Professional Documents
Culture Documents
Arus Sisa 1 PDF
Arus Sisa 1 PDF
Research papers
art ic l e i nf o a b s t r a c t
Article history: A reduced physics Ekman boundary layer solution is developed to infer the vertical structure of time-
Received 1 August 2012 mean circulation in a shallow tidal environment when the horizontal density and surface slope gradients
Received in revised form are misaligned. This generalisation of the classic Heaps (1972) model shows that the time-mean depth
18 February 2013
weighted flow, or the residual circulation, is usefully constrained by knowledge of the surface velocity,
Accepted 5 March 2013
Available online 27 March 2013
instead of freshwater flux, and the horizontal density gradient.
The generalised model is applied to Liverpool Bay. In regions where the Ekman depth scale is less
Keywords: than half the mean fluid depth the residual circulation is well modelled by a water column of uniform
Tidal residual density, constant eddy viscosity and linear bottom drag. Lateral variability in long-term mooring
ROFI
observations of depth varying residual flow are attributed to the misalignment of sea surface slope
HF radar
and haline controlled density gradients.
ADCP
Freshwater plume A method to infer 3D time-average residual currents in regions of misaligned freshwater density and
3D model sea surface slope gradients is presented. The method blends CTD survey data with HF radar surface
currents and simulation estimates of viscosity and friction. It is validated against ADCP data in Liverpool
Bay. It is speculated that this method could be applied more generally, to correct model biases, as part of
a coastal monitoring system.
& 2013 Elsevier Ltd. All rights reserved.
0278-4343/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.csr.2013.03.004
116 J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126
25.5
53.9
25
53.8
24.5
53.7 24
5
53.6
53.5 22.5
22
53.4
21.5
53.3
21
0.05 m/s
0.05 m/s
8 6 4 2
Fig. 1. Comparison between the observed mean density field (contoured on left) and model mean density field (contoured on right). Each map is overlayed with progressive
vectors plots of the residual volume flux (approximately every 3 m) at Sites A and B from the observations (left) and from the model (right). The model diagnostics are from
POLCOMS data for the years 2005–2007. The left hand panel also shows the extent of the HF radar coverage (green). (For interpretation of the references to color in this figure
caption, the reader is referred to the web version of this article.)
tidal constituent and are found to be less than 1 cm s−1. Prandle and distinguished from a divergent literature (Mathieu et al., 2002;
Ryder (1989) and Prandle and Matthews (1990) also report from Winant, 2006, 2007; Ponte, 2010) which is based on a similar idealised
around the same region that the observed net residual flow is model (Hendershott and Rizzoli, 1976; Ianniello, 1977; Winant, 2004)
broadly consistent with estimates of the density driven component where a surface stress is imposed but the bed has a no-slip condition.
and that the nonlinear Stokes advection is small. Here, by setting the surface stress to zero we implicitly propose that
On the largest scale (410 km) surface wind stress can have a the tidal processes are more important than the wind in maintaining
direct impact contributing both to the surface momentum and to the the Liverpool Bay system.
vertical mixing rate. In an early 3D numerical study Heaps and Jones Observational studies and monitoring systems are restricted in
(1977) found the wind to be of leading importance in determining the their spatial coverage. Often they are either restricted to only capturing
subtidal residual flow when it exceeds a modest threshold between the surface residual (Prandle, 1987), or vertical profiles at discrete
5 and 10 m s−1. This wind stress role is borne out in more contem- locations (Verspecht et al., 2009). Here we set out a novel methodol-
porary studies (e.g. Geyer, 1997; Janzen and Wong, 2002; Li and Li, ogy that combines observational data with model data to make new
2011; Ponte et al., 2012). Also at the larger scale, where the Earth's estimates of three dimensional residual circulation. In this study we
rotation is important, the horizontal pressure gradients maintain a develop a reduced physics Ekman boundary layer solution to infer the
residual circulation by balancing the Coriolis force and turbulent vertical structure of time-mean circulation in a shallow tidal environ-
‘viscous’ forces (Heaps, 1972). These pressure gradients are sustained ment where the horizontal density and surface slope gradients can be
by coastal freshwater input and by persistent sea surface height misaligned. We apply it in Liverpool Bay, where the tidal range is
gradients. Recently Verspecht et al. (2009) tested the original Heaps extreme, in order to correct a known bias in the numerical simulation.
(1972) estimates of residual circulation using Observatory data at Site In Section 2 we present the observational methods and numerical
A. They report that although the theory qualitatively captures the simulation setup. In Section 3 we develop the theoretical framework.
observations, quantitatively the solution is very sensitive to the choice In Section 4 this framework is used to explain the laterally varying
of eddy viscosity coefficient. This indicates that the classic theory, with nature of residual circulation and to investigate performance errors
a depth and time invariant eddy viscosity, is too crude to prognos- in the coastal ocean simulation POLCOMS. Finally in Section 5 a
tically model the residual circulation at that site. methodology is proposed to combine observational data with model
Analytic studies have tended to focus on certain components of the data to make 3D estimates of the time-mean circulation.
residual circulation (e.g. the density driven component (Heaps, 1972),
or tidal constituent decomposition (Prandle, 1982)) whereas numerical
simulations, while capturing a more complete physics set (Holt and 2. Material and methods
James, 2001) sacrifice transparency of individual processes. Whilst
effective complexity trade offs can be made in tackling the time This article draws together the surface current data derived from
varying problem (Ralston et al., 2008), in this manuscript we argue HF radar measurements (Gurgel and Antonischki, 1997), data from
that significant skill can be achieved (for the steady state problem) two fixed mooring sites which include Acoustic Doppler Current
using a modification to the basic analytic density driven model Profilers (ADCPs), CTD survey data, and numerical model data.
proposed by Heaps (1972). The success of this residual circulation
model approach is largely due to the ability to implicitly incorporate 2.1. Fixed Moorings
wind and nonlinear tidal advection with an imposed surface velocity
field. (Though ‘good’ skill is restricted to regions where the water The two fixed moorings, Sites A and B (Fig. 1), are detailed
depth exceeds twice the Ekman depth scale.) This model is in Howarth and Palmer (2011) and so are only summarised here.
J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126 117
0.02
10m interval
53.8
0.01
53.7
Velocity (m/s)
0.00
53.6
A
-0.01
53.5
B
-0.02
0 1 2 3 4 5 6 7 8 9 53.4
years
Fig. 2. The difference between the mean of the 9 year record and averages derived
53.3 0.10 m/s
from averaging over longer and longer periods, for both the east (blue) and north
(red) components at multiple bin depths. For reference the magnitude of the near
surface residual current is 0.04 m s−1. (For interpretation of the references to color -4 -3.8 -3.6 -3.4 -3.2 -3
in this figure caption, the reader is referred to the web version of this article.)
Fig. 3. Simulation mean surface currents (red) and filtered HF radar currents
(green) are overlaid on model bathymetry of Liverpool Bay. At sites A and B black
Site A is located at 53131:8′N 3121:6′W at the Mersey Bar, vectors denote the depth-weighted mean ADCP near surface velocity. (For inter-
pretation of the references to color in this figure caption, the reader is referred to
approximately 10 km west of the mouth of the River Mersey and the web version of this article.)
in a mean water depth of 23.5 m. The site has been maintained
continuously since the 7th August 2002. Site B was located 21 km
to the west-south-west of Site A at 53127′N 3138:4′W, where the is not common (24% of the time) so is considered weak when
mean water depth is 25.0 m and was maintained between the 5th compared to the persistent, horizontal density gradients (Polton
April 2005 and the 26th January 2010. Bottom mounted ADCPs are et al., 2011). Horizontal density gradients are computed from the
deployed at each location measuring horizontal velocities at CTD survey array to be ð−4:3; 0:7Þ 10−2 kg m−3 per km at Site A,
10 min intervals and 1 m vertical resolution between 2.7 m above and ð−4:4; 3:9Þ 10−2 kg m−3 per km at Site B. Howarth et al.
the bed and to within 3 m of the sea surface. (unpublished) evaluate a density gradient error associated with
This paper focusses on the steady residual circulation of Liver- this survey grid of 7:4 10−2 kg m−3 per km.
pool Bay. ADCP measurements (Fig. 2) suggest that a window of
1 year or longer is sufficient to determine mean currents to within 2.4. Calibrated POLCOMS model
0.01 m s−1 of its long term value while an increase beyond 2 years
can at best double the precision. A nested series of POLCOMS (B-grid, sigma levels) simulations
are used to simulate the Liverpool Bay region with a
2.2. Phased array HF radar 1=401 1=601∼1:8 km) horizontal resolution and 32 vertical
levels. This configuration is used to simulate the whole Irish Sea,
As described in Robinson et al. (2011), phased array HF radar but has known model bias in the Liverpool Bay freshwater
systems transmitting at 13 MHz were installed at Llanddulas and at horizontal density gradient (Polton et al., 2011; O'Neill et al.,
Formby Point (Fig. 1). The system measures the Lagrangian surface 2012) that results in an over prediction of the mean flow at Sites
current. Data were acquired in 100 cells spaced at 4 km intervals, at A and B (Fig. 1) and misalignment errors in the surface velocity
20 min resolution between August 2005 and November 2011. Since (Fig. 3). The model is forced by boundary conditions extracted
currents both in the numerical models and measured by the ADCP are from the POLCOMS Atlantic Margin Model (12 km resolution) that
in the Eulerian reference frame, the HF radar mean currents are covers the Northwest European Shelf, which in turn is forced by
converted to this reference frame, which require a calculation of the the MetOffice Atlantic (1/91 resolution) ocean forecast product.
Stokes drift of the surface waves. By calculating a linear regression The high resolution model is forced with 6 hourly, 1 11, ECMWF
between the tidal residual HF radar currents and the winds, here winds and with surface fluxes that are computed from COARE
measured at Hilbre Island at the mouth of the Dee, both the Stokes 3 bulk formulae (Fairall et al., 2003). River fluxes are derived from
drift and the surface current response to the wind stress are filtered Environment Agency river gauge data. The simulation is initialised
from the HF radar currents. The latter is also desirable since the simple with a climatological state in January 2004 and integrated through
model does not need to explicitly represent wind stress whose effects until December 2007. The final 3 years are used in the following
are instead captured through a barotropic sea level adjustment in the analysis.
system and through the surface boundary condition. (The correlation This configuration of POLCOMS uses the Canuto et al. (2001)
coefficient between the wind and the HF radar tidal residuals at Site A k-epsilon vertical mixing scheme. (Its performance against other
is 0.71, compared with a maximum of 0.42 for the ADCP currents.) The turbulence closure schemes is reported in Holt and Umlauf (2008).
HF radar mean surface velocity product is shown in Fig. 3. The tuneable coefficients are also presented in Holt and Umlauf
(2008).) In POLCOMS, horizontal advection is handled using the
2.3. CTD cruises Piecewise Parabolic Method (James, 1996). This scheme is extre-
mely good at preserving fronts, with only very weak numerical
The spatial distribution of density in Liverpool Bay is provided diffusion, and makes the model suitable for riverine freshwater,
from 8 years of near-monthly CTD survey data (Fig. 1). Survey or pollutant, dispersal problems (Phelps et al., submitted for
positions are not adjusted to account for tidal displacement since publication). Though in this study we advect only momentum,
such effects average out over the 68 surveys conducted over the T and S, a quadratic drag lower boundary condition is used in the
9 year period. Enduring (beyond tidal period) vertical stratification momentum equations (implemented in Holt and James, 1999).
118 J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126
Following Ruddick et al. (1995): C d ¼ ðκ=lnðz=z0 ÞÞ2 , where C d is These two components superimpose in the Heaps solution to
the bottom drag coefficient, κ is the von Karman constant, produce a velocity profile that rotates and changes magnitude with
z0 ¼ 0:005 m is the bed roughness length and z is the height above depth. Since the geostrophic velocity is linear in depth the solution
the bed. C d is constrained to have a floor value of 0.005. For more to ua can be written as the sum of two exponential functions.
details about the POLCOMS hydrodynamics refer to Holt and James The general solution is then
(2001) and Holt et al. (2005).
UðzÞ ¼ A expðαzÞ þ B expð−αzÞ þ ug ðzÞ ð4Þ
where α ¼ if =ν and A and B are complex constants that are
2
3. Theory/calculation determined by the choice of boundary conditions.
In the original work Heaps makes a boundary condition
Following Heaps (1972), we consider the steady problem when assumption of zero stress at the surface and a linear drag at the
pressure gradient forces are balanced by viscous and Coriolis bed. The linear drag boundary condition represents a departure
forces. Using complex notation, the linear momentum equations from the no slip condition used in much of the idealised modelling
for the time mean flow U can be written as literature (Hendershott and Rizzoli, 1976; Mathieu et al., 2002;
1 ∂2 U Sanay and Valle-Levinson, 2005; Winant, 2004, 2006, 2007; Ponte,
if U ¼ − ∇p þ ν 2 ð1Þ 2010) and is supported by the ADCP data (Fig. 4) and model
ρ ∂z
analysis (not shown). The bottom frictional stress boundary con-
where f is the Coriolis parameter, p is the hydrostatic pressure dition, where the stress (per unit density) is given to be propor-
given by p ¼ ρgðη−zÞ, g is gravitational acceleration, ρ is the depth tional to the bed flow velocity, with a scaling coefficient r such that
average density, ν is the viscosity (assumed depth invariant), and rU−∂U=∂z ¼ 0 at z ¼ −H. The dimensions of the friction coefficient
∇ ¼ ð∂=∂xÞ þ ið∂=∂yÞ is the natural complex analogue of the vector r are m−1 . Though a linear drag coefficient might be anticipated to
gradient operator. Sea level perturbations, η, are measured from have dimensions of velocity (i.e. the product rν), this choice
the mean sea level height, z ¼0 (which increases upwards). The ensures that the friction and viscosity coefficients are linearly
steady state assumption allows us to approximate the residual independent in later sensitivity analyses. For a representative
flow as the time invariant velocity U. Time independence is velocity scale V, the friction coefficient controls whether the
achieved by taking a sufficiently long time average, filtering out boundary condition is no slip (rν=V⪢1) or zero stress (rν=V - 0).
the tidal contributions that vary temporally whilst preserving The coefficients in (4) can be expressed as a function of surface
vertical structure. Deriving Eq. (1) in a volume conserving frame- slope and horizontal density gradient:
work (Mellor, 2003; Ardhuin et al., 2008; Winant, 2008) results in
an error of about 0.5 cm s−1 (following a perturbation expansion in ig hh r i i
A¼ 1 þ eαH −ð1 þ rHÞ ∇ρ−rρ∇η ð5Þ
tidal amplitude over mean depth, see Appendix). γf ρ α
ig hh r i i
3.1. Generalised Ekman solution B¼ 1− e−αH −ð1 þ rHÞ ∇ρ−rρ∇η ð6Þ
γf ρ α
To obtain solutions, the velocity is decomposed into two depth
where γ ¼ 2r coshðαHÞ þ 2α sinhðαHÞ: ð7Þ
varying terms, U ¼ ug þ ua . Firstly there is the geostrophic flow, ug ,
parallel to the pressure contours at each depth. Secondly there is an Substituting A and B into (4) gives the generalised Ekman solution
ageostrophic velocity, ua , which is an Ekman boundary layer flow. for the residual circulation as functions of ∇η and ∇ρ:
The geostrophic and ageostrophic velocities are defined as follows:
r coshðαzÞ ig
U ¼ 1− ρ∇η
i ig r coshðαHÞ þ α sinhðαHÞ f ρ
ug ðzÞ ¼ ∇p ¼ ½ρ∇η−z∇ρ ð2Þ
r sinhðαðz þ HÞÞ þ α coshðαðz þ HÞÞ−αð1 þ rHÞcoshðαzÞ ig∇ρ
fρ fρ þ −αz : ð8Þ
r coshðαHÞ þ α sinhðαHÞ αf ρ
i ∂ U i ∂ u 2 2 a
In practise ∇η is not known and so the surface geostrophic
ua ðzÞ ¼ − ν 2 ¼ − ν 2 ð3Þ
f ∂z f ∂z current is constrained by additional information. In the Heaps
0 0
−5 −5
−10 −10
Depth (m)
Depth (m)
−20 −20
−25 −25
−0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 −0.03 −0.02 −0.01 0 0.01 0.02 0.03 0.04 0.05 0.06
Velocity (m/s) Velocity (m/s)
Fig. 4. Residual circulation profiles for sites A (left) and B (right). The ADCP data (red) profiles are plotted with optimised profiles using the original Heaps model (grey) and
optimised profiles using the modified Heaps model (black). Solid and dashed lines denote the eastward and northward velocities respectively. The ADCP data are truncated at
the upper and lower levels. Circles denote the depth of the first subsurface observation. Surface currents generally flow offshore. Near bed currents generally flow onshore.
The error scores are given in brackets. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)
J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126 119
model these two degrees of freedom are constrained in terms of a perfectly reconstruct the residual flow and that there will be a
known diapycnal flux, q, which is riverine (and real and positive in certain amount of uncertainty in the optimally tuned parameters.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
the up density slope direction) 2
(Z ) The Ekman number, 2ν= fH , is defined as the ratio of the
0
∇ρ†
q ¼ Re U dz ð9Þ Ekman depth to the mean fluid depth. Circulations where the
−H j∇ρj
mean fluid depth exceeds the Ekman depth would be expected to
where ∇ρ† is the complex conjugate of ∇ρ and an assumption that be controlled by planetary rotation (Winant, 2004; Sanay and
the surface slope is parallel (and in the opposite direction) to the Valle-Levinson, 2005). Circulations where the mean fluid depth is
density gradient. Such that less than the Ekman depth are more heavily constrained by
viscous boundary layers (e.g. simulations in Mathieu et al.,
2rg αH sinhðαHÞ−coshðαHÞ þ 1 sinhðαHÞ 2002). In Liverpool Bay the error score corresponds very closely
q¼ ∇ρ Im −ρj∇ηjIm
fρ α γ
2 αγ to the Ekman number (see Section 5 and Fig. 9), that is not simply
ð10Þ a proxy for depth (Fig. 3), such that all sites with an Ekman number
Therefore, in the Heaps (1972) formulation, the sea surface less than 0.5 satisfy the ‘good’ error score criterion. This verifies that
slope is expressed as the Heapsian model is most effective in a low Ekman number regime
0 1 (which is consistent with expectation, since the Heaps parameterisa-
f 1 αH sinhðαHÞ−coshðαHÞ þ 1 tion of frictional processes is simplistic by design).
q− j∇ρjIm
∇ρ BB2rg ρ α2 γ C
C:
∇η ¼ ð11Þ The velocity decomposition into geostrophic and ageostrophic
j∇ρj @ sinhðαHÞ A
components is fundamental to understand the physical balances
Im
αγ that result in the differing depth varying behaviours at Sites A
We propose instead to constrain ∇η in terms of the two and B. In order to decompose the ADCP data into its component
dimensional surface velocity, U 0 . In (4), ∇η ¼ ifðA þ B−U 0 Þ=g and parts the parameters r, ν, ð1=ρÞ∇ρ, and U 0 or q are tuned in order to
the general solution (8) can be expressed in terms of ∇ρ and U 0 : find analytic profiles, determined by Eqs. (8), (11) and (12), that
best fit the data. This analysis and its sensitivity is explored in the
coshðαHÞ−coshðαH þ αzÞ þ ð1 þ rHÞðcoshðαzÞ−1Þ ig
U ¼ −z− ∇ρ next section.
r coshðαHÞ−r þ α sinhðαHÞ fρ
sinhðαHÞ−sinhðαH þ αzÞ þ sinhðαzÞ r ig
− ∇ρ 4. Understanding the observed profiles
rcoshðαHÞ−r þ α sinhðαHÞ αfρ
coshðαzÞ−1
þ 1−r U0: ð12Þ In this section, the reduced physics Ekman model of the vertical
r coshðαHÞ−r þ α sinhðαHÞ
profile of residual circulation is applied to the ADCP data from
Constraining the solution with a two dimensional surface Liverpool Bay. In Section (4.1) the dynamics are analysed and it is
velocity is consistent with relaxing the assumption that the sea shown that the original Heapsian formulation (8) and (11), with a
surface slope must be parallel to the density gradient. The solution prescribed riverine diapycnal flux, are not as effective as the
(12) is dependent on a number of tuneable parameters. These are modified solution (12), with prescribed surface current, in repro-
viscosity ν, bed friction r, surface velocity U 0 , and the density ducing the observed currents. In Section (4.2) the parameter
gradient term ∇ρ=ρ. sensitivity of the reduced physics models is investigated.
In order to evaluate the efficacy of a particular parameter set Error scores are minimised using both modified and original
relative to known velocity profile an error score is defined. This Heaps models against both Site A and Site B data. The optimised
score is a measure of the mean difference, over depth, between the profiles are shown in Fig. 4. On the left the original Heaps model
analytic solution U heaps (12) and the observed profile U ADCP , where achieves optimal scores of 0.7 and 0.6 for Sites A and B respec-
the velocity is a complex function of depth. The score is normal- tively. Alternatively, the modified formulation achieves ‘good’
ised by the variance of the observed profile and is given by the optimal scores of 0.2 and 0.1. The optimised parameter sets are
expression detailed in Table 1. The difference in score between the two
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi models is stark and is a consequence of the assumption that the
uR 0
u jU ADCP −U heaps j2 dz surface slope is parallel to the density gradient.
error score ¼ tR−H0
: ð13Þ
−H jU ADCP −U ADCP j dz
2
ug(nr surface)
ug(nr surface)
Northward velocity (m/s)
∇ ρ (CTD)
∇ ρ (CTD) ∇ ρ (Heaps)
∇ ρ (Heaps)
ADCP: Site B
ADCP: Site A
ug(nr bed) best fit: Modified
best fit: Modified
geostrophic
geostrophic
ageostrophic ug(nr bed) ageostrophic
Fig. 5. Residual flow decomposition for sites A (left) and B (right). The panels show the ADCP (red) and best fit estimate (black) hodographs as well as the geostrophic (light
green) and ageostrophic (blue) components that comprise the best fit estimate for the modified Heaps model. Near bed and surface geostrophic vectors (dark green) are
labelled to highlight ug which is not generally orthogonal to the density gradient (dashed arrows: CTD (red); Heaps model (green)). The hodograph data are truncated at the
upper and lower limits of where the ADCP data exists. Circles denote vector depth nearest the surface in all panels. Generally surface currents flow offshore. Near bed
currents flow is onshore. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)
Fig. 5 shows the laterally varying behaviour of the residual Finally either the up-density-gradient riverine diapycnal flux, q, or
current profiles at Site A and Site B, plotted as hodographs with the surface velocity, U 0 , is set. The original estimate in Heaps (1972)
the eastward velocity on the x-axis and the northward velocity on gave q ¼ 2 10−3 m3 s−1 per metre. An upper estimate for q can be
the y-axis. These plots also show the decomposition of the best fit calculated from average Liverpool Bay river flow rates (≈200 m3 s−1
solution of the modified Heaps model into the geostrophic (green) Polton et al., 2011). If these, average, rivers enter the Bay every 50 km
and ageostrophic (blue) components. At Site A (left panel) with of coastline and the flow rate is underestimated by a generous factor
decreasing depth, the observed flow (red) vector rotates predo- 10 (due to unaccounted ground water seepage), then an upper
minantly anticlockwise about the origin. At Site B (right panel) estimate of q ¼ 0:04 m2 s−1 . Therefore the range given by q ¼
with decreasing depth, the observed flow (red) vectors rotate ½0; 0:04 m2 s−1 is set. For the modified residual circulation model
clockwise. the surface velocity is used instead of imposing the diapycnal flux.
At both sites the sense of rotation is consistent with the rotary Here, because the error perturbations on any individual surface
sense of the geostrophic component (Fig. 5). However the geos- velocity estimate are anticipated to be smaller than the spatial
trophic component (2) is a linear function of depth and therefore variability of the quantity, parameter space is explored about an
only rotates if it does not pass through the origin at some depth. initial estimate with an envelope of size 7 0:03 m s−1 , with incre-
This is only possible if the sea surface slope and density gradients ments of 0:005 m s−1 , in each direction.
are not parallel. It therefore seems likely that the modified In tuning the parameters to minimise the error score there is an
Heapsian model that imposes the surface velocity would be implicit assumption that the optimising process brings the para-
preferable to the Heaps (1972) assumption of a parallel surface meters closer to their real value rather than that the tuning just
slope and pressure gradient. As a corollary to relaxing the compensates for errors in the reduced physics assumptions in the
assumption of parallel surface slope and density gradients, a momentum balance (1) that is being solved. Whilst it is not
component of the diapycnic transport q can be in geostrophic possible to definitively defend this assumption without more
balance. This results in apparent freshwater fluxes that are two accurate observational measurement of the parameters, an error
orders of magnitude larger than the riverine contribution and score sensitivity study, varying the parameters, can be instructive
must be balanced by surface buoyancy forces. as to whether this concern is important.
Since in practise there might be some considerable uncertainty
4.2. Sensitivity in the values of the controlling parameters (r; ν; ∇ρ and q or U 0 )
it is essential, in order to be practically useful, to explore the
As in the original Heaps work, a number of variables are sensitivity of the score to perturbations in the optimal parameters.
uncertain, so parameter space is explored, within realistic limits, The optimal parameters are given in Table 1.
to find the optimal set of parameters for the analytic solutions. The In the following subsections the error score sensitivity is
original estimate for bed friction was r ¼ 0:126 m−1 . So the range investigated across the parameter ranges. Site A and Site B are
r ¼ 10j m−1 is set where j ¼ ½−3; 2 with increments of 0.5. investigated in turn.
Similarly the original estimates of the viscosity parameter had a
range ν ¼ 0:0159–0:0636 m2 s−1 . Therefore, here, a wider range,
ν ¼ 10k m2 s−1 , where k ¼ ½−4; −1, with increments of 0.05, is set. 4.2.1. Site A sensitivity
The horizontal gradient of density is also uncertain (Howarth Score maps are computed by varying parameter pairs over their
et al., unpublished). Error envelopes in ð1=ρÞ∇ρ are centred about permitted range, whilst keeping the remaining parameters at their
the CTD derived estimates for Sites A and B with equal ranges in optimised values (Fig. 6) and compared to results from the original
both horizontal directions. For Site A the range is 7 8 10−8 m−1 Heaps formulation. This is useful since in the real ocean most of
and for Site B the range is 7 5 10−8 m−1 . these parameters have an uncertainty associated with any
J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126 121
Fig. 6. Site A sensitivity maps showing error scores for the original Heaps formulation (left hand panels) and the modified Heaps (right hand formulation). Score less than
0.5 are ‘good’. Maps show sensitivity of parameter deviations from the optimised set. The top panels show sensitivity to friction r, and viscosity ν, with the cross denoting the
minimum error solution. The middle row shows the sensitivity to vary the density gradient term, with the optimal vector (solid) and CTD estimate shown (dashed). The
lower left panel shows the sensitivity to freshwater flux q, with the marker denoting the original offshore Heaps estimate (2 10−3 m2 s−1 ). The lower right hand panel
shows the sensitivity to the choice of surface velocity U 0 . The optimal U 0 is shown as a solid vector.
measurements. This analysis can help us determine the confidence hand, the original solution has no good solution anywhere in this
ranges, on the optimal parameter set, that achieve good scores. friction–viscosity parameter space! The crosses denote the optimised
The scores are largely insensitive to the bottom bed boundary solution.
condition parameter rν=U, for representative velocity U ¼ 0:01 m s−1 , Linearly varying ð∇ρÞ=ρ from its optimised vector (Fig. 6, middle
suggesting that either a free slip or a no slip boundary condition row), with all other parameters being held constant, it has not been
could be used. This is partially a reflection that the observations do possible to produce a residual current profile with a good score using
not extend to the bed. In particular scores are more insensitive to the original Heaps solution since sufficiently low scores are unobtain-
variations in r than ν, implying that it is less important to reduce able. However, with the modified Heaps solution good scores are
uncertainty in r than in the viscosity parameter. Good scores are found and is still attained using the CTD estimates of the density
obtainable for both solutions though they occupy different regions of gradient. The magnitude of the discrepancy between the optimised
ðν; rÞ parameter space. The new solution exhibits good scores over and CTD derived density gradients are greater than the anticipated
the range ν ¼ 0:003–0:01ð10−2:5 –10−2 Þ m2 s−1 , though the range is measurement error when using the original Heaps formulation and
narrow and skewed to smallest values at the high friction end and within the anticipated measurement error when using the modified
skewed to the larger values at the low friction end. On the other formulation.
122 J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126
The score of the original Heaps formulation increases with In the new formulation q is replaced by surface velocity, U 0 .
increasing q (Fig. 6, lower row). The optimised value is constrained Exploring the sensitivity of the score to the choice of U 0 (Fig. 6,
to be positive (consistent with the concept that the net flux across lower right panel), good scores are restricted to a relatively small
density contours is balanced by the offshore riverine freshwater parameter window.
flux). However this sensitivity diagnostic suggests that an opti-
mally fitted q value would be negative, corresponding to an
onshore flux of freshwater! Clearly there must be a problem with 4.2.2. Site B sensitivity
the assumption that the depth integrated residual transport across The analysis is repeated at Site B (Fig. 7). The patterns are
density surfaces is balanced solely by the freshwater flux. Fig. 5 broadly similar. Notably the original Heaps formulation is still
(left panel) pictorially highlights the problem: the direction of the unable to fit ‘good’ profiles to the observational data. At Site B the
geostrophic flow (green) varies with depth and has a component, new formulation has a broader range of good scoring viscosities
balanced by the sea surface slope, that is not perpendicular to the and a slightly broader range of permissible density gradient
density gradient. This component contributes, along with the vectors. Again with the modified solution, the difference between
riverine freshwater flux, to the depth integrated diapycnal residual the good ∇ρ=ρ and the CTD estimate is small, such that the actual
transport. Clearly, therefore the freshwater transport is not an CTD derived gradients still have ‘good’ scores, and the difference
effective way to constrain the diapycnal residual transport. between them is within the subsampling error estimate.
Fig. 9. (a) Overlapping areas covered by the CTD array (red) and HF radar (green) are used to estimate depth varying residual flow hodographs (black). Crosses denote CTD
stations. The ADCP residual flow hodographs are shown (red) along with the 3D model simulations (blue) at sites A and B. In all hodographs the surface currents are more
northward than the bed current. The Ekman number (main panel) and error index (inset panel) are contoured within the CTD grid from zero with interval 0.5. Blended
hodographs are presented at CTD stations where there is HF radar data and where Ekman number ≤0:5. (b) Optimally fitted viscosity, log10 ðνÞ (m2 s−1 ). (c) Optimally fitted
friction, log10 ðrÞ (m−1 ). (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this article.)
The reduced physics model has a number of dependent para- well. We have demonstrated that the transition from high to low
meters both measurable (lateral density gradient and surface skill with the Heapsian model corresponds to an increase in
velocity) and immeasureable (viscosity and friction). It is shown Ekman number. Suggesting that the frictional component of the
that the error of the Heaps model is largely insensitive to friction Heaps model is not sufficient to capture the leading order
and that a generous range of viscosity parameter space satisfies dynamics at Site A. The dynamics at Site A are renowned for their
the good score criterion (Fig. 7). complexity (Verspecht et al., 2010). It is probable that wind stress,
We have introduced an error score analysis that shows the depth and/or time varying eddy viscosity may be required to
modified Heaps model which gives a significantly improved fit to appropriately represent the true behaviour there. Nevertheless,
the observed data at both Sites. We have also, for the first time, the fact that this reduced physics model, with a constant eddy
analysed the ADCP data in terms of the Heaps reduced physics viscosity profile and prescribed surface current, has such good skill
model and found that it is effective in explaining the spatial which is remarkable given the canonical view of the dominant role
variability of the observations. of periodic strain induced stratification.
Previous studies at Site A (Verspecht et al., 2009) have shown The density gradients are not sufficiently resolved on the CTD
that the original Heaps solution does not fit the observed data grid to make reliable predictions of the depth varying residual
J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126 125
circulation. It is anticipated that high spatial and temporal resolu- principle role of the long term current is to laterally transport
tion data, such as that provided by autonomous glider measure- material, a volume conserving expression for the residual circula-
ments of temperature and salinity deployed on multi-week tion is appropriate. Therefore we define
missions will significantly improve our current understanding.
∂s n
Since it was shown that 1 year of ADCP data is sufficient to capture U n ðζÞ ¼ u : ð14Þ
∂ζ
the steady residual current profile it is anticipated that at a
composite year of glider CTD survey data will equivalently be The star represents the variable being expressed in surface
adequate to satisfactorily resolve the horizontal density gradients. following vertical coordinates, ζ, e.g. uðx; z; tÞ ¼ un ðx; ζ; tÞ, which
It is found that the simulation errors in the residual circulation varies linearly from −H at the bed to 0 at the moving sea surface
are consistent with simulated density gradient being too strong. and the angle brackets represent phase averaging over the tidal
A naive cross-isopycnal advection–diffusion density budget scal- frequencies. The regular cartesian vertical coordinate is expressed
ing, where L is the density gradient length scale and νh is the as z ¼ sðx; ζ; tÞ, where H ¼ 〈η〉 þ hðxÞ is the sum of the mean surface
horizontal diffusion, gives qðΔρ=LÞ∼Hνh Δρ=L2 . If furthermore the elevation slope, 〈η〉, and the bottom depth, hðxÞ 4 0. The new
horizontal diffusion is dominated by the ‘effective diffusion’ expression for the cartesian vertical coordinate is given as
induced by vertical exchange in an oscillating shear flow a
(Bowden, 1965; Young et al., 1982), then νh ∼U 2 H 2 =ν varies inver- sðx; ζ; tÞ ¼ ηðx; tÞ þ ζ 1 þ cos ψ ð15Þ
H
sely with the vertical diffusion ν. Thus the cross-front length scale
where ψ is the phase of the tidal surface gravity wave, η, and a is
Ł∼U 2 H 3 =ðqνÞ could be too small because of overestimations in the
the amplitude.
diapycnal flux q and the vertical diffusion ν. It is also shown that
Since the wavelength of the tide is long compared to the fluid
the residual circulation is sensitive to the value of the prescribed
depth hyperbolic depth variations of the wave properties are
surface velocity (which are correlated to the wind with coefficient
linearised to give the last term in (15) (following Mellor, 2003),
of 0.7). It is therefore likely that the most significant improve-
similarly otherwise important bathymetry slopes (Ardhuin et al.,
ments in the simulation of the residual circulation could be made
2008) are neglected.
by (i) better constraining the accuracy of the diapycnal fluxes
Noting that sζ ¼ 1 þ ða=HÞcosψ and d=dz ¼ s−1
ζ d=dζ, the follow-
(riverine input and buoyancy forcing) in the domain, (ii) thor-
ing transformed and phase averaged momentum components can
oughly investigating the lateral dispersal of freshwater, and
be found:
(iii) investigating the sensitivity to the wind forcing.
Finally given the cost limitations in making long term current a2
〈sζ ∇p〉 ¼ ρg∇〈η〉− 1 þ gζ∇ρ ð16Þ
profile measurements and the accuracy issues with current high 2H 2
resolution numerical simulations, a hybrid method is proposed to
2 n
deduce the residual current whereby eddy viscosities are extracted ∂2 u 1 ∂2 un ∂2 U a a2 ∂ u
sζ 2 ¼ ¼ − 2 cos ψ þ 2 cos2 ψ−⋯ ;
from the model and combined with field estimates of density ∂z sζ ∂ζ 2
∂ζ 2 H H ∂ζ 2
gradient. This technique has remarkable success at Site B where, ð17Þ
using observationally derived CTD estimates and POLCOMS
derived viscosity and friction estimates, it predicts a profile that where the largest component of the last term, 〈2ða=HÞcos ψ
is closer to the independently observed ADCP estimates than it is ∂2 uðnÞ =∂ζ 2 〉 is found in the model to be about 10% the size of
to the POLCOMS simulation. In essence the lateral density gradient ∂2 U=∂ζ 2 at both Sites A and B. Therefore the momentum equations
in the POLCOMS simulation is corrected with the observed values. can be re-expressed in terms of phase average mean terms:
This tool cheaply combines multiplatform datasets to make ∇pn ∂2 U n a
n
improved predictions of the residual circulation in Liverpool Bay ifU ¼ − þν þO ð18Þ
ρ ∂ζ 2 H
that would otherwise be prohibitively expensive to directly
measure with multiple moorings, and are currently not well where ∇pn ¼ ρg∇〈η〉−gð1 þ πa2 =H 2 Þζ∇ρ≈ρg∇〈η〉−gζ∇ρ. Having trans-
represented in numerical models. It is speculated that this tool formed momentum terms into surface stretched coordinates we
could be operationally useful for managing dispersal of anthro- drop the angle bracket, star notation and Oða=HÞ terms. The problem
pogenic passive tracers. is reduced to that with a static free surface. In practise, the difference
between residual currents computed with this vertical coordinate
system and regular sigma coordinates is not large, with a depth mean
Acknowledgements velocity error of about 0:5 cm s−1 . Also, the truncation error is
smaller than might be expected from a direct a=H assessment
We are truly indebted to the conscientious reviews that because the truncation terms are also a function of the phase
resulted in a significantly improved manuscript. Data was col- correlation between the tidal wave amplitude and its velocity
lected and Palmer was supported under the National Oceanogra- (Pingree and Maddock, 1985; Winant, 2008), and in Liverpool Bay
phy Centre's Liverpool Bay Observatory (funded under NERC the phase correlation is small because the tide propagates like a
National Capability). Howarth was supported under EU FP7 standing wave. In the Rhine ROFI, however, the tide has a progressive
FIELD_AC project (EC Grant 242284). Polton was funded by a NERC wave character and so these discarded terms will not be so small.
New Investigator Award (NE/I002103/1).
References
Appendix A. Shallow water tidal phase averaging Ardhuin, F., Rascle, N., Belibassakis, K., 2008. Explicit wave-averaged primitive
equations using a generalized lagrangian mean. Ocean Modelling 20, 35–60.
Bowden, K.F., 1965. Horizontal mixing in the sea due to a shearing current. Journal
In the previous studies the time independence of the momen-
of Fluid Mechanics 21, 83–95.
tum equation (1) is achieved by taking a sufficiently long time Bowden, K.F., El Din, S.H.S., 1966. Circulation and mixing processes in the Liverpool
average, filtering out the tidal contributions that vary temporally Bay area of the Irish Sea. Geophysical Journal of Royal Astronomic Society 11,
but not vertically. In such shallow environments, however, it is 279–292.
Canuto, V.M., Howard, A., Cheng, Y., Dubovikov, M.S., 2001. Ocean turbulence. Part
also important to quantify the impact that the tidal averaging, or I: One-point closure model—momentum and heat vertical diffusivities. Journal
filtering, has on the vertical distribution of momentum. Since the of Physical Oceanography 31, 1413–1426.
126 J.A. Polton et al. / Continental Shelf Research 59 (2013) 115–126
Fairall, C.W., Bradley, E., Hare, J., Grachev, A., Edson, J., 2003. Bulk parameterization Journal of the Marine Biological Association of the United Kingdom 65,
of airsea fluxes: updates and verification for the COARE algorithm. Journal of 969–982.
Climate 16, 571–591. Polton, J.A., Palmer, M.R., Howarth, M.J., 2011. Physical and dynamical oceanogra-
Fischer, H.B., 1976. Mixing and dispersion in estuaries. Annual Review of Fluid phy of Liverpool Bay. Ocean Dynamics 61, 1421–1439.
Mechanics 8, 107–133. Ponte, A.L., 2010. Periodic wind-driven circulation in an elongated and rotating
Geyer, W.R., 1997. Influence of wind on dynamics and flushing of shallow estuaries. basin. Journal of Physical Oceanography 40, 2043–2058.
Estuarine, Coastal and Shelf Science 44, 713–722. Ponte, A.L., Gutiérrez de Velasco, G., Valle-Levinson, A., Winters, K.B., Winant, C.D.,
Gurgel, K.W., Antonischki, G., 1997. Remote sensing of surface currents and waves 2012. Wind-driven subinertial circulation inside a semienclosed bay in the Gulf
by the HF radar WERA. In: IEE Proceedings, Electronic Engineering in Oceano- of California. Journal of Physical Oceanography 42, 940–955.
graphy, pp. 211–217. Prandle, D., 1982. The vertical structure of tidal currents. Geophysical and Astro-
Heaps, N.S., 1972. Estimation of density currents in the Liverpool Bay area of the physical Fluid Dynamics 22, 29–49.
Irish Sea. Geophysical Journal of Royal Astronomical Society 30, 415–432. Prandle, D., 1987. The fine-structure of nearshore tidal and residual circulations
Heaps, N.S., Jones, J.E., 1977. Density currents in the Irish Sea. Geophysical Journal of revealed by H.F. radar surface current measurements. Journal of Physical
Royal Astronomic Society 51, 393–429. Oceanography 17, 231–245.
Hendershott, M.C., Rizzoli, P., 1976. The winter circulation of the Adriatic Sea. Deep Prandle, D., Matthews, J., 1990. The dynamics of nearshore surface currents
Sea Research 23, 353–370. generated by tides, wind and horizontal density gradients. Continental Shelf
Holt, J., James, I.D., 1999. A simulation of the Southern North Sea in comparison Research 10, 665–681.
with measurements from the North Sea Project. Part I: Temperature. Con- Prandle, D., Ryder, D., 1989. Comparison of observed (HF radar) and modelled
tinental Shelf Research 19, 1087–1112. nearshore velocities. Continental Shelf Research 9, 941–963.
Holt, J., James, I.D., 2001. An s-coordinate model of the northwest European Ralston, D.K., Geyer, W.R., Lerczak, J.A., 2008. Subtidal salinity and velocity in the
continental shelf. Part I: Model description and density structure. Journal of Hudson River Estuary: observations and modeling. Journal of Physical Oceano-
Geophysical Research 106, 14015–14034. graphy 38, 753–770.
Holt, J., Umlauf, L., 2008. Modelling the tidal mixing fronts and seasonal stratifica- Rippeth, T.P., Fisher, N.R., Simpson, J.H., 2001. The cycle of turbulent dissipation in
tion of the Northwest European Continental shelf. Continental Shelf Research the presence of tidal straining. Journal of Physical Oceanography 31,
2458–2471.
28, 887–903.
Robinson, A.M., Wyatt, L.R., Howarth, M.J., 2011. A two year comparison between
Holt, J.T., Allen, J.I., Proctor, R., Gilbert, F., 2005. Error quantification of a high-
HF radar and ADCP current measurements in Liverpool Bay. Journal of
resolution coupled hydrodynamic-ecosystem coastal-ocean model: Part
Operational Oceanography 4, 33–45.
1 model overview and assessment of the hydrodynamics. Journal of Marine
Ruddick, K.G., Deleersnijder, E., Luyten, P.J., Ozer, J., 1995. Haline stratification in the
Systems 57, 167–188.
Rhine-Meuse fresh water plume: a three-dimensional model sensitivity analy-
Howarth, M.J., Balfour, C., Player, R., Polton, J.A. Coastal density gradients near a
sis. Continental Shelf Research 15, 12507–12630.
macro-tidal estuary. Continental Shelf Research (unpublished).
Sanay, R., Valle-Levinson, A., 2005. Wind-induced circulation in semienclosed
Howarth, M.J., Palmer, M.R., 2011. The Liverpool Bay coastal observatory. Ocean
homogeneous, rotating basins. Journal of Physical Oceanography 35,
Dynamics 61, 1917–1926.
2520–2531.
Ianniello, J.P., 1977. Tidally induced residual currents in estuaries of constant
Simpson, J.H., 1997. Physical processes in the ROFI regime. Journal of Marine
breadth and depth. Journal of Marine Research 35, 755–786. Systems 12, 3–15.
James, I.D., 1996. Advection schemes for shelf sea models. Journal of Marine
Simpson, J.H., Brown, J., Matthews, J., Allen, G., 1990. Tidal straining, density
Systems 8, 237–254. currents, and stirring in the control of estuarine stratification. Estuaries 13,
Janzen, C.D., Wong, K.C., 2002. Wind-forced dynamics at the estuary-shelf interface 125–132.
of a large coastal plain estuary. Journal of Geophysical Research 107. Simpson, J.H., Burchard, H., Fisher, N.R., Rippeth, T.P., 2002. The semi-diurnal cycle
Li, Y., Li, M., 2011. Effects of winds on stratification and circulation in a partially of dissipation in a ROFI: model-measurement comparisons. Continental Shelf
mixed estuary. Journal of Geophysical Research 116. Research 22, 1615–1628.
MacCready, P., Geyer, W.R., 2010. Advances in estuarine physics. Annual Review of Verspecht, F., Rippeth, T.P., Simpson, J.H., Souza, A.J., Burchard, H., Howarth, M.J.,
Marine Science 2, 35–58. 2009. Residual circulation and stratification in the Liverpool Bay region of
Mathieu, P.P., Deleersnijder, E., Cushman-Roisin, B., Beckers, J.M., Bolding, K., 2002. freshwater influence. Ocean Dynamics 59, 765–779.
The role of topography in small well-mixed bays, with application to the lagoon Verspecht, F., Simpson, J.H., Rippeth, T.P., 2010. Semi-diurnal tidal ellipse variability
of Mururoa. Continental Shelf Research 22, 1379–1395. in a region of freshwater influence. Geophysical Research Letters 37.
Mellor, G.L., 2003. The three-dimensional current and surface wave equations. Visser, A.W., Souza, A.J., Hessner, K., Simpson, J.H., 1994. The effect of stratification
Journal of Physical Oceanography 33, 1978–1989. on tidal current profiles in a region of freshwater influence. Oceanologica Acta
O'Neill, C.K., Polton, J.A., Holt, J.T., O'Dea, E.J., 2012. Modelling temperature and 17, 369–381.
salinity in Liverpool Bay and the Irish Sea: sensitivity to model type and surface Winant, C.D., 2004. Three-dimensional wind-driven flow in an elongated, rotating
forcing. Ocean Science 8, 903–913. basin. Journal of Physical Oceanography 34, 462–476.
Palmer, M.R., 2010. The modification of current ellipses by stratification in the Winant, C.D., 2006. Three-dimensional wind-driven coastal circulation past a
Liverpool Bay ROFI. Ocean Dynamics 60, 219–226. headland. Journal of Physical Oceanography 36, 1430–1438.
Palmer, M.R., Polton, J.A., 2011. A strain induced freshwater pump in the Liverpool Winant, C.D., 2007. Three-dimensional tidal flow in an elongated, rotating basin.
Bay ROFI. Ocean Dynamics 61, 1905–1915. Journal of Physical Oceanography 37, 2345–2362.
Phelps, J.J.C., Polton, J.A., Souza, A.J., Robinson, L.A. Hydrodynamic timescales in a Winant, C.D., 2008. Three-dimensional residual tidal circulation in an elongated,
hyper-tidal region of freshwater influence. Continental Shelf Research (sub- rotating basin. Journal of Physical Oceanography 38, 1278–1295.
mitted for publication). Young, W.R., Rhines, P.B., Garrett, C.J.R., 1982. Shear-flow dispersion, internal waves
Pingree, R.D., Maddock, L., 1985. Stokes, Euler and Lagrange aspects of residual tidal and horizontal mixing in the ocean. Journal of Physical Oceanography 12,
transports in the english channel and the southern bight of the north sea. 515–527.