## 1. Introduction

The rate of turbulent kinetic energy dissipation ɛ is an important parameter in turbulent mixing theory. Knowledge of the magnitude of ɛ allows the estimation of the rate of vertical mixing (Gregg et al. 1977b; Ivey and Imberger 1991; and Saggio and Imberger 2000, hereafter SI).

It is well known that the higher wavenumber part of the one-dimensional temperature gradient spectrum due to turbulence (the Batchelor spectrum) is a function of the dissipation ɛ and the dissipation of temperature variance *χ*_{T}. Stewart and Grant (1962) describe a graphical method for fitting a nondimensionalized Batchelor spectrum to temperature gradient data to obtain the dissipation ɛ. Dillon and Caldwell (1980) and Dillon (1982) used a nonlinear least squares method to fit the Batchelor spectrum to well-resolved temperature gradient spectra with high signal-to-noise levels. Some temperature gradient spectra are difficult to fit in this way because of instrument noise at the high wavenumber end of the spectra (Gregg et al. 1986), internal wave and finestructure “contamination” at the low-wavenumber end of the spectra (Gregg 1977a,b), and the lack of a clearly defined viscous convective subrange (Sherman and Davis 1995). Figure 1 depicts typical spectra with these additional spectral components. Note that although temperature gradient spectra may be corrected for sensor response, uncertainty in the response is a source of distortion in measured spectra (Gregg 1999).

In this paper, we present an algorithm for obtaining ɛ by fitting the theoretical Batchelor spectrum to the measured spectrum. Key issues in fitting the theoretical Batchelor curve that are discussed in this paper are 1) what is the appropriate portion of the measured spectrum to use, 2) how should the dissipation of temperature variance *χ*_{T} be determined from the measured spectrum, and 3) what is an appropriate error function to use in determining the optimum fit? This paper focuses on the first two of these items and describes a fitting algorithm that has been developed and refined over a period of several years. This algorithm has been successfully applied to temperature gradient spectra from a wide variety of turbulent forcing in the field (e.g., Luketina and Imberger 1989; Imberger and Ivey 1991; SI) and in the laboratory (e.g. Teoh et al. 1997; and De Silva et al. 1997).

The data discussed in this paper were generally collected at 100 Hz using an FP07 thermistor mounted on a probe rising or falling at a nominal velocity of 0.1 m s^{−1}. However, the algorithm can be applied to a probe traveling in any direction and at velocities other than of 0.1 m s^{−1}.

The paper is divided into six sections. In section 2 the features of temperature gradient spectra are discussed. Data requirements, including preprocessing of data, are covered in section 3. The method for obtaining the dissipation ɛ by fitting the theoretical Batchelor spectrum to the measured temperature gradient spectrum (hereafter referred to as the measured spectrum) is outlined in section 4. Results from the application of the fitting method to measured and artificially generated spectra are discussed in section 5. Conclusions are presented in section 6.

Finally, it should be noted that wavenumbers are expressed as both cycles per meter (cpm) and radians per meter in this paper. The symbols *κ* and *k* are consistently used for wavenumbers in units of cpm and rad m^{−1}, respectively. Wavenumbers used in figures always have units of cpm.

## 2. Temperature gradient spectra

The typical temperature gradient spectrum shown in Fig. 1 has five different components as shown. These components are due to finestructure, internal waves, inertial convective turbulence (the inertial convective subrange), turbulence affected by viscosity and diffusivity (the Batchelor spectrum), and noise.

### a. Finestructure

*T*

*T*

_{o}

*β*

*z*

^{γ}

*z*

*γ*

*β*

*γ*and

*β*are constants,

*z*′ is the vertical ordinate whose origin is at the center of the region of interest and is positive upward, and

*T*

_{o}is the temperature at the center of the region. The temperature gradient is thus given by

Examples of stable 1-m-long temperature profiles with a 1°C change from top to bottom, which satisfy (1), are shown in Fig. 2a. By examining Fig. 2a it is clear that (2) describes a wide range of stable temperature profiles. In practice, it is expected that *γ* would most likely be in the range 0 ⩽ *γ* ⩽ 2 (*β* then depends on the temperature change from the top to the bottom of the profile).

*S*

_{f}(

*k*) of (2) is given by

*S*

_{f}(

*k*) = 2

*F*(

*k*)

*F**(

*k*), where

*F*denotes the Fourier transform of

*dT*/

*dz*′, the asterisk denotes the complex conjugate, and

*k*is the wavenumber (rad m

^{−1}). The Fourier transform of (2) is

*γ*+ 1) =

^{∞}

_{0}

*x*

^{γ}

*e*

^{−x}

*dx.*Thus

*S*

_{f}(

*k*) is given by

*S*

_{f}

*k*

*c*

_{f}

*k*

^{−2γ}

*γ*

*c*

_{f}= 8[(2

*π*)

^{−γ}

*γβ*cos(

*γπ*/2)Γ(

*γ*)]

^{2}. For

*γ*in the expected range of 0 ⩽

*γ*⩽ 2 the slope of the spectrum, as given by (4), will be between 0 and −4 in log–log space.

Band-averaged spectra *S*_{f}(*k*) of the gradients of the temperature profiles of Fig. 2a are shown in Fig. 2b. The temperature gradients were determined by applying a Gaussian gradient filter to the temperature profiles—this removed the problem of the temperature gradient as given by (2) becoming infinite at *z*′ = 0. The temperature gradient data record was then windowed (see Press et al. 1989, p. 423 for details) prior to applying a discrete Fourier transform to yield *F*(*k*). The filtering is responsible for the spectra with *γ* = 0 and *γ* = 0.5 deviating from the theoretical spectra given by (4) at wavenumbers above approximately 300 cpm. The spectrum with *γ* = 2 deviates because the high frequency content of the temperature gradient profile is located at the ends of the data record, which are subsequently modified by windowing. In any case, it is expected that the finestructure spectrum *S*_{f}(*k*) derived via discrete Fourier transforms will generally have a slope of between 0 and −4 at low wavenumbers in log–log space.

### b. Internal waves

*w*denotes internal waves and, based on the data of Gregg et al. (1973),

*c*

_{w}is an

*O*(1) constant. Internal wave effects can extend up to a maximum frequency of

*N,*where

*N*

^{2}= −(

*g*/

*ρ*)(∂

*ρ*/∂

*z*) and

*ρ*is the fluid density (in wavenumber space this will correspond to a maximum wavenumber of 2

*πN*/

*u,*where

*u*is the sensor velocity relative to the water).

### c. Inertial convective subrange

*πN*/

*u,*but large enough to be influenced by viscosity, we have the inertial convective subrange. The temperature gradient spectrum in the inertial convective subrange is given by (Monin and Yaglom 1975, 479–485)

*S*

_{ICS}

*k*

*c*

_{ICS}

*χ*

_{T}

^{−1/3}

*k*

^{1/3}

*c*

_{ICS}is an

*O*(1) constant and

*χ*

_{T}is the dissipation of temperature variance due to turbulence and is defined for the case of isotropic turbulence by

*D*

_{T}is the diffusivity of heat, and

*T*′ is the temperature fluctuation. The spectrum given by (6) is valid to a wavenumber (

*c*

_{ICS}/

*c*

_{VC})

^{3/2}

*k*

_{k}where the Kolmogorov wavenumber is

*k*

_{k}= (ɛ/

*ν*

^{3})

^{1/4}and

*c*

_{VC}is a constant related to the spectrum in the viscous convective subrange (see below).

### d. Batchelor spectrum

*S*(

*k*) of the temperature gradient

*k*has units of radians per meter,

*q*is a universal constant, Batchelor wavenumber

*k*

_{B}is defined as the inverse of the Batchelor length scale

*L*

_{B}= (

*D*

^{2}

_{T}

*ν*/ɛ)

^{1/4}, and

*α*is a dimensionless wavenumber given by (2

*q*)

^{1/2}

*kk*

^{−1}

_{B}

*S*

_{N}(

*α*) is given by

The one-dimensional form *S*(*k*) has been confirmed in laboratory experiments involving turbulence generated behind a grid (Gibson and Schwartz 1963), in the upper ocean (Dillon and Caldwell 1980), and in lakes (Imberger 1985).

*ν*/

*D*

_{T}> 1 is the viscous convective subrange and is given by (Batchelor 1959; Gibson and Schwartz 1963)

*c*

_{VC}is an

*O*(1) constant (Dillon and Caldwell 1980; Oakey 1982). Sherman and Davis (1995) have shown that the viscous convective subrange is only found for events with large values of the Cox number

*C*

_{T}=

*χ*

_{T}/2

*D*

_{T}

*T*

^{2}

_{z}

*T*

_{z}is the mean temperature gradient. We shall refer to the portion of the Batchelor spectrum at wavenumbers higher than that of the spectral maximum as the roll-off region (see Fig. 1a).

The value of the universal constant *q* was estimated to be 3^{1/2} < *q* < 2(3^{1/2}) from theoretical arguments advanced by Gibson (1968a,b). Grant et al. (1968) estimated *q* = 3.9 ± 1.5, and Dillon and Caldwell (1980) used *q* = 2(3^{1/2}) = 3.46. Oakey (1982) determined *q* = 3.67 ± 1.52 by comparing direct measurements of dissipation with those obtained from temperature gradient spectra. In this paper it is assumed that *q* = 2(3^{1/2}).

It should be noted that the Batchelor spectrum can be applied to any resolvable tracer that has a Prandtl reasonably greater than unity.

### e. Noise spectra

Noise is generated either at the sensor or by the processing circuitry. Commonly, the noise spectrum will have a slope of around 2 (this corresponds to white noise). In any case, the approach used in this paper is not sensitive to the exact form of the noise spectrum. The wavenumber at the intersection of the Batchelor and noise spectra is denoted *k*_{n}.

There have been various attempts to reduce the effects of noise. For example, Dillon and Caldwell (1980) used the in-phase power between redundantly amplified channels to reduce electronic noise. However, this technique is not necessarily useful for newer microstructure probes where electronic noise levels have been reduced resulting in sensor noise dominating electronic noise.

## 3. Data requirements

Before Batchelor curve fitting can be used to determine dissipation it is usually necessary that 1) the frozen field assumption be valid, 2) the temperature signal is corrected for the thermistor response, and 3) stationary data segments are selected. The frozen field assumption, or Taylor hypothesis (Taylor 1938), is used to convert the response of a thermistor from frequency *ω* to wavenumber *k* space using *k* = *ω*/*u,* where the frequency *ω* is measured in radians per second and wavenumbers given in radians per meter. The Taylor hypothesis is generally considered to be valid if variations in the fluctuating velocities are small compared with *u,* the probe velocity relative to the fluid (see Townsend 1976, p. 67). Ideally, the probe speed will be reasonably greater than the largest turbulent velocity fluctuations but not so great that sensor roll-off becomes a problem. Given that *u* ∼ (ɛ*l*)^{1/3} where *u* and *l* are the rms velocity and overturn size of the turbulence (Luketina and Imberger 1989), and *l* is typically less than 0.1 m and ɛ around 10^{−5} m^{2} s^{−3} in an energetic environment, results in *u* ∼ 10^{−2} m s^{−1}. Thus a nominal probe velocity of 0.1 m s^{−1} is a good compromise between satisfying the frozen field requirement and minimizing sensor roll-off. Higher probe velocities can be used in less energetic environments.

Because much of the spectral energy encountered by a thermistor is near the highest frequencies the thermistor can resolve, it is usually necessary to correct for the thermistor response. Generally, the response of a thermistor to changes in temperature is limited by the time for heat to diffuse through the boundary layer surrounding the sensor (Gregg and Meagher 1980). Response functions and correction methods for fast response thermistors have been published by Gregg and Meagher (1980), Dillon and Caldwell (1980), Fozdar et al. (1985), and Sherman and Davis (1995). Here the method of Fozdar et al. (1985) is used.

It is necessary to divide the data into statistically stationary segments or intervals if the measured spectrum *D*(*k*), or a portion of it, is to behave as a Batchelor spectrum (Imberger and Boashash 1986). An example of this is provided in section 4; arbitrarily dividing the data into short segments is not a guarantee that the turbulence is homogeneous within a segment.

Note that comments relating to measured spectra in subsequent sections of this paper refer to (wavenumber) band-averaged spectra. The number of spectral points *n*_{i} averaged to form the *i*th point in the band-averaged spectrum is a geometric series with *n*_{i} increasing from low to high wavenumbers. Thus confidence in the accuracy of the spectral estimate increases with increasing wavenumber (see Bendat and Piersol 1986, p. 285).

## 4. Method

When a theoretical Batchelor spectrum as given by (8) is fitted to a spectrum for which the variance has been estimated, the only free variable is the Batchelor wavenumber, which in turn, defines the dissipation. As mentioned previously, key issues in fitting the theoretical Batchelor curve are 1) what is the appropriate portion of the measured spectrum to use, 2) how should the variance *χ*_{T} be estimated from the measured spectrum, and 3) what is an appropriate error function to use in determining the optimum fit? These issues and other aspects of the method are discussed in this section.

### a. Background

*k*having units of radians per meter. If the spectra being fitted are based on cyclical wavenumbers

*κ*having units of inverse meters, then the nondimensional wavenumber

*α*is defined by

*α*= (2

*q*)

^{1/2}

*κκ*

^{−1}

_{B}

*κ*

_{B}= (2

*πL*

_{B})

^{−1}. For the remainder of this paper we shall only deal with cyclical wavenumbers. The measured spectrum will be denoted

*D*(

*κ*) to distinguish it from the theoretical or fitted spectrum

*S*(

*κ*). The universal constant

*q*in (7) is assumed to be 2(3

^{1/2}) and the nondimensional theoretical spectrum

*S*

_{N}(

*α*) as defined in (9) is approximated by

*π*

^{4}

*νD*

^{2}

_{T}

*κ*

^{4}

_{B}

### b. Overview of method

Once a value of the variance *χ*_{T} is estimated, the theoretical spectrum *S*(*κ*) can be fitted to an appropriate portion of the measured spectrum *D*(*κ*) with the Batchelor wavenumber *κ*_{B} being the free variable. Since we do not know a priori what the “appropriate portion” of the measured spectrum is, we use a trial and error approach of fitting the theoretical Batchelor spectrum to all possible portions (the method for determining possible portions is discussed later in this section). This entire process is repeated up to six times with the estimates of the noise wavenumber *κ*_{n} and the variance *χ*_{T} being progressively refined (see the following section). The dissipation is then found by substituting the wavenumber *κ*_{B}, associated with the best global fit into (14).

A nonlinear curve fitting algorithm could be used to try to find the best fit (i.e., that with the least error) of the theoretical temperature gradient spectrum to the measured data. However, experience has shown that these fitting algorithms often find only the best local rather than global fit. Further, they are computationally demanding. For this reason, a trial and error approach is used to find the best fit of the theoretical Batchelor spectrum to each of the possible portions of the measured spectrum. In this approach, we trial 200 geometrically spaced values of the Batchelor wavenumber *κ*_{B} such that ɛ ranges from 10^{−12} to 10^{−4} m^{2} s^{−3} (however, if *κ*_{B} exceeds *κ*_{n}, the noise wavenumber, then the upper limit for *κ*_{B} is taken to be *κ*_{n}). Of these 200 values, the wavenumber *κ*_{B} resulting in the least fitting error is assumed to be the Batchelor wavenumber.

The fitting procedure is done in log–log space as spectra are commonly presented in this manner and the roll-off region is well defined compared with a linear–linear format. There is no reason why the procedures that are presented here cannot be adapted for use with variance preserving spectra in a linear–linear format or the error function modified. In any case, the focus here is on developing an automated procedure for determining the Batchelor portion of a temperature gradient spectrum and an appropriate estimate of the variance. Until issues such as these are resolved, use of a statistically optimum error function is somewhat academic.

### c. Estimating variance

*κ*

_{L}and

*κ*

_{n}) and have fitted a theoretical spectrum to this part, we can estimate the temperature variance dissipation

*χ*

_{T}using

*D*

_{T}for water can be estimated from formulas given in Fofonoff and Millard (1983). However, we have a chicken and egg problem as the theoretical spectrum cannot be fitted until we have estimated the temperature variance dissipation

*χ*

_{T}which, in turn, requires the limiting wavenumbers,

*κ*

_{L}and

*κ*

_{n}, to be estimated. We get around this problem by an iterative procedure.

Prior to the first iteration, *κ*_{L} is set to the lowest wavenumber of the measured spectra *D*(*κ*). For each subsequent set of trials, *κ*_{L} is the lowest wavenumber at which the measured spectra *D*(*κ*) and the best-fit theoretical spectra *S*(*κ*) (from the set of fits done within the preceding iteration) intersect (see Fig. 3; if there is no intersection, *κ*_{L} does not change).

Prior to the first iteration, the noise wavenumber *κ*_{n} is set to the wavenumber corresponding to the minimum of the measured spectrum at wavenumbers greater than 30 m^{−1}. This generally ensures that the appropriate minima are selected. However, in some instances, this estimate of *κ*_{n} may be too high. By comparing the best-fit spectrum from the first iteration with the measured spectrum it is possible to reevaluate whether *κ*_{n} should be reduced. First, the highest wavenumber intersection of the theoretical (i.e., fitted) and measured spectra is found (see Fig. 3b). The measured spectrum is then scanned in the direction of higher wavenumbers until a positive gradient, indicating noise, is encountered. If the wavenumber at which this positive gradient is encountered is less than the previous value of the noise wavenumber *κ*_{n}, *κ*_{n} is replaced (see Fig. 3c). This is done only at the end of the first iteration and prior to the second iteration.

### d. Which portion of the measured spectrum should be used

Once wavenumbers higher than the noise wavenumber *k*_{n} have been excluded, the only basis for determining the Batchelor portion of the measured spectrum is that it is shaped like a Batchelor spectrum. It is assumed that some portion of the data at lower wavenumbers than *k*_{n} is Batchelor shaped. Accordingly, different portions of the data between the lowest wavenumber of the measured spectrum and the noise wavenumber *κ*_{n} are fitted. The portion of the data giving the best fit to the theoretical spectrum is deemed to be the Batchelor portion of the data.

*κ*

_{1}to

*κ*

_{2}, then the preceding restrictions can be written as

*S*(

*κ*

_{peak}) and

*κ*

_{peak}is equal to 0.176

*κ*

_{B}. Potential values of

*κ*

_{1}and

*κ*

_{2}satisfying the first two criteria are shown in Fig. 3.

### e. Error function

*χ*

_{T}is estimated, the theoretical curve as given by (8) is constrained to move along a slope of −1 in log–log space. For this reason, when working in log–log space, as is done here, the error should be the difference between the theoretical and measured spectrum in the direction of slope −1 (see Fig. 4). This error for the

*i*th point is given by

*λ*

_{i}

*D*

*κ*

_{i}

*S*

*κ*

_{Si}

^{2}

*κ*

_{i}

*κ*

_{Si}

^{2}

^{1/2}

*S*(

*κ*

_{Si}) is the point on the theoretical spectrum that lies on the −1 slope passing through

*D*(

*κ*

_{i}) (see Fig. 4). The error function that we have found to be effective involves a normalization of the error given by (17) and is

*w*

_{i}is a weight based on the number of raw spectral data points band averaged to produce

*D*(

*κ*

_{i}),

*m*is the number of data points (after band averaging) between wavenumbers

*κ*

_{1}and

*κ*

_{2},

*m*− 1 is the number of degrees of freedom,

*f*

_{r}is the fraction of the maximum possible number of data points used (i.e., all those at wavenumbers lower than

*κ*

_{n}), and

*c*is a constant.

The use of *f*_{r})^{c}*c* to 1.5 has been found to give the best fits to a wide range of measured spectra (the procedure is not sensitive to the value of *c* if the measured spectrum is clearly Batchelor shaped). Dividing by the degrees of freedom (*m* − 1) is a standard normalization. The weights *w*_{i} are based on the standard deviation of *n*_{i} points band averaged to a single point being reduced by a factor of (*n*_{i})^{−1/2}. More specifically, the weights are calculated using *w*_{i} = (*n*_{i})^{−1/2} and are then multiplied by a factor so that the average value of the weights is unity.

*S*(

*κ*

_{Si}), or more specifically

*κ*

_{Si}. The method is illustrated in Fig. 4 and is presented in terms of the normalized wavenumber

*α*

_{Si}= (2

*q*)

^{1/2}

*κ*

_{Si}/

*κ*

_{B}. Figure 4 shows a data point

*α*

_{i},

*D*

_{N}(

*α*

_{i}) on the normalized spectrum through which passes a line

*L*

_{N}(

*α*) of slope −1. The normalized spectrum is given by

*S*

_{N}(

*α*) with the −1 tangent point

*α*

_{tan},

*S*

_{N}(

*α*

_{tan}) indicated. Note that

*α*

_{tan}= 1.036. A parameter Δ is defined as

_{10}

*S*

_{N}

*α*

_{10}

*L*

_{N}

*α*

*α*

_{left}and

*α*

_{right}for Δ > 0 using the following series expansion:

*ζ*= lnΔ. The wavenumber

*α*

_{Si}is then taken to be the closer of

*α*

_{left}and

*α*

_{right}(in the example of Fig. 4,

*α*

_{Si}=

*α*

_{left}) or, if Δ < 0 (i.e., the line and fitted curve do not intersect),

*α*

_{Si}is set equal to

*α*

_{tan}.

Before moving onto the next section, we wish to make clear that we do not claim that the error function given by (18) is the statistically optimum error function (see Ruddick et al. 2000 for a discussion of the merits of different forms of the error function). However, the error function used here is convenient and is sufficient for our purposes as the error in the estimate of the log dissipation for synthetic spectra is within a few percent of the actual dissipation for the majority of cases (see section 5). Further, Ruddick et al. (2000) shows that a variety of error functions can give good results but that all error functions have some bias.

## 5. Discussion

### a. Examples of fitting to data

Figures 5 and 6 show the results of applying the curve fitting algorithm to a range of temperature gradient microstructure data. Figure 5 shows relatively good fits, while Fig. 6 shows poorer fits. These poorer fits are due to the measured spectrum lacking a clear Batchelor-shaped region. The value of the fitting error *δ*_{e} as a function of dissipation is also shown. It can be seen that the error function has a unique minimum for a measured spectrum that is clearly Batchelor shaped. As the measured spectrum becomes less Batchelor like, the minimum in the error tends to be flatter, indicating that the uncertainty in the value of the dissipation is increased. In general, poor Batchelor fits result in higher values of the minimum error and the error curve shape deviating from the shapes shown in Fig. 5—either by being flatter in the region of the minimum or more irregularly shaped.

The algorithm has been applied to many hundreds of sets of temperature gradient data and has proven itself to be robust in selecting the Batchelor portion of the temperature gradient spectrum. In measured spectra where there is a clear Batchelor portion, the placement of the theoretical Batchelor spectrum is always close to the best fit “by eye.”

### b. Accuracy

*η*(

*t*

_{i}) was calculated using

*θ*

_{j}is a uniformly distributed random variable in the range from 0 to 2

*π*;

*p*is the number of wavenumber (or Fourier) components being used to simulate

*η*(

*t*); and

*κ*

_{j}= (

*j*/

*p*)Δ

*κ,*where Δ

*κ*= (2

*p*Δ

*tu*) and Δ

*t*is the sampling interval of the time series. The magnitude of the component associated with wavenumber

*κ*

_{j}is

*A*

_{j}

*S*

*κ*

_{j}

*κ*

^{1/2}

*S*(

*κ*) is given by (8). The above relationships ensure that spectra based upon the time series will not be aliased as there is no energy in the time series above the Nyquist frequency.

Values of *D*_{T}, *p,* Δ*t* and *u* were fixed as 1.4 × 10^{−7} m^{2} s^{−1}, 300, 0.01 s, and 0.1 m s^{−1}, respectively. This left ɛ and *χ*_{T} to be specified so that a time series could be generated. Spectra were then derived by using adjacent segments of data of duration *τ* (or record length *L*_{r} = *uτ*) from a time series consisting of 51 200 points. Statistical quantities calculated for the log of the dissipation determined from Batchelor fitting were the mean *μ*_{logɛ} and the standard deviation *s*_{logɛ}. The originally specified dissipation was denoted ɛ.

Examples of artificially generated spectra and curve fits are shown in Fig. 7 for a range of dissipations. The segment length (in this instance 0.512 m) limits the amount of the low wavenumber data that can be captured, while the chosen sampling interval (Δ*t* = 0.01 s, which corresponds to 1 mm in space in this instance) limits the highest wavenumber to 500 m^{−1}. The viscous convective subrange is absent from the low dissipation spectrum due to the low wavenumber restriction, while the roll-off region is only just discernable for the high dissipation spectrum due to the high wavenumber limit.

Figure 8 shows the relationship between the actual ɛ and mean of the dissipation estimates *μ*_{logɛ} in the form of a percentage error 100(logɛ − *μ*_{logɛ})/logɛ. It is clear that the estimated and actual values are in good agreement in the range from 10^{−9} to 10^{−5} m^{2} s^{−3}. Below 10^{−9} m^{2} s^{−3}, dissipation tends to be underestimated due to a lack of data in the viscous convective subrange, while above 10^{−5} m^{2} s^{−3}, dissipation tends to be overestimated, due to little data being in the roll-off region. Figure 8 shows that the systematic error in the mean of the low (i.e., less than 10^{−9}) dissipation estimates is more than 5% when *κ*_{B}/*κ*_{r} < 3.2, where *κ*_{r} = *L*^{−1}_{r}^{−5}) the systematic error is generally larger than 5%. There is a slight reduction in the error as the (segment) record length is increased. Systematic errors can be corrected using Fig. 8 for a probe having sampling characteristics similar to the one used in this paper. For other probes, the preceding method can be applied to estimate errors.

In the field, there is rarely an opportunity to average large numbers of dissipation estimates. In this case, we are concerned with the likely error or standard deviation of a single dissipation estimate. Figure 9 shows the highest random error occurs at high values of dissipation due to the roll-off region not being resolved. However, the error reduces as the record length increases. More specifically, the random error associated with the log dissipation estimates is more than 5% for *κ*_{B}/*κ*^{−1/3}_{r}

### c. Statistical stationarity

Dillon and Caldwell (1980) claimed that dissipation estimates in nonstationary regions would be biased toward the higher dissipation values in the sample. In other words, the estimated dissipation would be greater than the volume averaged dissipation of the sample. Here we examine this assertion using artificially generated nonstationary datasets. More specifically, 2048 sample time (or depth) series was generated such that each half of the dataset had different values of dissipation. Figure 10 shows spectra of two of these artificial time series. From examining Fig. 10 it is evident that the spectrum for the entire profile (i.e., 2048 sample dataset) basically follows the spectrum of the lower dissipation portion of the profile at low wavenumbers and vice versa at high wavenumbers. In general, the spectrum of a temperature gradient profile with several statistically stationary regions is equivalent to taking, at each wavenumber, the average of the spectra for each stationary region. This is one reason why measured temperature gradient spectra may not be close in shape to the theoretical Batchelor spectrum.

The ratio of the estimated dissipation ɛ_{B} and the volume averaged dissipation ɛ_{vol} was calculated for a number of the artificial nonstationary datasets. The results are plotted in Fig. 11 where it can be seen that dissipation is overestimated at low values of ɛ_{vol} and underestimated at high values of ɛ_{vol}. This occurs because the low ɛ_{vol} spectra are well resolved (i.e., the roll-off region is well defined) as in Fig. 10b, while the high ɛ_{vol} spectra are dominated by the viscous convective region (e.g., Fig. 10a). These results are consistent with Dillon and Caldwell’s (1980) assertion that dissipation estimates in nonstationary regions would be biased toward the higher dissipation values in the sample because they were only using well-resolved spectra.

The dissipation value at which we switch from underestimating to overestimating dissipation will depend upon the bandwidth of the measured spectra. In turn, the bandwidth will depend upon the sampling frequency and the record length *L*_{r}. The presence of finestructure and internal waves will affect the effective bandwidth due to the fitting algorithm rejecting the lower wavenumber portion of the spectrum; this will tend to result in dissipation being overestimated rather than underestimated.

The biasing of dissipation estimates in nonstationary regions highlights the importance of being able to divide or segment data into stationary regions prior to estimating dissipation. To demonstrate that such segmentation is viable we have applied the segmentation algorithm of Imberger and Ivey (1991) to one of the artificially generated nonstationary temperature gradient datasets (see Fig. 12). The segmentation parameter shown in Fig. 12b has a large value wherever there is an appreciable change in the frequency content of the signal (the artificially generated dataset changes from ɛ = 10^{−6} to ɛ = 10^{−7} m^{2} s^{−3} at its midpoint). In particular, that the largest value of the segmentation parameter occurs at the midpoint transition. Thus, selecting a threshold of, say, 100 for the segmentation parameter would result in a segmentation boundary being placed at the transition, thereby dividing the data into stationary regions.

## 6. Conclusions

A relatively robust method for determining dissipation from Batchelor curve fitting has been described. The method is relatively robust with regard to determining and making use of the Batchelor portion of a temperature gradient spectrum and determining the variance associated with that portion. Overall, the use of the method to determine dissipation results in considerable time savings when compared to manual methods. The systematic and random errors associated with determining the dissipation have been shown to be a function of the dissipation and the record length. Of course, these errors are the best case, as the presence of internal waves, finestructure, noise and calibration errors will result in increased errors.

## Acknowledgments

We would like to thank Greg Ivey, Mike Barry, and the reviewers for providing constructive comments on this paper. Angelo Saggio provided the data shown and Carol Lam checked the coding of the algorithm. This work forms WRL Reference 198 and CWR Reference ED648DL.

## REFERENCES

Abramowitz, M., and I. A. Stegun, 1970:

*Handbook of Mathematical Functions.*Dover Publications, 1046 pp.Batchelor, G. K., 1959: Small-scale variations of convected quantities like temperature in turbulent fluid. Part I: General discussion and the case of small conductivity.

*J. Fluid Mech.,***5,**113–133.Bendat, J. S., and A. G. Piersol, 1986:

*Random Data—Analysis and Measurement Procedures.*2d ed. Wiley-Interscience, 566 pp.De Silva, I. P. D., J. Imberger, and G. I. Ivey, 1997: Localised mixing due to a breaking internal wave ray at a sloping bed.

*J. Fluid Mech.,***350,**1–27.Dillon, T. M., 1982: Vertical overturns: A comparison of Thorpe and Ozmidov length scales.

*J. Geophys. Res.,***87**(C12), 9601–9613.——, and D. R. Caldwell, 1980: The Batchelor spectrum and dissipation in the upper ocean.

*J. Geophys. Res.,***85,**1910–1916.Fofonoff, N. P., and R. C. Millard Jr., 1983: Algorithms for computation of fundamental properties of seawater. UNESCO Tech. Pap. in Marine Sciences 44, 53 pp.

Fozdar, F. M., G. J. Parker, and J. Imberger, 1985: Matching temperature and conductivity sensor response characteristics.

*J. Phys. Oceanogr.,***15,**1557–1569.Gibson, C. H., 1968a: Fine structure of scalar fields mixed by turbulence. Part I: Zero-gradient points and minimal-gradient surfaces.

*Phys. Fluids,***11,**2305–2315.——, 1968b: Fine structure of scalar fields mixed by turbulence. Part II: Spectral theory.

*Phys. Fluids,***11,**2305–2315.——, and W. H. Schwartz, 1963: The universal equilibrium spectra of turbulent velocity and scalar fields.

*J. Fluid Mech.,***16,**365–384.Grant, H. L., B. A. Hughes, W. M. Vogel, and A. Moilliet, 1968: The spectrum of temperature fluctuations in turbulent flow.

*J. Fluid Mech.,***34,**423–442.Gregg, M. C., 1977a: A comparison of finestructure spectra from the main thermocline.

*J. Phys. Oceanogr.,***7,**33–40.——, 1977b: Variations in the intensity of small scale mixing in the main thermocline.

*J. Phys. Oceanogr.,***7,**436–454.——, 1999: Uncertainty and limitations in measuring ɛ and

*χ**T*.*J. Atmos. Oceanic Technol.,***16,**1483–1490.——, and T. B. Meagher, 1980: The dynamic response of glass rod thermistors.

*J. Geophys. Res.,***85**(C5), 2779–2786.——, C. S. Cox, and P. W. Hacker, 1973: Vertical microstructure measurements in the central North Pacific.

*J. Phys. Oceanogr.,***3,**458–469.——, E. A. D’Asaro, T. J. Shay, and N. Larson, 1986: Observations of persistent mixing and near-inertial waves.

*J. Phys. Oceanogr.,***16,**856–884.——, D. P. Winkel, and T. B. Sanford, 1991: Dissipation and the vertical wavenumber spectra of internal waves in the oceanic thermocline. Preprints,

*Eighth Conf. on Atmospheric and Oceanic Waves and Stability,*Denver, CO, Amer. Meteor. Soc., 257–260.Imberger, J., 1985: The diurnal mixed layer.

*Limnol. Oceanogr.,***30**(4), 737–770.——, and B. Boashash, 1986: Application of the Wigner–Ville distribution to temperature gradient microstructure: A new technique to study small-scale variations.

*J. Phys. Oceanogr.,***16,**1997–2012.——, and G. N. Ivey, 1991: On the nature of turbulence in a stratified fluid. Part II: Application to lakes.

*J. Phys. Oceanogr.,***21,**659–679.Ivey, G. N., and J. Imberger, 1991: On the nature of turbulence in a stratified fluid. Part I: The energetics of mixing.

*J. Phys. Oceanogr.,***21,**650–658.Luketina, D. A., 1987: Frontogenesis of freshwater overflows. Ph.D. thesis, University of Western Australia, 139 pp. [Available from Centre for Water Research, University of Western Australia, Perth 6907, Australia.].

——, and J. Imberger, 1989: Turbulence and entrainment in a buoyant surface plume.

*J. Geophys. Res.,***94**(C9), 12 619–12 636.Monin, A. S., and A. M. Yaglom, 1975:

*Statistical Fluid Mechanics.*Vol. 2. 3d ed. MIT Press, 874 pp.Oakey, N. S., 1982: Determination of the rate of dissipation of turbulent kinetic energy from simultaneous temperature and velocity shear microstructure measurements.

*J. Phys. Oceanogr.,***12,**256–271.Pedlosky, J., 1987:

*Geophysical Fluid Dynamics.*Springer-Verlag, 710 pp.Press, W. H., B. P. Flannery, S. A. Teukolsky, and W. T. Vettering, 1989:

*Numerical Recipes—The Art of Scientific Computing (FORTRAN Version).*Cambridge University Press, 702 pp.Ruddick, B., A. Anis, and K. Thompson, 2000: Maximum likelihood spectral fitting: The Batchelor spectrum.

*J. Atmos. Oceanic Technol.,***17,**1541–1555.Saggio, A., and J. Imberger, 2000: Mixing and turbulent fluxes in the metalimnion of a stratified lake.

*J. Geophys. Res.,*in press.Sherman, J. T., and R. E. Davis, 1995: Observations of temperature microstructure in NATRE.

*J. Phys. Oceanogr.,***25,**1913–1929.Stewart, R. W., and H. L. Grant, 1962: Determination of the rate of dissipation of turbulent kinetic energy near the sea surface in the presence of waves.

*J. Geophys. Res.,***67,**3177–3180.Taylor, G. I., 1938: The spectrum of turbulence.

*Proc. Roy. Soc. London,***164A,**476–490.Teoh, S. G., G. I. Ivey, and J. Imberger, 1997: Laboratory study of the interactions between two internal wave rays.

*J. Fluid Mech.,***336,**91–122.Townsend, A. A., 1976:

*The Structure of Turbulent Shear Flow.*Cambridge University Press, 429 pp.

(a) Statically stable temperature profiles 1 m long with a 1°C change from top to bottom that satisfy Eq. (1) with *γ* = 0 (thin line), *γ* = 0.5 (thick line), and *γ* = 2 (dashed line), and (b) the corresponding spectra of the temperature gradients.

Citation: Journal of Atmospheric and Oceanic Technology 18, 1; 10.1175/1520-0426(2001)018<0100:DTKEDF>2.0.CO;2

(a) Statically stable temperature profiles 1 m long with a 1°C change from top to bottom that satisfy Eq. (1) with *γ* = 0 (thin line), *γ* = 0.5 (thick line), and *γ* = 2 (dashed line), and (b) the corresponding spectra of the temperature gradients.

Citation: Journal of Atmospheric and Oceanic Technology 18, 1; 10.1175/1520-0426(2001)018<0100:DTKEDF>2.0.CO;2

(a) Statically stable temperature profiles 1 m long with a 1°C change from top to bottom that satisfy Eq. (1) with *γ* = 0 (thin line), *γ* = 0.5 (thick line), and *γ* = 2 (dashed line), and (b) the corresponding spectra of the temperature gradients.

Citation: Journal of Atmospheric and Oceanic Technology 18, 1; 10.1175/1520-0426(2001)018<0100:DTKEDF>2.0.CO;2

A schematic showing the various stages in fitting a Batchelor spectrum (dark line) to a measured spectrum (gray line) to determine the kinetic energy dissipation ɛ. The dashed vertical line indicates the noise wavenumber *κ*_{n}. (a) The potential range of the wavenumbers *κ*_{1} and *κ*_{2} are indicated for a Batchelor spectrum being fitted during the first iteration. Following the first iteration, the wavenumber *κ*_{L} is determined and the noise wavenumber *κ*_{n} modified as shown in (b). (c) The potential range of the wavenumbers *κ*_{1} and *κ*_{2} are indicated for a Batchelor spectrum being fitted during the second iteration. Following the second iteration, the wavenumber *κ*_{L} is modified as shown in (d). Subsequent iterations repeat the procedure shown in (c) and (d), resulting in the wavenumber *κ*_{L} being progressively modified.

A schematic showing the various stages in fitting a Batchelor spectrum (dark line) to a measured spectrum (gray line) to determine the kinetic energy dissipation ɛ. The dashed vertical line indicates the noise wavenumber *κ*_{n}. (a) The potential range of the wavenumbers *κ*_{1} and *κ*_{2} are indicated for a Batchelor spectrum being fitted during the first iteration. Following the first iteration, the wavenumber *κ*_{L} is determined and the noise wavenumber *κ*_{n} modified as shown in (b). (c) The potential range of the wavenumbers *κ*_{1} and *κ*_{2} are indicated for a Batchelor spectrum being fitted during the second iteration. Following the second iteration, the wavenumber *κ*_{L} is modified as shown in (d). Subsequent iterations repeat the procedure shown in (c) and (d), resulting in the wavenumber *κ*_{L} being progressively modified.

A schematic showing the various stages in fitting a Batchelor spectrum (dark line) to a measured spectrum (gray line) to determine the kinetic energy dissipation ɛ. The dashed vertical line indicates the noise wavenumber *κ*_{n}. (a) The potential range of the wavenumbers *κ*_{1} and *κ*_{2} are indicated for a Batchelor spectrum being fitted during the first iteration. Following the first iteration, the wavenumber *κ*_{L} is determined and the noise wavenumber *κ*_{n} modified as shown in (b). (c) The potential range of the wavenumbers *κ*_{1} and *κ*_{2} are indicated for a Batchelor spectrum being fitted during the second iteration. Following the second iteration, the wavenumber *κ*_{L} is modified as shown in (d). Subsequent iterations repeat the procedure shown in (c) and (d), resulting in the wavenumber *κ*_{L} being progressively modified.

Finding the deviation between a data point on the normalized measured spectrum and the normalized theoretical spectrum along a slope of −1.

Finding the deviation between a data point on the normalized measured spectrum and the normalized theoretical spectrum along a slope of −1.

Finding the deviation between a data point on the normalized measured spectrum and the normalized theoretical spectrum along a slope of −1.

Measured temperature gradient spectra *D*(*κ*) (solid line) and fitted theoretical Batchelor spectra *S*(*κ*) (dashed line) for data from a single profile in Lake Kinneret, Israel. The portion of the measured spectra that was used for fitting is between the two gray vertical lines. (a) The measured spectrum follows the theoretical form fairly well. (c) The curve fitting algorithm has ignored finescale/internal wave contamination at the low wavenumbers. In both cases the algorithm has successfully fitted the Batchelor spectrum despite the presence of high frequency noise. Note that the noise decreases at the very high wavenumbers due to the use of a brickwall filter on the time series prior to the spectra being calculated. The fitting error *δ*_{e} as a function of dissipation ɛ for cases (a) and (c) is shown in (b) and (d), respectively. In both cases there are clear minima in the error function.

Measured temperature gradient spectra *D*(*κ*) (solid line) and fitted theoretical Batchelor spectra *S*(*κ*) (dashed line) for data from a single profile in Lake Kinneret, Israel. The portion of the measured spectra that was used for fitting is between the two gray vertical lines. (a) The measured spectrum follows the theoretical form fairly well. (c) The curve fitting algorithm has ignored finescale/internal wave contamination at the low wavenumbers. In both cases the algorithm has successfully fitted the Batchelor spectrum despite the presence of high frequency noise. Note that the noise decreases at the very high wavenumbers due to the use of a brickwall filter on the time series prior to the spectra being calculated. The fitting error *δ*_{e} as a function of dissipation ɛ for cases (a) and (c) is shown in (b) and (d), respectively. In both cases there are clear minima in the error function.

Measured temperature gradient spectra *D*(*κ*) (solid line) and fitted theoretical Batchelor spectra *S*(*κ*) (dashed line) for data from a single profile in Lake Kinneret, Israel. The portion of the measured spectra that was used for fitting is between the two gray vertical lines. (a) The measured spectrum follows the theoretical form fairly well. (c) The curve fitting algorithm has ignored finescale/internal wave contamination at the low wavenumbers. In both cases the algorithm has successfully fitted the Batchelor spectrum despite the presence of high frequency noise. Note that the noise decreases at the very high wavenumbers due to the use of a brickwall filter on the time series prior to the spectra being calculated. The fitting error *δ*_{e} as a function of dissipation ɛ for cases (a) and (c) is shown in (b) and (d), respectively. In both cases there are clear minima in the error function.

(a) The measured spectrum does not follow the theoretical form very well. Consequently, the corresponding error function shown in (b) is flatter in the region of the minima, and the values of the minima are larger than that for the more Batchelor shaped measured spectra of Fig. 5. (c) There is no discernable Batchelor shape. As a result, the the error function shown in (d) is lacking the smoother shape shown in error graphs of Fig. 5. Further, the minimum error is relatively high. See Fig. 5 for explanation of the different lines and symbols.

(a) The measured spectrum does not follow the theoretical form very well. Consequently, the corresponding error function shown in (b) is flatter in the region of the minima, and the values of the minima are larger than that for the more Batchelor shaped measured spectra of Fig. 5. (c) There is no discernable Batchelor shape. As a result, the the error function shown in (d) is lacking the smoother shape shown in error graphs of Fig. 5. Further, the minimum error is relatively high. See Fig. 5 for explanation of the different lines and symbols.

(a) The measured spectrum does not follow the theoretical form very well. Consequently, the corresponding error function shown in (b) is flatter in the region of the minima, and the values of the minima are larger than that for the more Batchelor shaped measured spectra of Fig. 5. (c) There is no discernable Batchelor shape. As a result, the the error function shown in (d) is lacking the smoother shape shown in error graphs of Fig. 5. Further, the minimum error is relatively high. See Fig. 5 for explanation of the different lines and symbols.

Band-averaged spectra corresponding to (a) artificially generated dataset 2048 samples long, the first half of the dataset has ɛ = 10^{−4} m^{2} s^{−3}, while the second half of the dataset has ɛ = 10^{−5} m^{2} s^{−3}; (b) is similar except that ɛ = 10^{−10} m^{2} s^{−3} and 10^{−11} m^{2} s^{−3} for the first and second parts of the dataset. The different lines and symbols are explained in Fig. 5.

Band-averaged spectra corresponding to (a) artificially generated dataset 2048 samples long, the first half of the dataset has ɛ = 10^{−4} m^{2} s^{−3}, while the second half of the dataset has ɛ = 10^{−5} m^{2} s^{−3}; (b) is similar except that ɛ = 10^{−10} m^{2} s^{−3} and 10^{−11} m^{2} s^{−3} for the first and second parts of the dataset. The different lines and symbols are explained in Fig. 5.

Band-averaged spectra corresponding to (a) artificially generated dataset 2048 samples long, the first half of the dataset has ɛ = 10^{−4} m^{2} s^{−3}, while the second half of the dataset has ɛ = 10^{−5} m^{2} s^{−3}; (b) is similar except that ɛ = 10^{−10} m^{2} s^{−3} and 10^{−11} m^{2} s^{−3} for the first and second parts of the dataset. The different lines and symbols are explained in Fig. 5.

Systematic error in estimates of log dissipation as a function of (a) Batchelor wavenumber *κ*_{B} (based on actual dissipation) and record wavenumber *κ*_{r}, which is the reciprocal of the record length and, (b) actual dissipation and record length *L*_{r}.

Systematic error in estimates of log dissipation as a function of (a) Batchelor wavenumber *κ*_{B} (based on actual dissipation) and record wavenumber *κ*_{r}, which is the reciprocal of the record length and, (b) actual dissipation and record length *L*_{r}.

Systematic error in estimates of log dissipation as a function of (a) Batchelor wavenumber *κ*_{B} (based on actual dissipation) and record wavenumber *κ*_{r}, which is the reciprocal of the record length and, (b) actual dissipation and record length *L*_{r}.

Absolute value of the random error (standard deviation) in estimates of log dissipation as a function of (a) Batchelor wavenumber *κ*_{B} (based on actual dissipation); record wavenumber *κ*_{r}, which is the reciprocal of the record length; and (b) actual dissipation and record length *L*_{r}.

Absolute value of the random error (standard deviation) in estimates of log dissipation as a function of (a) Batchelor wavenumber *κ*_{B} (based on actual dissipation); record wavenumber *κ*_{r}, which is the reciprocal of the record length; and (b) actual dissipation and record length *L*_{r}.

Absolute value of the random error (standard deviation) in estimates of log dissipation as a function of (a) Batchelor wavenumber *κ*_{B} (based on actual dissipation); record wavenumber *κ*_{r}, which is the reciprocal of the record length; and (b) actual dissipation and record length *L*_{r}.

Band-averaged spectra corresponding to (a) artificially generated dataset 2048 samples long (solid line), the first half of the dataset (line with short dashes), which has ɛ = 10^{−4} m^{2} s^{−3} and the second half of the dataset (line with long dashes), which ɛ = 10^{−5} m^{2} s^{−3}; (b) is similar except that ɛ = 10^{−10} m^{2} s^{−3} (line with short dashes) and ɛ = 10^{−11} m^{2} s^{−3} (line with long dashes).

Band-averaged spectra corresponding to (a) artificially generated dataset 2048 samples long (solid line), the first half of the dataset (line with short dashes), which has ɛ = 10^{−4} m^{2} s^{−3} and the second half of the dataset (line with long dashes), which ɛ = 10^{−5} m^{2} s^{−3}; (b) is similar except that ɛ = 10^{−10} m^{2} s^{−3} (line with short dashes) and ɛ = 10^{−11} m^{2} s^{−3} (line with long dashes).

Band-averaged spectra corresponding to (a) artificially generated dataset 2048 samples long (solid line), the first half of the dataset (line with short dashes), which has ɛ = 10^{−4} m^{2} s^{−3} and the second half of the dataset (line with long dashes), which ɛ = 10^{−5} m^{2} s^{−3}; (b) is similar except that ɛ = 10^{−10} m^{2} s^{−3} (line with short dashes) and ɛ = 10^{−11} m^{2} s^{−3} (line with long dashes).

Ratio of estimated dissipation ɛ_{B} to actual (volume averaged) ɛ_{vol} dissipation as a function of ɛ_{vol} when the data are nonstationary.

Ratio of estimated dissipation ɛ_{B} to actual (volume averaged) ɛ_{vol} dissipation as a function of ɛ_{vol} when the data are nonstationary.

Ratio of estimated dissipation ɛ_{B} to actual (volume averaged) ɛ_{vol} dissipation as a function of ɛ_{vol} when the data are nonstationary.

Segmentation of a nonstationary dataset: (a) an artificially generated temperature gradient dataset 2048 samples long where the upper and lower halves have ɛ = 10^{−6} m^{2} s^{−3} and ɛ = 10^{−7} m^{2} s^{−3}, respectively; and (b) the corresponding segmentation parameter calculated using the method of Imberger and Ivey (1991). The dashed line in (b) shows an arbitrary theshold of 100, which would result in a segmentation boundary being placed at the transition from ɛ = 10^{−6} to 10^{−7} m^{2} s^{−3}.

Segmentation of a nonstationary dataset: (a) an artificially generated temperature gradient dataset 2048 samples long where the upper and lower halves have ɛ = 10^{−6} m^{2} s^{−3} and ɛ = 10^{−7} m^{2} s^{−3}, respectively; and (b) the corresponding segmentation parameter calculated using the method of Imberger and Ivey (1991). The dashed line in (b) shows an arbitrary theshold of 100, which would result in a segmentation boundary being placed at the transition from ɛ = 10^{−6} to 10^{−7} m^{2} s^{−3}.

Segmentation of a nonstationary dataset: (a) an artificially generated temperature gradient dataset 2048 samples long where the upper and lower halves have ɛ = 10^{−6} m^{2} s^{−3} and ɛ = 10^{−7} m^{2} s^{−3}, respectively; and (b) the corresponding segmentation parameter calculated using the method of Imberger and Ivey (1991). The dashed line in (b) shows an arbitrary theshold of 100, which would result in a segmentation boundary being placed at the transition from ɛ = 10^{−6} to 10^{−7} m^{2} s^{−3}.