) over this frequency range, minimizing

*e*over the inertial-convective subrange, where

*f*

_{1}= 0.05 Hz, and

*f*

_{2}= 0.5 Hz;

*χ*

_{T}^{IC},

*ε*

_{χ}^{IC}, and

*K*

_{T}^{IC}(Table 1). However, because in most cases (e.g., Fig. 2b) the inertial-convective range is influenced by motions induced by surface gravity waves, we must first attempt to minimize the effects of surface wave contamination.

### c. Removal of surface wave contamination

*T*on a mooring cable that is vertically pumped by surface waves via the mooring’s float contain a contribution resulting from that motion through the vertical temperature gradient

_{t}*T*(Moum and Nash 2009). The vertical cable speed is deduced by integrating the vertical component of acceleration and denoted as

_{z}*w*. In the case of these particular measurements, surface wave contamination is apparent over the frequency band 0.05 <

_{c}*f*< 1 Hz (Fig. 2). The measured signal

*T*

_{t}^{meas}can be considered to be the sum of the environmental signal

*T*

_{t}^{envir}plus a contribution from the surface wave–induced pumping

*T*

_{t}^{wave}, Using Taylor’s frozen flow hypothesis, the spatial derivative of temperature is

*T*

_{x}^{meas}and

*w*/

_{c}T_{z}*u*.

Two examples are shown in Fig. 2. In Fig. 2a, *T _{z}* is relatively small and the surface wave contamination of the measured spectrum is nonexistent. In the frequency band 0.05 <

*f*< 0.5 Hz, the slope of the measured spectrum is close to

*f*

^{1/3}, which is expected for the inertial-convective subrange [Eq. (6)]. In contrast, the spectrum shown in Fig. 2b coincides with larger

*T*, and the surface wave pumping dominates the spectrum in the surface gravity wave band, 0.05 <

_{z}*f*< 1 Hz. However, correction of the measured spectrum via Eq. (12) results in a spectrum similar to that in Fig. 2a.

Inertial-convective subrange fits to the corrected spectra in Figs. 2a,b are shown in Fig. 3. Both *χ _{T}*

^{IC}and

*ε*

_{χ}^{IC}were estimated from 120-min spectral averages. These were compared to the 120-min averages of

*χ*and

_{T}^{o}*ε*in Fig. 3. Differences are about a factor of 2–3.

_{χ}^{o}## 4. Statistical comparison

To assess the skill of the combination of surface wave contamination removal and the inertial-convective subrange fitting procedure in estimating *χ _{T}*

^{IC},

*ε*

_{χ}^{IC}, and

*K*

_{T}^{IC}, we compare them to averaged values of

*χ*,

_{T}^{o}*ε*, and

_{χ}^{o}*K*from the 6-week records obtained at 0°, 140°W and 29-m depth in September–October 2005 (Fig. 4). The initial comparison is on 120-min interval estimates. Two distinct trends appear in Fig. 4b and they are detected by both

_{T}^{o}*χ*and

_{T}^{o}*χ*

_{T}^{IC}. The diurnal cycle of upper equatorial ocean mixing is clear, as is a longer period trend associated with changes in stratification and currents at 29 m. There also exist periods of time, however, when significant and prolonged differences between

*χ*and

_{T}^{o}*χ*

_{T}^{IC}occur. Large differences of up to a factor of 20 occur under various magnitudes of

*u*,

*U*, and

_{s}*T*(Figs. 4a,b).

_{z}We have been unable to diagnose the precise reasons for these discrepancies. However, one source of contamination that may adversely enhance the high wavenumbers, while perhaps not affecting the inertial-convective subrange of wavenumbers, may occur when mean flow speeds are low, that is, much smaller than cable speeds. In this case, the pumping of the *χ*pod and cable will create a wake that is not flushed from the measurement volume by the mean current. This will directly affect high wavenumbers, while larger scales may be advected through undisturbed. In this case, *χ _{T}*

^{IC}will provide a better estimate. This may explain some cases where

*χ*

_{T}^{IC}≪

*χ*and

_{T}^{o}*U*is small, such as the periods near 26 October and 4 November, which are associated with minima in

_{s}*U*(Fig. 4b).

_{s}Another potential contaminant is of a biological nature, which has concerned us from the beginning of the project. Growth of biology, including bacterial films on the thermistor, will also reduce the heat transfer rate from the fluid to the thermistor, thereby slowing the thermistor’s response. This will attenuate the thermistor’s high-wavenumber (fast) response but not the low-wavenumber (slow) response. Although we have taken considerable effort to minimize biological contamination, by making sensors using copper and adding protectors (Moum and Nash 2009), it is possible that this may occur on some deployments, and we continue to monitor the frequency response between beginning and ending deployments as a check. The attenuation ought to produce the result *χ _{T}^{o}* ≪

*χ*

_{T}^{IC}.

A statistical comparison derived from the full 6-week record that includes, but deemphasizes, the few examples of poor agreement, indicates no significant bias between the two estimates *χ _{T}^{o}* and

*χ*

_{T}^{IC}when computed over 120-min intervals (Fig. 5). Median values are within 40% of each other (Figs. 5a,d). Mean values differ by about a factor of 2, dominated by the relatively few very high values of

*χ*that are not predicted by

_{T}^{o}*χ*

_{T}^{IC}. This result is similarly reflected in comparative estimates of

*ε*(Figs. 5b,e) and

_{χ}*K*(Figs. 5c,f). A summary of the results in Fig. 5d indicates that 51% of the [

_{T}*χ*,

_{T}^{o}*χ*

_{T}^{IC}] pairs are within a factor of 2 of each other, 75% are within a factor of 3, 91% are within a factor of 5, and virtually all of the estimates (98%) are within a factor of 10.

We know that the natural variability of turbulence can contribute to variations in *χ _{T}* by several orders of magnitude on times shorter than 120 min. When this happens the spectrum of

*T*does not represent a stationary process, in violation of Batchelor’s original assumptions, so the form given by Eq. (7) is not necessarily representative. As a test, we recomputed

_{x}*χ*and

_{T}^{o}*χ*

_{T}^{IC}on time intervals of 15, 30, and 60 min. A statistical comparison indicates similar agreement at all intervals (Fig. 6). In all cases, approximately 90% of the data pairs are within a factor of 5 of each other. As a point of reference for the degraded results discussed in section 5, the kurtosis (

*K*) of the ratio

*χ*

_{T}^{IC}/

*χ*for the four different time intervals of 15, 30, 60, and 120 min is 3.4, 3.3, 3.2, and 3.1, respectively. Here, kurtosis of the normal distribution is 3. Distributions that are more outlier prone than the normal distribution have kurtosis greater than 3.

_{T}^{o}## 5. Estimates from degraded measurements

The *χ*pod measurements were intended to resolve the full spectrum of temperature gradient (within the limitations of the thermistors). To do this, the data were sampled at 120 Hz, and dynamic response corrections derived from separate field calibration measurements were then applied to spectra prior to scaling in the high-wavenumber turbulence subranges (Moum and Nash 2009). As well, the *χ*pod motion was measured to determine both vibrational contamination and cable speeds. However, if the inertial-convective subrange scaling is adequate, then we may be able to accept lower data rates and less-well-resolved motion measurements. We test this first by subsampling the temperature gradient and accelerometer signals at 2 Hz (thus at least ensuring resolution of the surface wave–induced motion), and using the cable speed determined by integration of the subsampled accelerometer record. We then suppose that we have no accelerometer, and determine the cable speeds through differentiation of the pressure record on the *χ*pod. This final test means that *u _{c}* and

*υ*are not determined (and set to 0 in our computation). However, an estimate of

_{c}*w*is still available, which is essential for removal of the surface wave–induced contamination of Φ

_{c}_{Tx}.

First, the records of temperature gradient (*T _{t}*) and acceleration (

*A*,

_{c}^{x}*A*,

_{c}^{y}*A*−

_{c}^{z}*xyz*coordinates in the

*χ*pod body reference frame), originally sampled at 120 Hz, were filtered at 0.75 Hz and resampled at 2 Hz. Spectra corresponding to the measured temperature gradient signal and the vertically pumped signal were computed (Fig. 7), and the environmental signal was recovered via Eq. (12). Then the procedure to compute

*χ*by scaling the inertial-convective subrange from Eq. (7) was applied. Resultant spectra are shown in Fig. 7 as solid lines. The results from this trial are designated as

_{T}*χ*

_{T}^{IC1},

*ε*

_{χ}^{IC1}, and

*K*

_{T}^{IC1}(Table 1).

The procedure was then repeated with the assumption that no accelerometer was available to compute cable speeds. In this case, we estimate *w _{c}* from the depth record derived from the

*χ*pod’s pressure signal

*w*=

_{c}*dz*/

*dt*, but we now have no means of estimating

*u*and

_{c}*υ*. The spectra are shown as dotted lines in Fig. 7. The results from this trial are designated as

_{c}*χ*

_{T}^{IC0},

*ε*

_{χ}^{IC0}, and

*K*

_{T}^{IC0}(Table 1).

Estimates of *χ _{T}*,

*ε*, and

_{χ}*K*from the degraded signals are compared as time series in Fig. 8. We first consider

_{T}*χ*

_{T}^{IC1}(blue trace in Fig. 8b). Both long-term trends and the diurnal cycle of mixing in the upper equatorial ocean are reproduced. However, the distribution of the ratio

*χ*

_{T}^{IC1}/

*χ*(Fig. 9a) shows that, in comparison to that of

_{T}^{o}*χ*

_{T}^{IC}/

*χ*(Fig. 6d),

_{T}^{o}*σ*is larger (cf. 2.6 –3.2); the median is significantly less than 1; and there is a significant number of outliers, predominantly at high values, as noted by the distribution kurtosis

*K*= 4.6, which is considerably larger than 3. For comparison, 82% of the data pairs are within a factor of 5 of each other.

The more seriously degraded signal *χ _{T}*

^{IC0}occasionally severely overestimates

*χ*for extended periods, particularly when

_{T}^{o}*U*is small (

_{s}*χ*

_{T}^{IC0}is the red trace in Fig. 8b). When

*U*is small, the contribution of cable velocities to

_{s}*u*are correspondingly large. Because in this case we have no measure of

*u*and

_{c}*υ*, our estimate of

_{c}*u*must be low, and hence our estimate

*χ*

_{T}^{IC0}must be high via Eq. (7), because Φ

_{Tx}∝

*u*

^{−4/3}. This is represented in the distribution of

*χ*

_{T}^{IC0}/

*χ*by a significantly larger standard deviation, high values of mean and median, and the large number of outliers at high values of the ratio (Fig. 9b).

_{T}^{o}## 6. Conclusions

To summarize, Figs. 5a,d show that estimated values of *χ _{T}*

^{IC}and

*χ*range over four orders of magnitude during the roughly 50-day period that we have examined for the purpose of this comparison. Figure 4b shows, with some exceptions, that

_{T}^{o}*χ*

_{T}^{IC}and

*χ*covary on daily and longer time scales. Mean values computed over the entire time period differ by a factor of 2 (Fig. 5a). These are weighted by the outliers because mean values of ratios differ by only 10% (Fig. 6d). It is not clear whether

_{T}^{o}*χ*

_{T}^{IC}or

*χ*represents the better estimate in any single sample, although we suspect that each will, in turn, provide a better estimate under different circumstances (as noted in section 4). These distributions do not show appreciable bias between the two estimates (10% at most in comparison of means of ratios). In comparison, (Moum and Nash 2009) estimated the uncertainty in individual 1-s estimates of

_{T}^{o}*χ*to be less than a factor of three 95% of the time. This uncertainty is largely because of uncertainty in

_{T}^{o}*u*. Averaging to 60-s intervals reduced the uncertainty to a factor of 2.

A somewhat different assessment is indicated in Fig. 5d, where it is noted that 91% of the data pairs agree within a factor of 5, or 75% of the data pairs agree within a factor of 3. Is this good enough? One metric for comparison is to the natural variability estimated from different measurements of the same process. From an experiment at the same location in 1991 (as part of the Tropical Instability Wave Experiment), Moum et al. (1995) compared estimates of *ε* from two different turbulence profilers deployed by two different groups on two ships over the same 3.5-day period and within 11 km of each other. They concluded that the systematic bias in *ε* was less than a factor of 2, which in turn was less than the estimated cumulative uncertainty in individual estimates of *ε*. Differences in hourly averaged data pairs at identical depths were attributed to natural variability. A reanalysis of the data pairs at 75-m depth (where the tightest statistical agreement was found) reveals that 51% of the data pairs were within a factor of 2 of each other, 73% within a factor of 3, 87% within a factor of 5, and 94% within a factor of 10. The agreement at 50 m was significantly poorer.

A direct measure of natural variability in *χ _{T}* at this same location comes from A. Perlin and J. N. Moum (2010, unpublished manuscript). The

*χ*pod measurements at the same depth (60 m) on moorings separated by 9 km, and CHAMELEON (a loosely tethered turbulence profiler) profiling measurements intermediate between the two moorings, yield three sets of data pairs from a 16-day period in October–November 2008. These show that roughly 30% of the data pairs were within a factor of 2 of each other, 40% within a factor of 3, 60% within a factor of 5, and 70% within a factor of 10. These ranges are markedly larger than those between our two estimates of

*χ*shown in Fig. 5d, which is due to nearly random variations. This is encouraging because as long as there is no bias between our two estimates, the natural variability that must be resolved to achieve meaningful averages is greater than the variations between the two methods of estimation.

_{T}Another consideration is the long-term variations in the turbulence caused by external forcing. This is emphasized by the demonstration of greatly enhanced mixing at the same location (0°, 140°W) associated with tropical instability waves (Moum et al. 2009), resulting in mean values of *ε* above the core of the Equatorial Undercurrent by a factor of 10 greater than any previous observations (Moum et al. 2009, their Fig. 2a). The differences in our two methods are much smaller than such large variations, permitting unambiguous detection.

The reasonable agreement between the two estimates *χ _{T}*

^{IC}and

*χ*leads us to conclude that our surface wave removal is effective. Because wave-contaminated spectral values are frequently two decades higher (e.g., Fig. 2b), we would expect a considerable bias in our estimate of

_{T}^{o}*χ*

_{T}^{IC}from corrected spectra. If we systematically did not remove enough of the contaminating signal, then

*χ*

_{T}^{IC}will be high, and if we systematically removed too much, then it will be low.

Our evaluation of the degraded signals indicates that the resolution of the wave field by both temperature sensors and independent motion sensors (all three components) is necessary. It is considerably more important to measure the motion sufficiently (wave-resolving independent estimates of *w _{c}*,

*u*, and

_{c}*υ*are required) than to oversample the records above the wave frequencies. By a proper and independent measurement of the wave-contaminated component of the temperature gradient signal, it can be effectively removed.

_{c}This work was funded by the National Science Foundation (0424133). Yanwei Zhang was supported by the State Key Laboratory of Marine Geology and Program for Young Excellent Talents in Tongji University, as well as Open Fund of the Key Lab of Ocean Circulation and Waves from Chinese Academy of Sciences. Thanks to Mike Neeley-Brown and Ray Kreth for their technical assistance and to A. Perlin for computing help. Jonathan Nash and A. Perlin have made important contributions to this project. Comments by Emily Shroyer, Aurelie Moulin, and Uwe Stoeber on an early draft are appreciated. Constructive comments by three anonymous reviewers helped to improve this paper. The continued involvement of NOAA’s National Data Buoy Center through the TAO Project Office and by the Pacific Marine Environmental Laboratory has been critical to the success of this project.

## REFERENCES

Batchelor, G. K., 1959: Small-scale variation of convected quantities like temperature in turbulent fluid.

,*J. Fluid Mech.***5****,**113–139.Dillon, T. M., , and Caldwell D. R. , 1980: The Batchelor spectrum and dissipation in the upper ocean.

,*J. Geophys. Res.***85****,**1910–1916.Gargett, A., 1985: Evolution of scalar spectra with the decay of turbulence in a stratified fluid.

,*J. Fluid Mech.***159****,**379–407.Gibson, C. H., , and Schwarz W. H. , 1963: The universal equilibrium spectra of turbulent velocity and scalar fields.

,*J. Fluid Mech.***16****,**365–384.Klymak, J. M., , and Moum J. N. , 2007: Oceanic isopycnal slope spectra. Part II: Turbulence.

,*J. Phys. Oceanogr.***37****,**1232–1245.Kraichnan, R., 1968: Small-scale structure of a scalar field convected by turbulence.

,*Phys. Fluids***11****,**945–953.Levine, E. R., , and Lueck R. G. , 1999: Turbulence measurement from an autonomous underwater vehicle.

,*J. Atmos. Oceanic Technol.***16****,**1533–1544.Luketina, D. A., , and Imberger J. , 2001: Determining turbulent kinetic energy dissipation from Batchelor curve fitting.

,*J. Atmos. Oceanic Technol.***18****,**100–113.Moum, J. N., , and Nash J. D. , 2009: Mixing measurements on an equatorial ocean mooring.

,*J. Atmos. Oceanic Technol.***26****,**317–336.Moum, J. N., , Gregg M. C. , , Lien R. C. , , and Carr M. E. , 1995: Comparison of turbulence kinetic energy dissipation rate estimates from two ocean microstructure profilers.

,*J. Atmos. Oceanic Technol.***12****,**346–366.Moum, J. N., , Lien R-C. , , Perlin A. , , Nash J. D. , , Gregg M. C. , , and Wiles P. J. , 2009: Sea surface cooling at the equator by subsurface mixing in tropical instability waves.

,*Nat. Geosci.***2****,**761–765.Nash, J. D., , and Moum J. N. , 1999: Estimating salinity variance dissipation rate from conductivity microstructure measurements.

,*J. Atmos. Oceanic Technol.***16****,**263–274.Nash, J. D., , and Moum J. N. , 2002: Microstructure estimates of turbulent salinity flux and the dissipation spectrum of salinity.

,*J. Phys. Oceanogr.***32****,**2313–2333.Osborn, T. R., 1980: Estimates of the local rate of vertical diffusion from dissipation measurements.

,*J. Phys. Oceanogr.***10****,**83–89.Osborn, T. R., , and Cox C. S. , 1972: Oceanic fine structure.

,*Geophys. Fluid Dyn.***3****,**321–345.Smyth, W. D., 1999: Dissipation range geometry and scalar spectra in sheared, stratified turbulence.

,*J. Fluid Mech.***401****,**209–242.Sreenivasan, K. R., 1996: The passive scalar spectrum and the Obukhov-Corrsin constant.

,*Phys. Fluids***8****,**189–196.

Summary of notation used in this paper for thermal variance dissipation rate *χ _{T}*. Corresponding turbulent kinetic energy dissipation rate (

*ε*) and eddy diffusivity (

_{χ}*K*) are designated similarly.

_{T}