MARCH 2020 JABBARI ET AL.

517

Dissipation of Turbulent Kinetic Energy in the Oscillating Bottom Boundary

Layer of a Large Shallow Lake

AIDIN JABBARIa AND LEON BOEGMAN

Environmental Fluid Dynamics Laboratory, Department of Civil Engineering, Kingston, Ontario, Canada

REZA VALIPOUR
Water Science and Technology, Canada Centre for Inland Waters, Environment and Climate Change Canada,
Burlington, Ontario, Canada

DANIELLE WAIN
7 Lakes Alliance, Belgrade Lakes, and Colby College, Waterville, Maine

DAMIEN BOUFFARD
Surface Waters—Research and Management, Swiss Federal Institute of Aquatic Science and Technology (EAWAG),
Kastanienbaum, Switzerland

(Manuscript received 12 May 2019, in final form 8 January 2020)

ABSTRACT

Mixing rates and biogeochemical fluxes are commonly estimated from the rate of dissipation of tur-
bulent kinetic energy « as measured with a single instrument and processing method. However, differ-
ences in measurements of « between instruments/methods often vary by one order of magnitude. In an
effort to identify error in computing «, we have applied four common methods to data from the bottom
boundary layer of Lake Erie. We applied the second-order structure function method (SFM) to velocity
measurements from an acoustic Doppler current profiler, using both canonical and anisotropy-adjusted
Kolmogorov constants, and compared the results with those computed from the law of the wall, Batchelor
fitting to temperature gradient microstructure, and inertial subrange fitting to acoustic Doppler velo-
cimeter data. The « from anisotropy-adjusted constants in SFM increased by a factor of 6 or more at 0.2 m
above the bed and showed a better agreement with microstructure and inertial method estimations. The
maximum difference between SFM «, computed using adjusted and canonical constants, and micro-
structure values was 25% and 50%, respectively. This difference was 30% and 55%, respectively, for those
from inertial subrange fitting at times of high-intensity turbulence (Reynolds number at 1 m above the bed
of more than 2 3 104). Comparison of the SFM « to those from law of the wall was often poor, with errors
as large as one order of magnitude. From the considerable improvement in « estimates near the bed,
anisotropy-adjusted Kolmogorov constants should be applied to compute dissipation in geophysical
boundary layers.

1. Introduction and Madsen 1986; Boegman and Ivey 2009), and sedi-
ment water interface biogeochemistry (e.g., Scalo et al.
Turbulence in the bottom boundary layers (BBLs) of
2012; McGinnis et al. 2014; Schwefel et al. 2017). BBL
lakes and oceans regulates basin-scale mass flux (e.g., Munk
turbulence is typically characterized according to the
1966; Imberger 1998), sediment resuspension (e.g., Grant
rate of dissipation of turbulent kinetic energy «, which
enables parameterization of turbulent mixing and bio-
a
Current affiliation: Physical Ecology Laboratory, Department of geochemical fluxes (e.g., Lorke 2007; Bouffard et al.
Integrative Biology, University of Guelph, Guelph, Ontario, Canada. 2013). Precise calculation of « (i.e., « 5 n›ui /›xj ›uj /›xi ,
where n is the kinematic viscosity and ui is the velocity
Corresponding author: Aidin Jabbari, 0saj@queensu.ca component in the direction i and the overbar denotes

DOI: 10.1175/JTECH-D-19-0083.1
Ó 2020 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright
Policy (www.ametsoc.org/PUBSReuseLicenses).
518 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 37

FIG. 1. Time series of the dissipation from TMM (s), IDM [from Nortek (m) and RDI (j)
measurements], LES of the oscillating flow (solid line), IDM (from LES calculations; dashed–
dotted line), and log law (from LES calculations; dashed line) at a height of 1 m above the
sediment for Lake Alpnach (13–14 Aug 2002). Field data are from Lorke and Wüest (2005).
Modified from Jabbari (2015).

averaging with respect to the time) requires simulta- (TMM), and log law. Typically, a single method is
neous measurement of the instantaneous velocity gra- applied in isolation, and so the present objective is to
dient tensor, which is observationally infeasible (e.g., reduce error in computation of «, through inter-
Doron et al. 2001). Therefore, dissipation is usually comparison, so the various methods will converge
estimated from indirect methods with the exact ob- toward a true estimate of «. This will allow present-day
served value remaining unknown due to measurement benchmarks to be set.
and methodological error. For example, typically ap- The SFM and IDM rely on measurement of turbu-
plied methods to compute « assume the flow is steady lent velocity fluctuations, which may be affected by
and the turbulence is isotropic and homogeneous (e.g., anisotropy in the boundary layer. They are based on
Doron et al. 2001), conditions that are rarely met in Kolmogorov’s second similarity hypothesis that re-
geophysical boundary layers. lates spatial velocity correlations to the dissipation in
These errors can be seen in data from Lake Alpnach the inertial subrange where the eddies scale with the
(Fig. 1), where « at 1 m above the bed from Batchelor distance from the wall (log layer; Townsend 1976).
fitting to temperature microstructure are compared These methods have the advantage of generating long-
to those from inertial fitting applied to data from two term ‘‘instantaneous’’ dissipation time series (IDM)
acoustic Doppler current profilers (ADCPs) (Lorke and profiles (SFM) (e.g., Lorke 2007). However, the
and Wü est 2005). Also shown are « calculated from anisotropic nature of turbulent flows within boundary
wall-resolved large-eddy simulations (LESs) of an layers limits their applicability (Monin and Yaglom
oscillating flow with similar Reynolds number, as well 1975). Jabbari et al. (2015, 2016) showed that the
as « computed from both log law and inertial fitting Kolmogorov constants, used to fit to the theoretical
applied to the LES data (Jabbari 2015). Order of energy spectrum (IDM) and to spatial correlations of
magnitude discrepancies are evident between the velocity (SFM), both in the inertial subrange, change
different calculations (mean standard deviation of substantially from the canonical values within the
2 3 1029 W kg21) resulting from methodological er- boundary layer, where near-wall anisotropy becomes
rors (« from log law and inertial fits applied to LES) significant.
and methodological plus measurement errors (« from The field-based IDM data shown in Fig. 1 also has
field instruments). error, because the ADCP data did not resolve along-
To better quantify error in computation of «, this and cross-stream velocity spectra, leading to an un-
study investigates the factors contributing to the dis- certainty in « up to a factor of (4/3)23/2 ’ 0.65, due to
crepancies between « estimated from four commonly anisotropy (Lorke and Wüest 2005). Moreover, these
applied field methods: the inertial fitting dissipation data were subject to inhomogeneity, from the beam-
method (IDM), structure function method (SFM), spreading angle of 1.58 and system spatial resolution
Batchelor fitting to temperature microstructure method (width of the acoustic beam at a particular depth bin),
MARCH 2020 JABBARI ET AL. 519

FIG. 2. (a) Map of Lake Erie and its bathymetry. The square shows the location of the
mooring: Sta. 341, 418470 N, 828160 W. Bathymetric contours are in meters and the axis is based
on Universal Transverse Mercator coordinate system (UTM) in the zone 17-North. (b) The
tripod equipped with ADCPs, an ADV, and RBR TR-1060 s before deployment on the bottom
at Sta. 341. Modified from Valipour et al. (2015a).

and unsteadiness in the turbulence field, which con- computed using both the canonical (e.g., Wiles et al.
tribute to uncertainty in « calculations. For example, 2006; Monin and Yaglom 1975) and anisotropy-adjusted
Lorke (2007) found that SFM-ADCP and IDM acoustic (Jabbari et al. 2016) Kolmogorov constants. This is the
Doppler velocimeter (ADV) computed « was highly first application of these numerically derived anisotropy
affected by the amount of temporal and spatial averag- corrections to field data. The calculated « values were
ing in nonstationary turbulence. also compared to those from the log-law and Batchelor
It is, therefore, not surprising that the log law often fits to temperature gradient microstructure, which were
gives poor results when applied to unsteady boundary not affected by anisotropy.
layer flows (e.g., Grant and Madsen 1979; Lorke et al.
2002); although logarithmic velocity profiles are often
2. Study site and measurements
observed (e.g., Valipour et al. 2015a; Troy et al. 2016).
For example, Huntley and Hazen (1988) and Huntley Lake Erie (Fig. 2; 388 km long and 92 km wide) is the
(1988) found that the log law underestimates the fric- shallowest of the Laurentian Great Lakes and consists
tion velocity in combined wave and steady flow con- of distinct western, central, and eastern basins, which
ditions, where the wave flow enhances the bottom have maximum depths of 11, 25, and 64 m, respec-
shear stress. Conversely, in Lake Michigan Cannon tively. In the summers of 2008 and 2009, extensive field
and Troy (2018) found ‘‘remarkable’’ agreement be- measurements (Bouffard et al. 2012; Bouffard and
tween SFM and log-law «, with log law becoming a Boegman 2013; Bouffard et al. 2013; Valipour et al.
much less reliable predictor at low current speeds. 2015a,b) were carried out in the central basin to ad-
These field data may be contrast with numerical data dress the hypoxia problem (e.g., Rao et al. 2008; Scavia
(Fig. 1), showing log law to provide—at selected times et al. 2014). As part of the measurements, a 1.8 m tripod
other than the flow reversal—a better estimate of (Figs. 2a,b) was deployed at Sta. 341a depth of 17.5 m.
« than the IDM, in comparison to the LES calculations. The tripod was equipped with a downward-looking
Given the discrepancies in « estimates, resulting from 2 MHz pulse coherent acoustic Doppler current pro-
both the measurement and methodological errors, filer (HR-ADCP, Nortek with accuracy 61% of
described above, there is a need for further inter- measured values) at 1.84 m above the bed, that mea-
comparison of « estimates. sured the velocity down to ;3 cm above the bed,
In the present study, we compared estimates of « from and a Nortek Vector ADV at 1 m above the bed. At
four different methods (ADV-IDM, ADCP-SFM, log 15 min intervals, the HR-ADCP burst recorded ve-
law, and TMM) in the bottom boundary of a large lake, locity at 1 Hz over 256 s in 3 cm bins to the sediment–
in an effort to converge on the most accurate dissipation water interface. The ADV measured the velocity at
estimate and assess what a reasonable error should be in 16 Hz for 5 min every 20 min (Fig. 2b). Here, dissi-
calculating «. To account for anisotropy, the SFM was pation profiles were calculated by applying the SFM
520 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 37

and log law to the HR-ADCP measurements and the velocity based on the Kolmogorov turbulent cascade
IDM to the ADV data. theory in the inertial subrange
Since the adjustments in dissipation methods we ap-
plied are based on neutrally stratified flows, we minimize D(z, r) 5 C«2/3 r2/3 , (2)
the contribution of density stratification on anisotropy
where C is the Kolmogorov 2/3 constant (Pope 2000),
near the bed (Sarkar 2003) by concentrating on data
which in atmospheric studies has been found to be
collected between 3 and 6 May 2009 (days of year
between 2.0 and 2.2 (Sauvageot 1992). Typically, the
122.23–125). At this time, the flow was weakly stratified
dissipation «SFC is computed with the canonical value
with the surface seiche (;14 h period) being the pre-
C 5 CC 5 2.1 (e.g., Wiles et al. 2006; Lorke 2007;
dominant process that energized the bottom boundary
Bouffard et al. 2013). These dissipation values dif-
layer (e.g., Boegman et al. 2001; Rao et al. 2008). The
fer from those («SFA) calculated, as described below,
significant period of surface waves was less than the
using the anisotropy-adjusted Kolmogorov constants CA
theoretical threshold (;5 s; Valipour et al. 2017) re-
[Eq. (3) and Fig. 3a; Jabbari et al. 2016], which have been
quired for orbital velocities to reach the bottom.
adjusted for turbulence anisotropy through the bound-
The dissipation from velocity measurements were
ary layer. The latter constants were computationally
compared with those from Batchelor fitting to tem-
evaluated over 20 # z1 # 2000, where z1 5 zut/y is the
perature gradient microstructure measured with a
height above the bed normalized by the friction velocity
self-contained autonomous microstructure profiler
ut and kinematic viscosity. The constant CA was given by
(SCAMP; PME Inc.; Bouffard and Boegman 2013) on
the fitting functions
28 May and 2 July 2009 (days 147.13 and 182.64, re-
8
spectively) when the flow was weakly stratified. This >
> 0:048(z1 ) ,
0:65
z1 # 160
>
>
subsample was picked from 26 days of measurements >
>
>
> 0:42 1 0:45 ln[(z1 ) ],
0:4
in July and August 2008–09, when the boat was >
> 160 , z1 # 500
>
>
moored within 30 m of Sta. 341 (to prevent entan- >
< 0:44
glement) for ;1 h capturing 5–12 microstructure casts CA 5 1:118 1 , 500 , z1 , 2000 ,
>
> 0:5 1 24(z 1 )20:61
on each day. The SCAMP profiled vertically through >
>
>
>
the water column at a speed of 0.1 m s21 in a falling >
>
> 0:699 1 0:455 ln[(z1 )
0:295
], 2000 # z1 # 16 200
>
>
mode and sampled at 100 Hz with a time response of >
>
:
7 ms, resolving water column structure with vertical 2, z1 . 16 200
scales as small as 1 mm. Batchelor fits, to estimate (3)
dissipation, were performed in 0.25 m bins through
the water column up to 1 m above the bed. The sensor where CA for z1 # 2000 was obtained from numerical
guard prevented profiling all the way to the sedi- simulation of a steady turbulent open channel flow
ments. These results have been published, with the (Jabbari et al. 2016) and has been extrapolated to z1 $
methodology and data described in detail in Bouffard 2000 using CA 5 2 from Saddoughi and Veeravalli’s
and Boegman (2013). (1994) experimental results that found CA 5 2.0 6 0.1
for z1 5 16 200 and 62 000.
3. Methodology The dissipation was estimated by fitting the measured
structure function from the HR-ADCP [D; Eq. (1)] to
a. Structure function method Eq. (2) within the inertial subrange and beam averaging.
We applied the second-order structure function Individual fits, at each height, were over 0.03 , r/2 ,
0.18 m, where the maximum separation distance was six
D(z, r) 5 [u0b (z 1 r) 2 u0b (z)]2 , (1) bins (Lorke 2007) and the minimum range of r was given
by the distance between two adjacent bins (i.e., 2 times
which is the correlation of the in-beam fluctuation the bin size of 0.03 m). The bottom flow velocity and the
velocity u0b , (obtained by Reynolds averaging each friction velocity were ,0.1 and ,0.01 m s21, respectively.
256 s burst) between two points separated by a beam
b. Inertial dissipation method
distance r. In Eq. (1) D(z, r) represents the structure
function at height z above the bed. Here, the centered Similar to the structure function method, the iner-
difference scheme described by Wiles et al. (2006) was tial dissipation method is based on fitting the energy
applied. In isotropic turbulent flows, at high Reynolds spectrum E to the theoretical form within the inertial
numbers, the SFM relates the rate of dissipation of subrange. From Kolmogorov’s universal equilibrium
turbulent kinetic energy to the spatial correlations of hypothesis, at sufficiently high Reynolds number, the
MARCH 2020 JABBARI ET AL. 521

FIG. 3. (a) The vertical velocity component Kolmogorov 2/3 constant adjusted for anisotropy obtained from
numerical simulation CA. Circles are from numerical analysis and the solid line is from the fitting function [Eq. (3)].
The data in z1 5 16 000 are from Saddoughi and Veeravalli (1994). The dashed line shows the canonical constant
CC 5 2.1. (b) An example of the burst-averaged along-beam structure function [Eq. (1)] from the three HR-ADCP
beams at z1 5 2113 (z 5 1.32 m) on day 123.98. The dashed line shows a 2/3 slope.

 
turbulence is isotropic and the one-dimension longi- 1 z
uL 5 ut ln , (6)
tudinal and transverse spectra are k z0

E11 (k1 ) 5 a11 «IDM

2 /3
k125/3 (4) where k ’ 0.41 is the von Kármán constant and z0 the
roughness length, calculated by fitting the velocity pro-
and files to Eq. (6). The log-law dissipation is
4
E22 (k1 ) 5 E33 (k1 ) 5 a11 «IDM
2/3
k25
1
/3
, (5) «L 5
u3t
, (7)
3 kz
where bb 5 1, 2, 3 represents the streamwise, spanwise,
which is based on a local equilibrium between the dis-
and vertical (or x, y, z) directions, respectively, k1 is the
sipation and production P of turbulent kinetic energy in
longitudinal wavenumber, and a11 5 0.5 is the canonical
stationary flows (Lorke and Maclntyre 2009).
Kolmogorov 25/3 constant (Monin and Yaglom 1975).
To be consistent with the in-beam SFM calculations and d. Friction velocity
minimize noise, we used only the vertical velocity com-
The friction velocity was calculated from log law,
ponents from the ADV for calculation of spectra. The
which is common in field studies. Following Valipour
inertial subrange was considered the wavenumber range
et al. (2015a), we first located the logarithmic
with 20% error from the plateau of E33 (k1 )k15/3 , that is, the
constant-stress layer, by finding the depth range
region where E33 (k1 )k15/3 is independent of the wave-
over which
number and equal to 4/3a11«2/3 [Eq. (5); Saddoughi and
 
Veeravalli 1994]. For z1 , 160 the vertical constant ad- dut d du
justs with anisotropy (Jabbari et al. 2015); however, this 5 kz 5 0 6 0:1 s21 . (8)
dz dz dz
condition was not observed at 1 m above the bed; indi-
cating the flow at 1 m was isotropic. The local mean ve- Then ut and z0 were evaluated by least squares fitting
locity was used to relate the wavenumber and frequency Eq. (6) to the profiles of u over the constant-stress layer
spectra. (e.g., Kundu and Cohen 2002).
Deviation from log-law behavior will lead to error in
c. Logarithmic law of the wall
CA [i.e., Eq. (3)], through the usage of ut in computation
The observed velocity profiles (u) were compared of z1. To examine the accuracy of the log-law-based
with those from logarithmic law of the wall (log law; uL) friction velocity, we also computed ut by three other
for a steady turbulent wall flow methods, applied to the ADV data. In these methods,
522 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 37

FIG. 4. (a) Predicted relationship («p,1m; solid line) between the dissipation and the flow velocity at 1 m above the
bed with 95% error (dashed lines). (b) Ratio of the calculated dissipation from SFM («SFA,1m) to the predictions
from (a) («p,1m) vs velocity direction at 1 m above the bed (u1m).

the bed shear stress t was calculated from (i) the qua- with period T (Jabbari 2015). The numerical model
dratic law has been widely used for the simulation of similar
geophysical flows (e.g., Scalo et al. 2012; Yuan and
t 5 rCd U 2 (9) Piomelli 2014).

using a drag coefficient Cd 5 4.5 3 1023, as determined 4. Results

by Valipour et al. (2015a) for this site in Lake Erie
through log-law fits to an extended HR-ADCP dataset; a. Flow interference
(ii) the turbulent kinetic energy (TKE) The field data were first evaluated to identify when
spurious dissipation measurements resulted from
t 5 rCt w0 w0 , (10)
frame interference at 1 m above the bed, where the
where Ct 5 0.9 (Bluteau et al. 2016), and (iii) the ADV battery canister was located (Fig. 2; see also
Reynolds stress (e.g., Zulberti et al. 2018) Valipour et al. 2015a). As expected from theory (Pope
2000; Lorke and Maclntyre 2009) the observed dissi-
t 5 ru0 w0 . (11) pation correlated with the flow velocity. We followed
McGinnis et al. (2014) and related «SFA to the mean
The shear velocity was then computed from bed flow at 1 m above the bed (Fig. 4a). Dividing «SFA by
pffiffiffiffiffiffiffi
stress ut 5 t/r. the predicted dissipation from the correlation allowed
identification of the flow directions associated with
e. Numerical simulations
outlier data (Fig. 4b). The largest deviations occurred
The numerical data in this study are from LES of when flow was from 2758 to 3608, resulting in discrep-
turbulent flows on a smooth wall using a well-validated ancy of the dissipation by a factor of up to 23. The
numerical code (Keating et al. 2004a,b). The top and same results were achieved from correlation of dis-
the bottom boundaries were free-slip symmetry and sipation with the third power of velocity [e.g., Eq. (7);
no-slip boundary conditions, respectively, and periodic not shown here]. Therefore, HR-ADCP and ADV
conditions were applied in the streamwise and span- data from these directions were neglected in the
wise directions. The data here are from LES of a fully analysis.
developed turbulent and steady open channel flow with
b. Time variations of friction velocity and
Ret 5 Lzut/n 5 2000, where Lz 5 1 is the height of the
Kolmogorov constant
channel (Jabbari et al. 2015, 2016) and an oscillating flow
with zero-mean freestream current with Red 5 dU/n 5 In general, ut from all methods, were within one
3600, where U was ffi the maximum free freestream ve-
pffiffiffiffiffiffiffiffiffi standard deviation (2 3 1024 m s21) of the mean
locity, d 5 2n/v was the Stokes layer thickness, and (Fig. 5a). The value of ut from log-fits was within 6%,
v 5 2p/T was the angular frequency of the oscillation 16%, and 21% of that from the quadratic law, TKE,
MARCH 2020 JABBARI ET AL. 523

FIG. 5. (a) Flow velocity (m s21) at 1 m above the bed (dashed black line) and 10 times the friction velocity
calculated by log law [Eq. (8); solid black], the quadratic law [Eq. (9); blue], the TKE [Eq. (10); green], and the
Reynolds stress method [Eq. (11); red]. (b) Time series of dissipation of turbulent kinetic energy at 1 m above the
bed from the structure function method using the anisotropy-adjusted constants («SFA; filled circles) and canonical
constants («SFC; hollow circles), log law [«L; Eq. (7); solid line], and inertial dissipation method («IDM; red dashed–
dotted line). (c) Variation of the Reynolds number at 1 m above the bed (Re1m 5 u1m 3 1/n) and the ratio of shear
length scale to the Kolmogorov length scale B 5 Ls/h. The data with flow direction 2758–3608 were removed
because of the frame interference (see Fig. 3b).

and Reynolds stress methods, respectively, giving ,17% at 0.2 m above the bed, z1 5 50 and z1 5 320 on low
error in «SFA due to uncertainties in ut [based on a (day 123.98, ut 5 0.000 25 m s21, Figs. 6a,c) and high
maximum discrepancy of 22% in z1; Eqs. (2) and (day 123.07, ut 5 0.0016 m s21, Figs. 6d,f) turbulence
(3)]. These errors were similar to the 4%–11% error intensity days, with corresponding Kolmogorov con-
found by Zulberti et al. (2018) in applying the same stants (Fig. 3a) of CA 5 0.61 and 1.46, respectively.
methods to estimate bottom stress at more energetic The Kolmogorov constant at 1 m above the bed
ocean site. These uncertainties do not affect «SFC , changed from 1.41 on day 123.98 (z1 5 253, Figs. 6a,c)
which uses the canonical constant that is independent to 1.69 on day 123.07 (z1 5 1600, Figs. 6d,f).
of u t.
c. Dissipation of turbulent kinetic energy
The friction velocity was oscillatory, following the
barotropic mode-one seiche with a period of 14 h Time series of velocity and flow direction measured
(Fig. 5a), leading to periodicity in ut as is typically by the HR-ADCP are shown (Figs. 7a,b, respectively),
observed (e.g., Valipour et al. 2015a) and here ut ; along with the computed «SFA (Fig. 7c), «SFC (Fig. 7d),
0.07u1m (e.g., Lorke and Maclntyre 2009). The baro- and «L (Fig. 7e). The velocity and dissipation time
tropic flow caused time variations of z1, and conse- series exhibited an oscillatory pattern with a period
quently CA that varied with both time and height of ;14 h consistent with the surface seiche (Boegman
above the bed due to changes in z1. For instance, et al. 2001; Valipour et al. 2015a). From these figures,
524 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 37

FIG. 6. (a),(d),(g),(j) Velocity profiles from HR-ADCP (circles) compared with log law [uL; Eq. (6); solid line]. (b),(e),(h),(k) Reynolds
stress: ru0 w0 (filled circles) and ry 0 w0 (hollowed circles). Note the change in x-axis limits in these panels. (c),(f),(i),(l) Dissipation of
turbulent kinetic energy from structure function method using the anisotropy-adjusted constants («SFA; filled circles) compared with
those from canonical constants («SFC; hollow circles), inertial dissipation method («IDM; filled red circles), and log law [«L; Eq. (7);
solid line]. Also included in (i) and (l) is the dissipation from temperature microstructure method («TMM; dashed–dotted line). Rows
are from (a)–(c) day 123.98 (ut 5 0.000 25 m s21 and z0 5 0.003 m), (d)–(f) day 123.07 (ut 5 0.0016 m s21 and z0 5 0.002 m), (g)–(i) day
182.64 (ut 5 0.0032 m s21 and z0 5 0.0016 m), and (j)–(l) day 147.13 (ut 5 0.005 m s21 and z0 5 0.0019 m). In (c), (f), (i), and (l) the
width of the shaded area shows the difference between maximum and minimum dissipation predictions from the three beams using
canonical constants.

it is striking how «L  «SFA . «SFC, particularly during «SFA (9.55 3 1028 W kg21) compared to «SFC (3.02 3
periods of strong mean flow, when log-law behavior 1028 W kg21; Fig. 8a).
was expected, suggesting «L fundamentally overesti- During these measurements, at 1 m above the bed
mates dissipation at this site. That «SFA . «SFC, during (z1 . 160, Fig. 6c) where the canonical value of the
strong flow, shows how anisotropy near the bed en- vertical 25/3 Kolmogorov constant is valid, «IDM was
hanced dissipation. within ;30% and ;55% of «SFA and «SFC, respectively,
Close to boundary, the difference between «SFC and at times of high-intensity turbulence when the Reynolds
«SFA was .6 (e.g., «SFA/«SFC 5 6.8 at z1 5 50 or z 5 0.2 m, number was .2 3 104 (Re1m 5 u1m 3 1/n) (Figs. 5b,c).
Fig. 6c), but the values converged away from the bed The difference between SFM and IDM increased during
where CA approached CC (Fig. 3a) (e.g., «SFA/«SFC 5 1.82 flow reversal (Fig. 7a), when turbulence intensity
at z1 5 253 or z 5 1 m, Fig. 6c; «SFA/«SFC 5 1.16 at z1 5 was low and both methods were inaccurate (e.g.,
8000 or z 5 1.62 m, Fig. 6l). As a result, the average «IDM/«SFA 5 3 and «IDM/«SFC 5 5.5 at 1 m above the bed
value of the lognormal distribution of the dissipation on day 123.98, Fig. 6c). Comparison of the dissipation
within 0.5 m above the bed was 1.9 times greater for calculated from SFM with those from Batchelor fits to
MARCH 2020 JABBARI ET AL. 525

FIG. 7. Time series of the flow on days 122.23–125:(a) flow direction, (b) flow velocity, and log of the dissipation of turbulent kinetic
energy calculated from structure function based on (c) anisotropy-adjusted Kolmogorov constants [CA; Eq. (3)] and (d) canonical constant
(CC 5 2.1) as well as (e) that calculated from log law [Eq. (7)]. Note that the data with flow direction from 2758 to 3608 have been removed
from (b)–(e) (Fig. 3). The dissipation is in W kg21.

SCAMP data («TMM; Figs. 6i,l and Table 1) shows that the log law overestimated the dissipation close to the
«SFA had better agreement with «TMM than «SFC («SFA boundary (z , 0.5 m) on most days (Figs. 6f,i,l and 7e).
was within 25% of «TMM, while «SFC had differences This difference can be as high as a factor of 4 or more
up to 50%; «IDM was within 25% of «TMM at 1 m above (Fig. 6f) and may be due to the lack of equilibrium
the bed). between production and dissipation of turbulent ki-
Velocity profiles showed better logarithmic charac- netic energy, which is not always satisfied in bound-
teristics during high-intensity turbulence (Figs. 9b–d) ary layers of oscillatory seiche-induced lake flows
compared to those during low turbulence flow reversal (Lorke 2007; Lorke and Maclntyre 2009; Jabbari
(Fig. 9a). The dissipation from the log law showed 2015). Consequently, the mean value of the lognor-
discrepancies relative to that calculated from the other mal dissipation «L ; 2 3 1027 W kg 21 from the entire
methods (e.g., «L/«SFA 5 4.4 at z1 5 2592 or z 5 measurement region (0.18 , z , 1.65 m) was an order
1.62 m, Fig. 6f). The log law underestimated the dissi- of magnitude greater than those from «SFA 54.8 3
pation during the flow reversal (Fig. 6c), due to lami- 1028 W kg21 and «SFC 5 3.02 3 1028 W kg21 (Fig. 8b).
nar flow during these phases of oscillation. However, This difference was magnified near the bed (z , 0.5 m;
526 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 37

FIG. 8. Frequency distributions of dissipation rate (normalized by the total number of observations) from theh structure function based
on anisotropy-adjusted Kolmogorov constants [red circles; Eq. (3)], based on canonical constants (blue circles), and calculated from log
law (black circles) (a) within 0.5 m from the bed, (b) in the whole region (0.18 , z , 1.65 m), and (c) at 1 m above the bed. The frequency
distributions are calculated for log10(«) based on the entire dataset of 260 bursts at 15 min intervals (days of year 122.23–125 in 2009).
Solid lines show a fit to a lognormal distribution.

z1 , 2685), where the boundary leads to anisotropy scale of bed-induced mean-flow shear that drives near-
(Fig. 8a) and decreased where the flow became iso- bed turbulence. Here, u(z) is the burst-averaged mean
tropic (z 5 1.0 m; z1 , 5188), and the three methods flow velocity at height z (Fig. 11). For Res , 2 3 104 (z ,
predicted very similar dissipation (Fig. 8c). 0.5 m) the log law overestimated the dissipation by more
than two orders of magnitude; however, farther from the
bed «L / «SFA (0.5 , z , 1.8 m; Res . 2 3 104). This
5. Discussion and conclusions
highlights the poor performance of log law near the wall,
This study computed dissipation of turbulent kinetic in this complex geophysical flow. The log regression on
energy within an unsteady boundary layer of a large lake «L/«SFA versus Res in Fig. 11 (R2 5 0.64) provides a
by applying four common methods: log law, inertial measure for accuracy of log-law calculations. Usage of a
subrange fitting, structure function, and Batchelor fitting. Stokes’s Reynolds number (Jabbari 2015; Cannon and
The log law overestimated dissipation near the wall Troy 2018) did not collapse the data.
(e.g., at z , 0.5 m in Figs. 6f,i,l). This can be explained In Lake Erie it has been shown that the log layer can
from analysis of turbulent open channel flows. For ex- extend 0.5–3 m above the bed during the measurements
ample, in Figs. 10a and 10b the log law is only valid for in this study and the velocity profiles ,1 m above the
z1 . 30, where the flow is in equilibrium (production ; bed follow the log law with 20% error (Valipour et al.
dissipation), but it overestimates dissipation for z1 , 30, 2015a). Here, we showed that error in friction velocity,
where the shear from the bed affects the flow (Pope from fitting velocity profiles to the log law over the
2000). To evaluate the deviation of near-bed dissipation constant-stress layer, can result in #17% uncertainty in
from log-law predictions, we calculated the ratio of dissipation from anisotropy-adjusted constants using
dissipation from log law to that from the structure the structure function.
function («L/«SFA) versus a local shear-flow Reynolds Farther away from the wall, however, in highly un-
number Res(z) 5 u(z)z/n, which captures the large-eddy steady flows such as Lake Erie with oscillatory forcing

TABLE 1. Rate of dissipation of turbulent kinetic energy (W kg21) from Batchelor fitting («TMM), SFM based on anisotropy-adjusted
(«SFA) and canonical («SFC) constants, log law («L), and the inertial dissipation method («IDM).

Day of year (2009) z (m) z1 «TMM «SFA «SFC «L «IDM

29 29 29 29
182.64 1 3240 7.65 3 10 5.09 3 10 3.99 3 10 8.74 3 10 6.13 3 1029
182.64 1.25 4050 1.23 3 1028 8.87 3 1029 7.12 3 1029 7.03 3 1029
182.64 1.5 4860 8.14 3 1029 6.02 3 1029 4.93 3 1029 5.88 3 1029
147.13 1 4980 1.02 3 1026 6.76 3 1027 5.78 3 1027 3.08 3 1027 8.98 3 1027
147.13 1.25 6225 2.37 3 1026 1.65 3 1026 1.39 3 1026 2.42 3 1027
147.13 1.5 7470 2.42 3 1026 2.33 3 1026 2.00 3 1026 2.03 3 1027
MARCH 2020 JABBARI ET AL. 527

FIG. 9. Velocity profiles from the HR-ADCP normalized by ut compared with log law [uL; Eq. (6); solid line] at
(a) day 123.98, (b) day 182.64, (c) day 123.07, and (d) day 147.13, which correspond with Figs. 6a, 6e, 6c, and 6g,
respectively.

from the basin-scale seiches (Valipour et al. 2015b), (iv) the averaging time scale in the IDM. These are
nonboundary layer processes [e.g., baroclinic shear discussed in detail below:
(Bouffard et al. 2012), entrainment (Ivey and Boyce
1982)] may be generating/damping turbulence, and (i) ADCP beam angle: We applied vertical velocity
therefore, the velocity and dissipation profiles can component Kolmogorov constants to in-beam
deviate from the logarithmic behavior due to these HR-ADCP velocities measured at 258 to the vertical.
energetic events (e.g., Lorke et al. 2002; Cannon and The horizontal velocity component coefficients
Troy 2018). This can be seen in increased Reynolds differ from the vertical, leading to a potential over-
stress (ru0 w0 and ry0 w0 ; Figs. 6b,e,h,k), where the dissi- estimation of dissipation by a factor of 1.33 (Cannon
pation increases significantly above log-law behavior, and Troy 2018). This error is within the range of that
which can only be attributed to locally generated tur- from other methods (Lorke and Wüest 2005).
bulence. For example, at z . 1.25 m, «L , 4«SFA (Fig. 6l) (ii) Inhomogeneity between instruments: The IDM fits
as the Reynolds stress increases by . 0.1 N m22 (Fig. 6k). velocity fluctuations at a single point to the inertial
Evaluation of the impacts of these events on bound- subrange, where the data are converted to wave-
ary layer dynamics requires further study that in- number space using the mean velocity. Conversely,
cludes consideration of the basin-scale processes in the SFM relies on spatial correlations of velocity
the water column. fluctuations in the inertial subrange with height
Time series of «IDM were in reasonable agreement with (e.g., 0.03 , r/2 , 0.18 m in this study). In unsteady
the «SFA at 1 m above the bed (Fig. 5b); however, the boundary layer flows, where turbulence can be gen-
predictions between these two methods had ;30% dif- erated from nonboundary layer processes (e.g., baro-
ference at times when the Reynolds number was .2 3 clinic seiches, wave–wave interactions, and shear
104 (Fig. 5c). That difference increased during flow re- instability) inhomogeneous turbulence may cause
versal (Fig. 5b). The difference may result from several variability in the estimation of dissipation between
factors, including (i) the HR-ADCP beam angle, (ii) these methods.
point versus spatial acoustic observations, (iii) spatial (iii) Inhomogeneity between ADCP beams: Beam av-
differences in dissipation along HR-ADCP beams, and eraging the dissipation in the SFM relies on the
528 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 37

FIG. 10. (a) Velocity profile from LES of a steady turbulent open channel flow (solid line) vs log law
1 1
[u1 1
L 5 (1/k) lnz 1 5:1; dashed line] and laminar sublayer flow (u 5 z ; dashed–dotted line). (b) Production (dotted
line) and dissipation (solid line) of turbulent kinetic energy from LES of Ret 5 2000 vs «L [dashed line; Eq. (7)].
Nondimensional production P1 and dissipation «1 are normalized by u4t /y and the velocity u1 is normalized by ut.

assumption of homogeneous turbulence, the dif- anisotropy. Differences in dissipation from each
ference between the dissipation calculated from the beam are common when the SFM is applied to
beams (Fig. 3b) shows that this condition was not ADCP data, and the results are typically beam
fulfilled. Wiles et al. (2006) showed that upstream averaged (Wiles et al. 2006; Lorke 2007), con-
facing beams can give greater dissipation in tidal tributing to measurement error.
flows and attributed that to inhomogeneity and (iv) Frozen time assumption: The IDM ideally fits over
anisotropy in the shear and Reynolds stresses. one to two decades of wavenumber space (Bluteau
Here, D [Eq. (2)] in each of the three beams et al. 2011), with the mean velocity typically ap-
was—at times —different, which gave differing plied as a convection velocity to convert frequency
realizations of dissipation (e.g., 4.65 3 1029, 3.12 3 spectra into wavenumber space. Since the IDM is
1029, and 1.61 3 1029 W kg21 for the top, middle, strictly valid in steady-state turbulence, by applying
and the bottom lines, respectively, in Fig. 3b). The the IDM to oscillating flows, the Reynolds-averaging
difference between the dissipation from beams time scale should be short enough to allow unstead-
typically increases with the distance from the bed iness in the mean velocity to be neglected, but long
according to the divergence of the ADCP beams enough to capture the inertial subrange. In this study
(Figs. 6c,f,i,l). The difference between the maximum the 5-min Reynolds averaging (burst length) was
(«max) and minimum («min) dissipation between the relatively short compared to the ;14 h surface
three beams can overshadow the improvement in seiche or ;17 h Poincare wave, giving about one
dissipation from using anisotropy-adjusted constants; decade of inertial subrange during strong turbu-
for example, at z1 5 4050 (z 5 1.25 m) in Fig. 6i, lence (Re1m . 2 3 104), for example, day 122.51
«max/«min 5 2.26, whereas «SFA/«SFC 5 1.24 and (Figs. 5c and 12). However, during flow reversal,
«TMM/«SFA 5 1.40, which shows that inhomoge- with a decrease in velocity and turbulence intensity,
neity errors can be greater than the errors due to the shear reduced the extent of the inertial subrange
MARCH 2020 JABBARI ET AL. 529

FIG. 11. The log of the ratio of the dissipation from log law
[Eq. (7)] to those from the structure function based on anisotropy-
adjusted Kolmogorov constants CA [Eq. (3)] vs log of parameter
Res 5 u(z)z/n. The plot is colored by the height above bottom.
The solid black line is a fit {logRes 5 0.047[log(«L/«SFA)]2 2
0.77[log(«L/«SFA)] 1 2.74, R2 5 0.64} and the dotted black lines
show 95% prediction interval.

(Bluteau et al. 2011; Jabbari et al. 2015), for exam-

ple, day 123.39 (Figs. 5c and 11). Figure 5c shows FIG. 12. Spectral density of ADV velocity measurements at 1 m
the ratio of shear length scale Ls 5 («IDM/S3)1/2, above the bed on days 123.39 (dashed–dotted line) and 122.51
where S 5 (2Sij Sij )1/2 is the mean velocity shear, to (solid line). The dashed line shows the inertial subrange 25/3 slope.
the Kolmogorov length scale (h 5 n3/«IDM)1/4) as
B 5 Ls/h at 1 m above the bed. In general, B , 100
In case of an oscillating flow (Fig. 1), IDM and log law
during the deployment, which is significantly less
have error of 40% and 66%, respectively, in comparison
than B . 3000, as required to observe more than
to the LES, with RMSE ;10210 W kg21 (Table 2). When
decades of inertial subrange (at much higher
measurement error is included, the percentage ranges
Reynolds numbers and farther from the boundary,
between 72% and 436% for the TMM and IDM, with
Bluteau et al. 2011; Jabbari et al. 2015). However,
RMSE ;1029 W kg21. The errors are larger in the
at 1 m above the bed about one-half decade of in-
laminar phases (08 , f , 908 and 1808 , f , 2708, where
ertial subrange was observed during weak turbu-
f is phase) before the turbulent bursts during the de-
lence in this study.
celeration phases (908 , f , 1808 and 2708 , f , 3608;
Density stratification may also be causing anisotropy Akhavan et al. 1991). These benchmarks against nu-
through the Ozmidov scale. The anisotropy-adjusted merical simulations may be applied to better inter-
constants were calculated from LES of the boundary pret the accuracy of the results from the present
layer in a neutrally stratified flow, where the anisotropy study. Here, we compare against the mean of the
was due to bed shear (Jabbari et al. 2016). In this study observational realizations («IDM 1 «TMM 1 «SFA)/3,
the measurements were during weak spring stratification, which appear convergent (Fig. 6 and Table 2). Error
N2 , 1024 s22 (average N2 ; 1026 s22), where N is the ranges from 4% to 51%, with RMSE from ;1028 to
buoyancy frequency (Valipour et al. 2015a). Considering ;1027 W kg21 while log law remains least accurate.
the average dissipation «SFA ; 1028 W kg21 (Fig. 7c) and From these comparisons we may ascertain that method
average friction velocity ; 0.003 ms21 (Fig. 5a), the ratio of errors can be as low as ;10210 W kg21 (Table 2). This
the Ozmidov length scale [L0 5 («/N3)1/2] to Kolmogorov is a negligible level, as it is close to the noise floor for the
length scale [h 5 (n3/«)1/4] [I 5 L0/h 5 («/nN2)3/4] was TMM applied to SCAMP data (J. Imberger 2004, per-
;1000, which is much less than 3000, as required for sonal communication). Adding measurement error re-
stratification to influence anisotropy (Bluteau et al. 2011). sults in an order of magnitude increase in RMSE.
A combination of measurements and methodological Comparison with DNS and LES has shown that
errors can result in scatter in the dissipation estimates. methodological errors can be reduced by anisotropy
530 JOURNAL OF ATMOSPHERIC AND OCEANIC TECHNOLOGY VOLUME 37

TABLE 2. Dissipation error analysis. The top section shows an evaluation of method and measurement error between dissipation
computed from the large-eddy simulations («LES) and other methods shown in Fig. 1. The bottom section shows an evaluation of relative
error from dissipation estimates in the present study.

RMSE Error in laminar Error in turbulent

Dissipation estimate (Fig. 1) (W kg21) Error (%) phases (%) phases (%)
«LES
IDM vs «
LES
(Method error) 5.2 3 10210 40 61 15
«LES
L
LES
vs « (Method error) 7.8 3 10210 66 83 47
«RDI
IDM vs «
LES
(Measurement and method error) 2.8 3 1029 436 640 54
«Nortek
IDM vs «
LES
(Measurement and method error) 2.1 3 1029 74 80 62
«SeaBird
TMM vs «LES
(Measurement and method error) 1.9 3 1029 72 75 55
Dissipation estimate (this study) RMSE (W kg21) Error (%)
27
«SFA vs. [(«IDM 1 «TMM 1 «SFA )/3] (Measurement 1.5 3 10 22
and method error)
«IDM vs. [(«IDM 1 «TMM 1 «SFA )/3] (Measurement 3.2 3 1028 4
and method error)
«TMM vs. [(«IDM 1 «TMM 1 «SFA )/3] (Measurement 1.2 3 1027 21
and method error)
«L vs. [(«IDM 1 «TMM 1 «SFA )/3] (Measurement 3.9 3 1027 51
and method error)

adjustment with regard to the flow conditions (Jabbari and inertial fits provide estimates of dissipation that
et al. 2015, 2016). Using field measurements, this study converge to within ;20% of each other.
showed that application of anisotropy-adjusted con-
stants in the SFM can result in higher dissipation Acknowledgments. The authors thank Ram Yerubandi
values, which are closer to those from IDM and TMM, at Environment and Climate Change Canada and the
and consequently reduce the discrepancy between captain and crews the Limnos for deploying the instru-
different methods (Table 1). For example, the mean ments in Lake Erie. This research was funded by an
standard deviation at 1m above the bed decreases from NSERC Discovery Grant to LB.
;6.41 3 1026 and 3.87 3 1029 W kg21 in Figs. 5c and 8,
respectively, using canonical constants to ;4.13 3 1026 and REFERENCES
1.01 3 1029 W kg21, respectively, using anisotropy-adjusted Akhavan, R., R. D. Kamm, and A. H. Shapiro, 1991: An investi-
constants. In general, it is recommended that anisotropy- gation of transition to turbulence in bounded oscillatory
adjusted coefficients be applied when z1 , ;10 000 Stokes flows. Part 2. Numerical simulations. J. Fluid Mech.,
(Jabbari et al. 2016), leading to significant changes 225, 423–444, https://doi.org/10.1017/S0022112091002112.
Bluteau, C. E., N. L. Jones, and G. N. Ivey, 2011: Estimating tur-
(improvements) in near-bed estimation of dissipation. bulent kinetic energy dissipation using the inertial subrange
In conclusion, because the observed dissipation can- method in environmental flows. Limnol. Oceanogr. Methods,
not be obtained directly from the dissipation tensor 9, 302–321, https://doi.org/10.4319/lom.2011.9.302.
(e.g., as in DNS) and there is no ‘‘true’’ observational ——, S.-L. Smith, G. N. Ivey, T. L. Schlosser, and N. L. Jones, 2016:
Assessing the relationship between bed shear stress estimates
value to which various methods, each with inherent
and observations of sediment resuspension in the ocean.
assumptions and limitations, may be compared. As a 20th Australasian Fluid Mechanics Conf., Perth, Australia,
consequence, we must rely on assessing accuracy Australasian Fluid Mechanics Society.
through intercomparing our results in the context of Boegman, L., and G. N. Ivey, 2009: Flow separation and resuspension
the underlying physics. Our results show that the log beneath shoaling nonlinear internal waves. J. Geophys. Res.,
law overestimates the dissipation close to the bound- 114, C02018, https://doi.org/10.1029/2007JC004411.
——, M. R. Loewen, P. F. Hamblin, and D. A. Culver, 2001:
ary with unreliable differences as large as one order of Application of a two-dimensional hydrodynamic reservoir
magnitude in comparison to the other methods. The model to Lake Erie. Can. J. Fish. Aquat. Sci., 58, 858–869,
dissipation from the structure function, using isotropy https://doi.org/10.1139/f01-035.
adjusted constants, increased by a factor of 6 or more Bouffard, D., and L. Boegman, 2013: A diapycnal diffusivity model
close to the bed and showed improved agreement in for stratified environmental flows. Dyn. Atmos. Oceans, 61–62,
14–-34, https://doi.org/10.1016/j.dynatmoce.2013.02.002.
comparison to dissipation from microstructure and iner- ——, ——, and Y. R. Rao, 2012: Poincaré wave induced mixing in a
tial fitting. We can conclude that, when carefully applied, large lake. Limnol. Oceanogr., 57, 1201–1216, https://doi.org/
the isotropy-adjusts structure function, microstructure 10.4319/lo.2012.57.4.1201.
MARCH 2020 JABBARI ET AL. 531

——, J. D. Ackerman, and L. Boegman, 2013: Factors affecting the ——, L. Umlauf, T. Jonas, and A. Wüest, 2002: Dynamics of tur-
development and dynamics of hypoxia in a large shallow bulence in low-speed oscillating bottom-boundary layers of
stratified lake: Hourly to seasonal patterns. Water Resour. stratified basins. Environ. Fluid Mech., 2, 291–313, https://
Res., 49, 2380–2394, https://doi.org/10.1002/wrcr.20241. doi.org/10.1023/A:1020450729821.
Cannon, D. J., and C. D. Troy, 2018: Observations of turbulence McGinnis, D. F., S. Sommer, A. Lorke, R. N. Glud, and P. Linke, 2014:
and mean flow in the low-energy hypolimnetic boundary layer Quantifying tidally driven benthic oxygen exchange across per-
of a large lake. Limnol. Oceanogr., 63, 2762–2776, https:// meable sediments: An aquatic eddy correlation study. J. Geophys.
doi.org/10.1002/lno.11007. Res. Oceans, 119, 6918–6932, https://doi.org/10.1002/2014JC010303.
Doron, P., L. Bertuccioli, J. Katz, and T. R. Osborn, 2001: Turbulence Monin, A. S., and A. M. Yaglom, 1975: Statistical Fluid Mechanics:
characteristics and dissipation estimates in the coastal ocean Mechanics of Turbulence. Vol. 2. MIT Press, 874 pp.
bottom boundary layer from PIV data. J. Phys. Oceanogr., 31, Munk, W. H., 1966: Abyssal recipes. Deep-Sea Res. Oceanogr. Abstr.,
2108–2134, https://doi.org/10.1175/1520-0485(2001)031,2108: 13, 707–730, https://doi.org/10.1016/0011-7471(66)90602-4.
TCADEI.2.0.CO;2. Pope, S. B., 2000: Turbulent Flows. Cambridge University Press, 771 pp.
Grant, W. D., and O. S. Madsen, 1979: Combined wave and current Rao, Y. R., N. Hawley, M. N. Charlton, and W. M. Schertzer,
interaction with a rough bottom. J. Geophys. Res., 84, 1797– 2008: Physical processes and hypoxia in the central basin
1808, https://doi.org/10.1029/JC084iC04p01797. of Lake Erie. Limnol. Oceanogr., 53, 2007–2020, https://
——, and ——, 1986: The continental-shelf bottom boundary layer. doi.org/10.4319/lo.2008.53.5.2007.
Annu. Rev. Fluid Mech., 18, 265–305, https://doi.org/10.1146/ Saddoughi, S. G., and S. V. Veeravalli, 1994: Local isotropy in tur-
annurev.fl.18.010186.001405. bulent boundary layers at high Reynolds number. J. Fluid Mech.,
Huntley, D. A., 1988: A modified inertial dissipation method for 268, 333–372, https://doi.org/10.1017/S0022112094001370.
estimating seabed stresses at low Reynolds numbers, with Sarkar, S., 2003: The effect of stable stratification on turbulence
application to wave/current boundary layer measurements. anisotropy in uniformly sheared flows. Comput. Math. Appl.,
J. Phys. Oceanogr., 18, 339–346, https://doi.org/10.1175/1520- 46, 639–646, https://doi.org/10.1016/S0898-1221(03)90022-8.
0485(1988)018,0339:AMIDMF.2.0.CO;2. Sauvageot, H., 1992: Radar Meteorology. Artech House, 366 pp.
——, and D. G. Hazen, 1988: Seabed stresses in combined wave and Scalo, C., U. Piomelli, and L. Boegman, 2012: High-Schmidt-
steady flow conditions on the Nova Scotia continental shelf: number mass transport mechanisms from a turbulent flow to
Field measurements and predictions. J. Phys. Oceanogr., 18, absorbing sediments. Phys. Fluids, 24, 085103, https://doi.org/
347–362, https://doi.org/10.1175/1520-0485(1988)018,0347: 10.1063/1.4739064.
SSICWA.2.0.CO;2. Scavia, D., and Coauthors, 2014: Assessing and addressing the re-
Imberger, J., 1998: Flux paths in a stratified lake: A review. Physical eutrophication of Lake Erie: Central basin hypoxia. J. Great
Processes in Lakes and Oceans, J. Imberger, Ed., Coastal and Lakes Res., 40, 226–246, https://doi.org/10.1016/j.jglr.2014.02.004.
Estuarine Studies, Vol. 54, Amer. Geophys. Union, 1–18. Schwefel, R., M. Hondzo, A. Wüest, and D. Bouffard, 2017: Scaling
Ivey, G. N., and F. M. Boyce, 1982: Entrainment by bottom cur- oxygen microprofiles at the sediment interface of deep strat-
rents in Lake Erie. Limnol. Oceanogr., 27, 1029–1038, https:// ified waters. Geophys. Res. Lett., 44, 1340–1349, https://
doi.org/10.4319/lo.1982.27.6.1029. doi.org/10.1002/2016GL072079.
Jabbari, A., 2015: Analysis of idealized numerical simulations to Townsend, A. A., 1976: The Structure of Turbulent Shear Flow.
calibrate and validate boundary layer models. Ph.D. thesis, 2nd ed. Cambridge University Press, 440 pp.
Queen’s University, 148 pp. Troy, C., D. Cannon, Q. Liao, and H. Bootsma, 2016: Logarithmic
——, L. Boegman, and U. Piomelli, 2015: Evaluation of the in- velocity structure in the deep hypolimnetic waters of Lake
ertial dissipation method within boundary layers using nu- Michigan. J. Geophys. Res. Oceans, 121, 949–965, https://
merical simulations. Geophys. Res. Lett., 42, 1504–1511, doi.org/10.1002/2014JC010506.
https://doi.org/10.1002/2015GL063147. Valipour, R., D. Bouffard, and L. Boegman, 2015a: Parameterization
——, A. Rouhi, and L. Boegman, 2016: Evaluation of the structure of bottom mixed layer and logarithmic layer heights in central
function method to compute turbulent dissipation within bound- Lake Erie. J. Great Lakes Res., 41, 707–718, https://doi.org/
ary layers using numerical simulations. J. Geophys. Res. Oceans, 10.1016/j.jglr.2015.06.010.
121, 5888–5897, https://doi.org/10.1002/2015JC011608. ——, ——, ——, and Y. R. Rao, 2015b: Near-inertial waves in
Keating, A., U. Piomelli, K. Bremhorst, and S. Nesić, 2004a: Large- Lake Erie. Limnol. Oceanogr., 60, 1522–1535, https://doi.org/
eddy simulation of heat transfer downstream of a backward-facing 10.1002/lno.10114.
step. J. Turbul., 5, N20, https://doi.org/10.1088/1468-5248/5/1/020. ——, L. Boegman, D. Bouffard, and Y. R. Rao, 2017: Sediment
——, ——, E. Balaras, and H.-J. Kaltenbach, 2004b: A priori and a resuspension mechanisms and their contributions to high-
posteriori tests of inflow conditions for large-eddy simulation. turbidity events in a large lake. Limnol. Oceanogr., 62,
Phys. Fluids, 16, 4696–4712, https://doi.org/10.1063/1.1811672. 1045–1065, https://doi.org/10.1002/lno.10485.
Kundu, P. K., and I. M. Cohen, 2002: Fluid Mechanics. Academic Wiles, P. J., T. P. Rippeth, J. H. Simpson, and P. J. Hendricks, 2006:
Press, 759 pp. A novel technique for measuring the rate of turbulent dissi-
Lorke, A., 2007: Boundary mixing in the thermocline of a large pation in the marine environment. Geophys. Res. Lett., 33,
lake. J. Geophys. Res., 112, C09019, https://doi.org/10.1029/ L21608, https://doi.org/10.1029/2006GL027050.
2006JC004008. Yuan, J., and U. Piomelli, 2014: Numerical simulations of sink-flow
——, and A. Wüest, 2005: Application of coherent ADCP for turbulence boundary layers over rough surfaces. Phys. Fluids, 26, 015113,
measurements in the bottom boundary layer. J. Atmos. Oceanic https://doi.org/10.1063/1.4862672.
Technol., 22, 1821–1828, https://doi.org/10.1175/JTECH1813.1. Zulberti, A. P., G. N. Ivey, and N. L. Jones, 2018. Observations of
——, and S. Maclntyre, 2009: Hydrodynamics and mixing in near-bed stress beneath nonlinear internal wave trains in
lakes, reservoirs, wetlands, and rivers. Encyclopedia of the ocean. Proc. 21st Australasian Fluid Mechanics Conf.,
Inland Waters, G. E. Likens, Ed., Elsevier, 505–514. Adelaide, Australia, Australasian Fluid Mechanics Society.