Development and application of a three-dimensional baroclinic model to the study of the seasonal circulation in the Celtic Sea

ARTICLE IN PRESS

Continental Shelf Research 24 (2004) 13–36

Development and application of a three-dimensional baroclinic model to the study of the seasonal circulation in the Celtic Sea E.F. Younga,*, J. Browna,1, J.N. Aldridgea, K.J. Horsburghb, L. Fernanda a

The Centre for Environment, Fisheries & Aquaculture Science, Pakefield Road, Lowestoft, Suffolk NR33 0HT, UK b School of Ocean Sciences, University of Wales Bangor, Menai Bridge LL59 5EY, UK Received 8 November 2002; received in revised form 11 August 2003; accepted 15 September 2003

Abstract A three-dimensional density-resolving model based on the Princeton Ocean Model (POM) has been developed for the simulation of tide, wind and density-driven flows in a region of the northwest European shelf extending from the Celtic Sea to the Sea of the Hebrides. Predicted co-tidal charts of the region are in good agreement with published charts based on observed tidal elevations and from previous modelling studies. Comparisons of observed and predicted M2 and S2 tidal elevations and currents suggest that in general, this model is of comparable or greater accuracy than previous large area models of the region, and tidally generated mixing is of the correct magnitude to enable accurate simulations of the location of tidal mixing fronts. The ability of the model to predict observed temperatures in the region was assessed by comparison with a comprehensive seasonal hydrographic data set collected in the Irish Sea in 1995. The modelled seasonal cycle of thermal stratification at a site in the western Irish Sea was in good agreement with a time series of observed temperatures. A statistical evaluation of model accuracy using all available data showed mean and root mean square (RMS) errors in near-surface temperatures of 0.25
C and 0.72
C, and of 0.05
C and 0.44
C in near-bed temperatures. A consideration of the geographical and temporal distribution of errors showed that the largest errors were near the start of the simulation when the model was consistently too warm, suggesting an inadequate representation of the initial conditions. Model predictions for the Celtic Sea in summer 1998 also compared well with a comprehensive spatial survey, with mean and RMS errors in near-surface temperatures of 0.18
C and 0.70
C, and of 0.01
C and 0.74
C in near-bed temperatures. Errors are in part due to inaccuracies in the initial conditions and the neglect of salinity variations in the model, in particular freshwater inputs in the Bristol Channel. Using the modelled flow fields to drive a particle-tracking model, the broad pattern of observed Lagrangian transport, an essentially cyclonic circulation pattern following the contours of bottom density, was successfully predicted. The baroclinic component of flow contributed 91% to the net westward flux along the frontal region in St. George’s Channel, clearly demonstrating the importance of including the baroclinic component in models of the northwest European shelf. Whilst the model ably reproduces the broad scale features of the hydrodynamics observed in the Irish and Celtic Seas region, comparisons of the predicted vertical structure of thermal stratification with high frequency data sets suggest that the vertical resolution of the model is insufficient to accurately reproduce strong vertical gradients. Reproduction

*Corresponding author. Current address: Proudman Oceanographic Laboratory, Bidston Observatory, Bidston Hill, Prenton, Merseyside CH43 7RA, UK. Tel.: +44-151-653-1532; fax: +44-151-653-6269. E-mail address: [email protected] (E.F. Young). 1 Current address: British Oceanographic Data Centre, Bidston Observatory, Bidston Hill, Prenton, Merseyside CH43 7RA, UK. 0278-4343/$ - see front matter Crown Copyright r 2003 Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.csr.2003.09.003

ARTICLE IN PRESS 14

E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

of the fine scale detail revealed by these data is currently the most stringent test for numerical models of the shelf seas. With increasing computer power and the development of higher resolution models, such data will act as the template against which to assess model performance. Crown Copyright r 2003 Published by Elsevier Ltd. All rights reserved. Keywords: Thermal stratification; Jets; Residual flow; Drifters; Tidal prediction; Irish Seas

1. Introduction There is a growing body of observational evidence (for example, Hill et al., 1994; Horsburgh et al., 2000; Brown et al., 1999a; Brown et al., 2003) that circulation on the northwest European shelf is strongly seasonal in character. In summer, localised but intense baroclinic flows develop associated with the margins of dense bottom-water pools. These are often formed when cold winter water is left trapped and isolated in topographic depressions after the onset of summer stratification. The discovery of the flows and understanding of the factors which drive them have led to a significant redefinition of the seasonal circulation of the shelf seas. It is apparent that extensive areas of the continental shelf are dominated by strong, persistent and stable jet-like currents transporting water and material over many hundreds of kms. These flows can provide rapid and organised pathways for the transport of pollutants and the larval stages of marine fish. They may also act as retention mechanisms about topographic depressions. For example, a cyclonic gyre observed in the western Irish Sea in spring and summer provides a physical mechanism whereby the pelagic larvae of the commercially valuable Norway lobster, Nephrops norvegicus, are retained close to the muddy substrate they require for settlement (Brown et al., 1995; Hill et al., 1996). The well-defined boundaries between mixed and stratified waters also act as important and extensive sites for phytoplankton and fisheries production, and play an important role in nutrient and contaminant dynamics. Historically, shelf models used for the prediction of contaminant transport and as tools for the management of fisheries have used a simplistic approach with water velocities averaged in the vertical from a tidal data base, a simple para-

meterisation of wind-driven flows, and no consideration of baroclinic flows. Recent progress in observational studies with the collection of very high resolution data (for example, Brown et al., 1999a, 2003), has highlighted the inadequacies of these models, which fail to capture many of the observed features of the circulation. With advances in computing power, it has become desirable to develop more sophisticated models that include more of the essential physics. The application of three-dimensional, density-resolving models to the simulation of the complex dynamical processes in shelf regions has thus increased over recent years (for example, Evans & Middleton, 1998; Holt and James, 1999, 2001; Xing & Davies, 2001; Horsburgh & Hill, 2003). In this study, a three-dimensional model based on the Princeton Ocean Model (POM) (Blumberg & Mellor, 1987) has been developed. POM is a three-dimensional, primitive equation, s co-ordinate, free surface model. It has been successfully used to simulate tidal and density-driven circulations for a wide range of applications (e.g. Oey et al., 1985; Galperin & Mellor, 1990; Yang & Weisberg, 1999; Zavatarelli & Mellor, 1995) and includes the essential physics required to model the seasonal evolution of baroclinic flows in the European shelf seas. In this paper, the application of the model to the prediction of tidal, wind and density-driven flows in a region of the northwest European shelf extending from the Celtic Sea in the south to the Sea of the Hebrides in the north (Fig. 1) will be described. As the extent of stratified regions is strongly dependent on the strength of tidal mixing, it is important that the model accurately reproduces the main tidal constituents (M2 and S2) in the region. This will be demonstrated by a comparison with observed elevations and velocities. The ability of the model to predict

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

15

Fig. 1. Map showing the location and bathymetry of the model grid, indicating locations referred to in the text and CTD sites A and B.

the seasonal cycle of stratification will be assessed by comparison with an extensive hydrographic data set collected in the Irish Sea in 1995. Finally, the model will be used to predict the development and main characteristics of the summer circulation in the Celtic Sea, and the importance of the baroclinic component of the flow will be assessed.

2. Model description The basic model used Princeton Ocean Model which may be found in (1987). Briefly, the POM

for this study was the (POM), full details of Blumberg and Mellor is a three-dimensional,

fully non-linear, primitive equation model. It has a free surface and a bottom-following s co-ordinate system, and includes the second moment turbulence closure scheme of Mellor and Yamada (1982) to provide vertical mixing coefficients. The equations are solved on an Arakawa C grid in the horizontal (Mesinger and Arakawa, 1976) using a mode splitting technique, whereby the governing equations are split into depth-integrated (external mode) and three-dimensional (internal mode) parts. The external mode equations are solved using a short time step to satisfy the Courant– Friedrichs–Lewy (CFL) condition, whilst the internal mode calculations use a much longer time step of 10 or more times the external time step. The

ARTICLE IN PRESS 16

E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

equations are approximated using leap-frog time differencing, and a weak filter is used to prevent time splitting of the solutions at the odd and even time steps, according to Asselin (1972). Horizontal mixing was included using a parameterisation due to Smagorinsky (1963). The default scalar advection scheme in POM (a forward-time, centred-space (FTCS) scheme) was considered unsuitable for this study: it is not positive-definite and can introduce spurious concentrations and unphysical ripples at fronts. Therefore, the Smolarkiewicz (1983) positivedefinite, low implicit diffusion scheme was implemented. Here, an upwind scheme is recursively applied, with numerical diffusion corrected at each iteration using an antidiffusion velocity based on the local, first-order truncation error. Tests with an idealised basin scenario suggested that 3 iterations produced the optimum compromise between accuracy and speed of simulation. Full details of the model implementation specific to this study may be found in Young (2002); a brief summary is provided here. Surface wind stresses are calculated using ðtsx ; tsy Þ ¼ 1:25CDS ðu210 þ v210 Þ1=2 ðu10 ; v10 Þ;

ð1Þ

where tsx ; tsy are the surface wind stress components in the x and y directions, respectively, the drag coefficient CDS is dependent on the wind velocity following the formulation of Large and Pond (1981), and u10 ; v10 are the near surface wind velocities in the x and y directions, respectively. In the model simulations described in this paper, salinity was assumed constant. The surface flux of heat was calculated using the bulk formulae discussed in Gill (1982), with heat losses due to longwave radiation corrected for cloud cover following Reed (1976). Spatially varying winds, air temperatures and dewpoint temperatures were obtained from the European Centre for Mediumrange Weather Forecasting (ECMWF) on a horizontal resolution of 1.125
at 6-h intervals. These were spatially and temporally interpolated to provide forcing data for each model grid point at each time step. For 1995, solar shortwave radiation was specified using hourly observations from Dublin Airport. Spatially varying daily

averages of both solar short-wave radiation and cloud cover from ECMWF were used for 1998. At the open boundary, for the barotropic (depth mean) flow, a radiation condition is applied of the form (Aldridge and Davies, 1993) rffiffiffi g ð2Þ ðz  zT Þ; q ¼ qT þ h where q is the normal component of depth mean current and z is the sea surface elevation. The imposed tidal elevations, zT ; and depth mean currents, qT ; were obtained from the Proudman Oceanographic Laboratory (POL) storm surge model (Flather et al., 1991) for the two dominant tidal constituents (M2 and S2) and interpolated to provide forcing terms at each open boundary point. Use of this radiation condition allows the open boundary to adjust to internally generated waves. Tangential velocities at the open boundary are set to zero. The transport of heat and salt at boundaries where outflow occurs is determined using an upwind scheme. When the flow is into the model domain, a zero gradient boundary condition is applied. The model domain is shown in Fig. 1. To include the effects of far field forcing and limit the influence of the open boundaries on the region of interest, the model extends from the Celtic Sea in the south to the Sea of the Hebrides in the North. The grid had 150 columns and 250 rows with cell dimensions of 1/20
of longitude and 1/30
of latitude to give a resolution of approximately 3.3 km (Dx)  3.7 km (Dy). Fifteen sigma layers were used in the vertical, arranged as follows: s=(0, 0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.85, 0.9, 0.95, 1.0). This irregular spacing provided enhanced resolution in the high-shear surface and near-bed layers. Model depths were obtained by smoothing a fine resolution (approximately 1 km) bathymetry digitised from Admiralty Fair Sheets (where available) and Admiralty Charts supplied by the UK Hydrographic Office (Brown et al., 1999b). Solutions were generated from initial conditions of zero elevation and motion, with model output commencing after 5 days to allow for the dissipation of initial transients. Salinity was assumed constant throughout the model simulations. The

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

implications of this are discussed in later sections. The initial temperature field was obtained from AVHRR images of weekly averaged sea surface temperatures (NASA Physical Oceanography Distributed Active Archive Center at the Jet Propulsion Laboratory, California Institute of Technology) for the week prior to the start of the model simulations (21st March). It was assumed that the water column is well-mixed at this time of year, and the temperatures obtained from the night-time pass were used as they were considered to be a more suitable indicator of temperature throughout the water column. Sea surface temperatures from AVHRR are accurate to within 0.5
C, and a discussion of the errors may be found at http://podaac.jpl.nasa.gov. Unfortunately, due to the high incidence of cloud cover at these latitudes on the northwest European shelf, temperatures over some of the model domain could only be obtained by interpolation, which invariably introduces errors into the initial temperature fields. The potential influence of the initial conditions on model predictions is discussed in later sections. All model integrations were started on 21st March using a time step of 15 s for the external mode and 300 s for the internal mode. The two specific years considered in this study for which adequate data were available were 1995 and 1998.

3. Model validation 3.1. Tides Thermal stratification occurs where there is insufficient tidally generated turbulent energy to maintain mixing against the input of surface buoyancy through solar heating (Simpson and Hunter, 1974). Thus, for the prediction of seasonal cycles of stratification in the shelf seas, it is important to use a model that accurately describes the main tides in the region. For the tidal validation simulations, the model was run with fixed temperature and salinity for a period of 18 days, including an initialisation period of 3 days, which is sufficient to extract the two dominant

17

tidal constituents in the Celtic and Irish Sea region, M2 and S2. 3.1.1. M2 tidal elevations and currents The predicted amplitude and phase diagrams for M2 are shown in Figs. 2a and b. These are in good agreement with charts of the region based on observations (Robinson, 1979) and with previous modelling studies (e.g. Pingree and Griffiths, 1987; Davies and Jones, 1992; Sinha and Pingree, 1997; Young et al., 2000). Notable are the positions of the degenerate amphidrome in the Irish Sea on the east coast of Ireland, and a second amphidrome in the North Channel. M2 tidal amplitudes increase rapidly in the shallower regions such as the eastern Irish Sea and the Bristol Channel, reaching a maximum of approximately 4.2 m in the Bristol Channel. The co-tidal chart reveals a weak local instability within 3 grid cells of the open boundary. In the northwest of the model, this propagates into the model domain for a further 10 grid cells before dissipating. However, it does not affect the solution in the main regions of interest (the Irish and Celtic Seas) so was considered unimportant in the context of this study. Predicted elevations and currents were compared with previous CEFAS observations and those held at the British Oceanographic Data Centre (BODC). A total of 95 tide gauge and 142 current metre observations at variable depth in the water column were used in the comparison, a detailed description of which may be found in Young (2002). A summary of the model performance is shown in Tables 1 and 2 in the form of error histograms and RMS errors. As found in previous modelling studies (e.g. Aldridge and Davies, 1993; Jones and Davies, 1996; Young et al., 2000) validating the model for accurate elevations resulted in an over prediction of current magnitude. In this study, the location of tidal mixing fronts was a primary consideration, placing a greater emphasis on the accuracy of modelled currents. The results presented here are therefore a compromise; whilst the currents are still slightly over predicted, the elevations tend to be too low. This model does not reproduce the observed elevations and currents as accurately as the fine resolution model of Young et al. (2000) due to the

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

18

(a)

(b)

Fig. 2. (a) Predicted amplitude (cm) of the M2 tidal constituent. Contours are plotted at 15 cm intervals. (b) Predicted phase (
) of the M2 tidal constituent. Contours are plotted at 20
intervals.

Table 1 Error frequency histogram for M2 tidal elevations showing the number of points where the computed amplitude or phase exceeds or is below the observed value Number of points

Number of points

30 25 20 15 10 5 5 h 1 g 0

0 1

0 1

Table 2 Error frequency histogram for M2 tidal currents showing the number of points where the computed semi-major axis amplitude, phase or orientation exceeds or is below the observed value

3 2

30 9

30 37

10 15 20 25 30

11 11 2 36 3 3

1 0

0 2

6 1

Amplitude h expressed as % error relative to the observed value, phase g as degrees. Error is modelled minus observed value. RMS errors in h and g are 13.23% and 7.14
, respectively.

coarser model grid used in this study. This is of particular relevance for currents where the high spatial variability and dependence on adequate representation of complex bathymetry in the model is of greater importance (Jones & Davies, 1996). However, these predictions are of improved

30 25 20 15 10 5 5 A 6 g 0 O 3

1 1 2

5 5 3

3 4 3

10 13 7

12 18 16

10 15 20 25 30

11 19 15 9 51 24 6 8 39 32 14 8

16 4 4 2 3 4

Amplitude A as % error relative to the observed value, phase g and orientation O in degrees. Error is modelled minus observed value. RMS errors in semi-major axis and phase are 17.28% and 11.81
, respectively.

or comparable accuracy to equivalent large area models of the region (e.g. Davies and Jones, 1992; Sinha and Pingree, 1997) and suggest that the model is sufficient for the purposes of this study.

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

3.1.2. S2 tidal elevations and currents The predicted amplitude and phase diagrams for S2 are shown in Figs. 3a and b. As with the M2 tidal constituent, these are in good agreement with charts of the region based on observations (Robinson, 1979) and with previous modelling studies (e.g. Pingree and Griffiths, 1987; Davies and Jones, 1992; Young et al., 2000). The amphidromic system appears very similar to that of M2 shown in Figs. 2a and b with a degenerate amphidrome in the Irish Sea and a second amphidrome in the North Channel. The latter is located slightly further to the south east for the S2 tidal constituent. S2 tidal amplitudes are approximately one third the magnitude of the M2 amplitudes and have a similar spatial distribution with increasing amplitudes in the shallow regions of the eastern Irish Sea and the Bristol Channel. The maximum amplitude of approximately 1.5 m is achieved in the Bristol Channel. As for M2, the

19

co-tidal chart reveals a weak local instability near the open boundaries of the model; this is discussed in the previous section. Modelled elevations and currents were compared with observations from 68 tide gauges and 142 current meters, a detailed description of which may be found in Young (2002). A summary of the model performance is shown in Tables 3 and 4 in the form of error histograms and RMS errors. As for the M2 tidal constituent, the results presented here represent a compromise, with currents and elevations slightly over and under predicted, respectively. This model reproduces the observed elevations more accurately than the fine resolution model of Young et al. (2000). Current phases are also more accurate although the amplitudes are not, probably due to the relative coarseness of the grid as discussed for the M2 tidal constituent. However, these are of improved or comparable accuracy to equivalent large area models of the

Fig. 3. (a) Predicted amplitude (cm) of the S2 tidal constituent. Contours are plotted at 5 cm intervals. (b) Predicted phase (
) of the S2 tidal constituent. Contours are plotted at 20
intervals.

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

20

Table 3 Error frequency histogram for S2 tidal elevations showing the number of points where the computed amplitude or phase exceeds or is below the observed value Number of points 30 25 20 15 10 5 5 h 0 g 0

1 0

1 3

0 1

17 9

30 24

10 15 20 25 30

7 3 25 4

2 2

3 0

1 0

3 0

Amplitude h expressed as % error relative to the observed value and phase g as degrees. Error is modelled minus observed value. RMS errors in h and g are 12.01% and 5.68
, respectively.

Table 4 Error frequency histogram for S2 tidal currents showing the number of points where the computed semi-major axis amplitude, phase or orientation exceeds or is below the observed value Number of points 30 25 20 15 10 5 5 A 4 g 2 O 3

4 7 2

5 1 1

6 3 7

7 11 9

9 23 17

10 15 20 25 30

15 12 11 8 8 32 29 11 10 4 38 28 10 5 6

7 3 4

Amplitude A as % error relative to the observed value, phase g and orientation O in degrees. Error is modelled minus observed value. RMS errors in semi-major axis and phase are 20.77% and 10.09
, respectively.

region (e.g. Davies and Jones, 1992) and overall, the model is sufficiently accurate for the purposes of this study. 3.2. Temperature: 1995 The baroclinic component of the model was validated using the extensive data set presented by Horsburgh et al. (2000) collected on 14 cruises spanning January 13–December 11, 1995, involving the CEFAS research vessels Corystes and Cirolana, and the University of Wales Bangor (UWB) research vessel Prince Madog. These hydrographic data were obtained with profiling CTDs, the Scanfish towed undulating CTD (Brown et al., 1996; Fernand, 1999) and the UWB undulating CTD, Searover (Bauer et al., 1985). Near-surface and near-bed temperatures

were extracted at each profile location for comparison with model predictions. For the Scanfish and Searover data sets, values at the deepest point in each undulation and at approximately 4 m depth were extracted. A full description of the validation exercise is beyond the scope of this paper; a summary is presented here. The predicted near-surface and near-bed temperatures at location A (Fig. 1) in the centre of the western Irish Sea gyre are shown in Fig. 4 and illustrate the development and breakdown of thermal stratification in the gyre. These show good agreement with a time series of observed temperatures compiled by Horsburgh (1999) using data from this location and neighbouring station B (Fig. 1). The near-bed temperatures in particular agree well with the observed trend. The gyre is formed when relict winter water is left trapped in a topographic depression after the onset of thermal stratification in spring. This trapped water is warmed only very slowly by weak diffusion of heat across the seasonal thermocline (Hill et al., 1997) and, compared to the mixed surface waters, is relatively isolated from the influences of surface and model boundary forcing. The good agreement with observations suggests that the model realistically represents the dynamics of this cold pool. A more quantitative assessment of model performance is given by the statistical comparison of pairs of observed temperatures and the corresponding predicted values for each location and time. Pairs of observed and predicted near-surface and near-bed temperatures are plotted in Figs. 5a and b, respectively, and Table 5 shows the mean error, RMS and correlation coefficient for nearbed and near-surface temperatures. The agreement of both surface and bottom temperatures is good, with slightly better agreement for bottom temperatures. Here, it is important to note that there is an essential difference between modelled and observed temperatures; whilst the model outputs diurnally averaged temperature, the observations of surface temperatures show a distinct diurnal signal with fluctuations of the order 1
C. Thus some discrepancy between observed and modelled surface values is to be expected. Insight into the possible sources of model errors may be gained by considering the geographical

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

21

Fig. 4. Predicted surface and near-bed temperatures for site A in the western Irish Sea (Fig. 1) in 1995. Observed temperatures are indicated by } (surface) and  (near-bed).

distribution of errors and their temporal variability. Figs. 6a and b show the errors in modelled near-surface and near-bed temperatures for 2 weeks after the start of the simulation and are thus an indication of errors in the initial conditions. It can be seen that the predicted temperatures are generally too warm, which suggests that the initial temperature field was too warm. The temperature fields used to initialise the model are interpolated from the available AVHRR data. These may be rather sparse due to cloud cover, thus introducing a source of error. In addition, AVHRR data are a measure of the surface temperatures and by applying these values throughout the water column it is implicitly assumed that the Irish and Celtic Seas are well

mixed at the start of the simulation (March 21). However, near-coast regions may be stratified due to significant freshwater inputs (DickeyCollas et al., 1997; Horsburgh et al., 2000). This is likely to be the cause of the under prediction of the surface to bed temperature difference (TSB ) (Fig. 6c) near the Irish coast at the start of the simulation. Figs. 7a and b show the errors in modelled nearsurface and near-bed temperatures during spring (days 80–151). Predicted temperatures in the Eastern Irish Sea are no longer consistently warmer than observed, which suggests that the effect of errors in the initial conditions is of limited duration in this region. Near-bed temperatures are generally in good agreement with observations in

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

22

(a)

(b)

Fig. 5. Comparison of observed and modelled near-surface (a) and near-bed (b) temperatures for data collected in the Irish Sea in 1995. Table 5 Statistical comparison of observed temperatures and the corresponding predicted values for the Irish Sea in 1995, and Celtic Sea in 1998 RMS

R2

Irish Sea 1995 (n ¼ 4422) Near-surface 0.25 Near-bed 0.05

0.72 0.44

0.905 0.959

Celtic Sea 1998 (n ¼ 7256) Near-surface 0.18 Near-bed 0.01

0.70 0.74

0.685 0.882

Mean

Results for near-surface and near-bed temperatures presented as mean error, root mean square (RMS) error, and correlation coefficient, R2 :

the Western Irish Sea although they tend to be too warm near the Irish coast. Whilst surface temperatures are also too warm in this near coast region, the TSB (Fig. 7c) is weaker than observed suggesting a continuing influence of the freshwater inputs discussed earlier. Surface temperatures are generally too warm over the western gyre region, which corresponds to the over prediction of surface temperatures at site A (Fig. 4) at this time.

As near bed temperatures are in good agreement here, the errors are likely due to inadequate representation of the atmospheric forcing. During the summer period (days 152–243) predicted near-surface and near-bed temperatures tended to be slightly lower than observed (not shown). However, there was no geographical bias, as illustrated by the TSB errors shown in Fig. 8a. This implies that the imposed atmospheric forcing was generating insufficient heat flux into the model across the whole domain during the summer months. However, comparisons of modelled and observed temperatures during autumn (days 244– 334) show very good agreement, with the majority of near-surface and near-bed temperatures within 0.5
C of the corresponding observed value (not shown), and predicted TSB also in good agreement (Fig. 8b).

4. Model simulations: 1998 The comparison of model predictions with the spatially and temporally extensive data set available for 1995 demonstrated the ability of the

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

(a)

23

(b)

(c) Fig. 6. Geographical distribution of errors in temperature prediction (modelled minus observed) for the first 2 weeks of model simulation (days 80–94). (a) Near-surface temperatures, (b) near-bed temperatures and (c) surface to bed temperature difference. Errors (E) expressed as E>0.5
C (red +), 0.5
CXEX0.5
C (black &), Eo  0:5
C (blue  ).

model to accurately predict the seasonal cycle of thermal stratification in shelf waters. The model was now applied to the Celtic Sea region where there were two CEFAS cruises in July and August 1998. Although these did not obtain such a

temporally extensive data set as available for the Irish Sea in 1995, hydrographic observations using both profiling CTDs and the Scanfish towed undulating CTD were collected. In addition, 23 satellite-tracked drifters, with holey-sock

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

24

(b)

(a)

(c) Fig. 7. Geographical distribution of errors in temperature prediction (modelled minus observed) during spring (days 80–151). (a) Near-surface temperatures, (b) near-bed temperatures and (c) surface to bed temperature difference. Errors (E) expressed as E > 0:5
C (red +), 0:5
CXEX  0:5
C (black &), Eo  0:5
C (blue  ).

drogues 5.5 m long and 1.5 m diameter centred at 30 m, were deployed at locations shown in Fig. 9 to observe the Lagrangian transport in the region.

4.1. Evolution of thermal fronts Thermal stratification in the Celtic Sea develops rapidly, as illustrated in Fig. 10, which shows daily

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

(a)

25

(b)

Fig. 8. Geographical distribution of errors in surface to bed temperature difference (modelled minus observed) for: (a) summer (days 152–243) and (b) autumn (days 244–334). Errors (E) expressed as E > 0:5
C (red +), 0:5
CXEX  0:5
C (black &), Eo  0:5
C (blue  ).

averaged surface to bed temperature differences at weekly intervals. The thermocline first develops to the south of Ireland and over the Celtic Deep, strengthening and spreading quickly eastwards and northwards to penetrate the mouth of the Bristol Channel and the English Channel within a month. These predictions are in broad agreement with the theoretical description of Pingree (1980), although the initial thermocline development is predicted further east than suggested by Pingree. The timing is also somewhat later in 1998, with stratification developing towards the end of April, several weeks later than the general outline in Pingree (1980). However, model simulations of 1995 predict the onset of stratification during the first week of April. This demonstrates the sensitivity of the timing of the stratification development to interannual variability in atmospheric forcing. This has implications for larval transport in the region. The strong flows associated with the fronts between regions of stratified and well mixed waters (Brown et al., 2003) will not only provide rapid transport routes along the front but will also act as a barrier for material passing through

St. George’s Channel into the Irish Sea. Thus, the interaction between variability in the onset of thermal stratification and the timing of spawning may be a significant contributor to interannual variability in the recruitment of the larvae of marine species to the Irish Sea. 4.2. Evaluation of fluxes The predicted depth-averaged density-driven currents in the Celtic Sea, averaged over the period 21 July–20 August 1998, are shown in Fig. 11. They suggest a residual speed in the frontal regions of between 5 and 10 cm s1. To more clearly visualise the importance of these flows, the relative contribution of tides, winds and density to predicted flows through section AB to the northeast of the Celtic Deep (Fig. 9) was calculated for the same period in 1998. The location of the section was along longitude 5.9
W, and between latitudes 51.65
N and 52.1
N, thus encompassing the width of the front. The predicted fluxes and their relative magnitudes are shown in Table 6. The contribution due to the

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

26

53.0

25

52.5

A

25

202

75

52.0

25

25

B

75

Latitude

51.5

51.0

178

182

50.0

125

50.5

75 75

49.5 -9

-8

-7 Longitude

-6

-5

Fig. 9. Location of Scanfish sections 178, 182 and 202 and section AB used for flux estimates. Deployment positions of the drogued satellite tracked drifters are indicated by ’.

baroclinic component of flow, at 91% of the total westward flux and a relative magnitude of 86%, is clearly the most important influence on transport through this region during this summer period. This demonstrates the necessity to include a consideration of baroclinic flows in modelling studies of the European shelf waters. 4.3. Comparison with CTD data Modelled and observed temperatures were compared as described in Section 3.2. The results for mean error, RMS and correlation coefficient for near-bed and near-surface temperatures are shown in Table 5 and pairs of observed and

predicted near-surface and near-bed temperatures are plotted in Figs. 12a and b, respectively. The results show that model predictions of temperatures in the Celtic Sea in 1998 are slightly less accurate than those predicted for the Irish Sea in 1995, with a tendency to under predict the observed values. An understanding of potential sources of error in the model predictions can be obtained by considering the spatial distribution of errors. There are three main regions of error. The first is in the southwest where near-bed temperatures are generally too warm. Stratification and isolation of bottom water in this region can occur as early as the beginning of April (Pingree, 1980), which

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

(a)

27

(b)

(c)

(d)


Fig. 10. Predicted surface to bed temperature difference ( C) at weekly intervals from 29 April 1998 (Day 119) illustrating the development of thermal stratification.

suggests that modelled near-bed temperatures are too high from soon after the start of the simulation. This is probably due to errors in the initial temperature field, for reasons discussed in Section 3.2. The second and third regions of error are to the northwest of the Scilly Isles and in St. George’s Channel. In St. George’s Channel predicted near-surface temperatures are too cool,

and near-bed temperatures are too warm, whilst the converse is true to the northwest of the Scilly Isles. This suggests that the predicted region of stratification extends too far south and east near the Scilly Isles and not far enough north in St. George’s Channel. This may be due to inadequate representation of the bathymetry, the predicted levels of tidal and wind mixing may be

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

28

Fig. 11. Predicted density (thermal) driven depth-mean currents in the Celtic Sea, averaged over the period 21 July to 20 August 1998. Table 6 Predicted mean residual fluxes and relative magnitudes for tide, wind, and density driven flows through section AB to the north east of the Celtic Deep (Fig. 4), 21 July to 20 August 1998

Tides Wind Density

Predicted flux (  104 m3 s1)

Relative magnitude (%)

1.387 0.292 10.465

11.4 2.4 86.2

inaccurate, or buoyant inputs (either by surface heating or freshwater inputs) may be insufficient. The topography around the Scilly Isles is complex and poorly resolved by the 3.5 km grid of the model, where the averaging of the source

1 km bathymetric data set results in a gentler, shallower bathymetry. The potentially weaker vertical mixing is likely to contribute to the inaccurate prediction of the location of fronts in this region. However, this is less likely to produce the erroneous frontal position in St. George’s Channel. The model was carefully validated for accurate tidal flows in the Irish and Celtic Seas to ensure that the level of tidal mixing was appropriate. However, as discussed in Section 3.1, accurate simulations of tidal elevations tend to result in an over prediction of the current magnitudes. As shown in Tables 2 and 4, the compromise solution presented here still has a small tendency for over prediction of currents

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

(a)

29

(b)

Fig. 12. Comparison of observed and modelled near-surface (a) and near-bed (b) temperatures for data collected in the Celtic Sea in 1998.

and this may be generating too much mixing in St. George’s Channel resulting in errors in the frontal location. However, a reduction in tidal forcing at the open boundary of 5% produced insignificant change in the frontal position, thus inaccuracies in the tidal flows are unlikely to be the main source of error in this region. It is possible that insufficient heat is entering the model at the surface. Solar irradiance data were daily averaged values from the ECMWF. Errors in the atmospheric model predictions, and the potential errors introduced by using a daily averaged value, will contribute to erroneous predictions of surface heat flux. Insufficient surface heating may have resulted in the predictions of cooler surface temperatures than observed, and the lower levels of buoyancy input would contribute to errors in the predicted geographical location of the front in St. George’s Channel. The other source of buoyant input is fresh water, in particular in the Bristol Channel, which in this model is neglected. This is likely to be the main source of error in the prediction of the location of fronts and strength of frontal flows in the Celtic Sea. For most of the Celtic Sea, temperature

dominates (>85%) the vertical stratification (Brown et al., 2003). However, in some regions, in particular in the east around the Bristol Channel, salinity contributes up to 50% and in these areas the model will not perform as well. Fresh water from the Bristol Channel will increase buoyant inputs to the St. George’s Channel region, thus strengthening the density stratification. This will increase both the extent of the stratified area, and thus the location of the fronts, and the strength of the subsequent density-driven flows. The neglect of salinity variability in the model may also be impacting the region proximate to the Scilly Isles, in this case due to poor representation of the characteristics of inflowing North Atlantic water. Incorporation of freshwater influxes and salinity variability is a primary objective for the continuing development of this model. 4.4. Comparison with Scanfish data Whilst it is informative to assess model performance by a statistical analysis of predicted and observed near-bed and surface temperatures, a further and perhaps more insightful test is a

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

30

qualitative comparison of the predicted thermal structure with high spatial frequency data sets collected using a towed undulating CTD (Scanfish). Scanfish is a computer-controlled vehicle towed at 3.0–4.5 ms1, which was set to profile between 4 m below the sea surface and about 5 m above the sea bed. Profiles were at a separation of approximately 150–500 m in the horizontal, with a resolution of 1 m in the vertical following averaging of the 17 Hz raw data. A comprehensive description of the use of Scanfish and the subsequent data analysis can be found in Brown et al. (1996) and Fernand (1999). Between 26 August and 5 September 1998 19 Scanfish sections were completed, providing a comprehensive description of the physical oceanography of the Celtic Sea region. Here, we concentrate on three sections that capture the principal features of the water column structure. The locations of the

sections are shown in Fig. 9, numbered 178, 182 and 202, and the observed thermal structure is shown in the upper panels of Figs. 13a–c. A complete description of the water column structure and circulation inferred from these sections may be found in Brown et al. (2003); a summary of the main features is given here. Leg 178, a transect across the southwest of the region between the Irish coast and the Scilly Isles (from left to right in Fig. 13a) shows an intense seasonal thermocline with a temperature gradient of approximately 0.9
C m1. Below this is an extensive central pool of cold water (o11
C), bounded to the north and south by strong surface and bottom fronts. The water column structure along this section is dominated by temperature (Brown et al., 2003), hence this model is able to reproduce the gross features observed (Fig. 13a, bottom panel). The cold pool of bottom water is

(b)

(a)

(c) Fig. 13. Comparison of observed and modelled temperatures for Scanfish sections: (a) 178, (b) 182 and (c) 202 (Fig. 4).

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

successfully predicted although it is about 0.5
C warmer than observed, probably due to errors in the initial temperature field as discussed earlier. The downward slope of the thermocline from the north to the south of the section is also successfully reproduced, following the slope of the bottom topography. Predicted surface temperatures show a peak of about 17
C to the south of the section as observed, although they are lower than observed to the north of the section where the overly diffuse model thermocline impacts the surface temperatures. Here the value of comparing model predictions with high frequency data sets is clear; whilst near-bed and surface temperatures generally compare well with those observed, it is evident that the model is unable to reproduce the very strong thermal gradient. It is possible that this is caused by too much vertical mixing, through diapycnal diffusion by the horizontal mixing scheme, or the vertical turbulence formulation. However, it is likely that the vertical resolution of the model is insufficient. The 15 sigma layers described in Section 2 give a resolution at mid-depth of only 10 m in a water depth of 100 m, and this is insufficient to accurately resolve the strong thermocline. Unfortunately, with the computing power available it was not possible to increase the number of vertical levels, and therefore the relative roles of vertical resolution and diapycnal diffusion could not be assessed. However, although the intense thermocline is poorly predicted, the gradient of the bottom fronts is similar to that observed. As it is the bottom fronts that govern the location and strength of the strong jetlike flows in the frontal region (Brown et al., 2003), the model is able to reproduce the observed transport pathways in this region. Leg 182 was positioned to the northeast of leg 178 (Fig. 9), traversing from the Irish to the Cornish coast over the Celtic Deep (from left to right in Fig. 13b). This section shows a dome of cold bottom water (o11
C) centred over the deep trough of the Celtic Deep, bounded to the south by a series of bottom fronts extending to the Cornish coast. There is a suggestion of a secondary dome to the north. The thermocline is comparatively intense in the centre and to the south, whilst it is more diffuse to the north. As with leg 178, the

31

water column structure is dominated by temperature (Brown et al., 2003). The model again reproduces the gross features of the observed temperatures, predicting a cold bottom water dome with temperatures o11
C, and a peak in surface temperatures of about 17
C to the south of the section. However, it again fails to adequately resolve the strong thermocline, leading to weaker bottom fronts at the margins of the cold pool than observed. To the south of the section, the bottom fronts are predicted closer to the Cornish coast, thus the frontal flows are further south than observed. This is the source of the region of error to the northwest of the Scilly Isles discussed in Section 4.3. The model does not reproduce the bottom fronts at the north margin of the cold pool, or the secondary dome in the predicted temperature field. The observed strong southwestward geostrophic flows associated with these fronts (Brown et al., 2003) are thus poorly predicted. This is further illustrated in the comparison with drifter data described in the next section. Leg 202 (Fig. 9) illustrates the structure along the deep axis of St. George’s Channel. At the north of the section (to the right in Fig. 13c) is the vertically well-mixed water of the Southern Irish Sea. Heading south along the section are series of bottom fronts which mark the transition to the cold isolated bottom water of the Celtic Deep. Toward the northern end, the boundary between stratified and well-mixed water is marked by the Celtic Sea surface front, a feature observed in satellite imagery. The model successfully reproduces the observed transition from well mixed water in the north with a temperature of about 15.5
C, to the thermally stratified region in the south with temperatures in the cold bottom pool of o11
C. However, the model predicts the water column to become well mixed further south than observed, with the strongest gradient in bottom fronts (and the associated westward geostrophic flow (Brown et al., 2003) also occurring further south. Consequently, the predicted transport path does not extend as far north into the Southern Irish Sea as observed. This is further illustrated by the Lagrangian transport discussed in the next section. Possible causes for the

ARTICLE IN PRESS 32

E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

predicted fronts in St. George’s Channel occurring too far south include excessive tidal mixing and inadequate representation of buoyant inputs; these are discussed in the previous section. The strong thermocline is again poorly predicted, as discussed earlier, resulting in a surface front located further south and more diffuse than observed. However, the surface front is dynamically relatively unimportant and will not impact the ability of the model to predict the main transport pathways. 4.5. Comparison with drifters Of the 23 drifters deployed in 1998, the trajectories of 14 are shown here that between them illustrate the principal elements of the circulation pathways. The release sites for these buoys are shown in Fig. 9 and their observed transport paths (filtered to remove the high frequency component) are shown in Fig. 14a. The drifters describe an essentially anti-clockwise (cyclonic) circulation pattern, particularly evident over the Celtic Deep, following the contours of bottom density (Brown et al., 2003). To assess the ability of the model to predict the Lagrangian transport paths, the predicted flow fields were used to drive a particle tracking model. Particle release times and locations, and the durations of particle transport were chosen to correspond to the 14 drifters, with particles assumed to remain at a depth of 30 m. The modelled transport paths are shown in Fig. 14b. Although the details of the observed drifter tracks are not predicted due to sub-grid scale variability, the broad features of the transport patterns are reproduced. However, the two drifters released to the northeast of St. George’s Channel (k and l) are predicted to travel north, in contrast to the observed southwestward transport, the latter following the contours of bottom density. This illustrates that the predicted region of stratification does not extend far enough north, as discussed in Sections 4.3 and 4.4. In addition, the model does not adequately represent the observed cyclonic transport pattern over the Celtic Deep. This is mainly due to the failure of the model to reproduce the observed strong southwestward flows on the northern margins of the basin, as discussed in the previous section. There is

a suggestion that the predicted drifter velocities in the frontal regions are also slightly slower than observed (for example, buoy c). As discussed in Section 4.4, the model does not adequately resolve the strong thermocline above the central cold pool. The bottom fronts at the margins of the cold pool, and their associated geostrophic flows, are therefore weaker than observed and the predicted drifter velocities are correspondingly slower. Given the historic neglect of baroclinic flows in shelf sea models, it is interesting to consider the effect of this assumption on these predictions of Lagrangian transport. The hydrodynamic model simulation was therefore repeated using the same tidal and wind forcing, but with a constant density field (salinity and temperature fixed at 34.0 and 7.5
C, respectively), and the resultant velocity fields were used for the prediction of the drifter tracks as described earlier. The results are shown in Fig. 14c and clearly bear little resemblance to the observed transport paths, in particular drifters c, g, h and m. These simulations aptly demonstrate the importance of including baroclinic flows in the modelling of seasonal transport pathways in the shelf seas.

5. Discussion and conclusions The application of a three-dimensional densityresolving model based on the Princeton Ocean Model (POM) to the prediction of tide, wind and density-driven flows in a region of the northwest European shelf extending from the Celtic Sea to the Sea of the Hebrides has been described. The location of fronts between stratified and wellmixed regions is dependent on the local relationship between depth and tidal mixing (Simpson and Hunter, 1974). It was, therefore, important to ensure that the tidally generated mixing was of the correct magnitude. A comparison of predicted and observed elevations and currents for the two dominant tidal constituents in the region (M2 and S2) demonstrated that the model reproduces the observed tides and is of comparable or greater accuracy than previous large area models of the region.

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36 53.0

33

80

80

4600

4600

53.0

52.5

52.5

lk

g

l

51.5

m h 100

51.0

c

80

d

100

f

40

g

10

60

m

0 100

d

51.0

d

c

b a

d

40

100

b

e

e

b

100

100

a

hc

80

f

20

60

i

2040 600h 8

80

g

10 0

50.5

40

20

j i

h

fi

e

k

40 6 0

40

100

80 c

2040 600 8

20

Latitude

i

n

g

j

l

n 51.5

52.0

Latitude

n

n j

20

100

52.0

k

jk 40 60f

20

20

l

50.5

m

m 0

0

10

60

a

10

20

20

60

50.0

50.0

60

60

80

80

b -7

-6

-5

0

-8

10

0

(a)

49.5 -9

10

49.5 -9

-8

-7

(b)

Longitude

-6

-5

Longitude

80

4600

53.0

52.5 l l 20

k

e

52.0

k

20

40 60

n

n j h

e

j

f

40

g

i

10

0

d

100

80

c

100

a

b

40

b

100

50.5

g

c

f

d

51.0

60 h

20

Latitude

51.5

100

i 80

2040 600 8

m 0

10

m

20

60

a

50.0

60

80

(c)

0 10

49.5 -9

-8

-7

-6

-5

Longitude

Fig. 14. Drifter trajectories for buoys released in the Celtic Sea in 1998. (a) Observed, (b) predicted and (c) predicted (assuming constant density). Lighter and dotted lines indicate model bathymetry. Start and end positions are indicated by  and , respectively.

ARTICLE IN PRESS 34

E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

An extensive hydrographic data set collected in the Irish Sea in 1995 was used to assess the ability of the model to reproduce the spatial and temporal (seasonal) variability in temperature. Model predictions of the development of thermal stratification were in good agreement with a series of observations compiled by Horsburgh (1999). Using all the available data, a total of 4422 data points, a statistical evaluation of model accuracy revealed mean and RMS errors in near-surface temperatures of 0.25
C and 0.72
C, and of 0.05
C and 0.44
C in near-bed temperatures. The largest errors occurred near the start of the simulation when model temperatures were consistently higher than those observed. This implies that the initial temperature field was too warm. Initial temperatures were derived from satellite observations of sea surface temperature and errors may have been introduced due to cloud cover. In addition, applying observed surface temperatures at all model depths implicitly assumes the water column to be well mixed in early spring. However, surface to bed temperature differences were under predicted near the Irish coast at the start of the model simulation and stratification along the Irish coast in spring due to riverine influxes of fresher water has been observed (Dickey-Collas et al., 1997; Horsburgh et al., 2000). These suggest that the water column cannot be assumed to be well mixed throughout the model domain in early spring, specifically in regions of significant freshwater influence, and this may have contributed to errors in the initial temperatures. Simulating the seasonal circulation of the Celtic Sea in 1998, the validated model predicted lower temperatures than observed, with mean and RMS errors in near-surface temperatures of 0.18
C and 0.72
C, and of 0.01
C and 0.74
C in nearbed temperatures (7256 data points). The spatial distribution of model errors suggested that the predicted location of fronts was inaccurate in St. George’s Channel and near the Scilly Isles, in part due to the neglect of salinity variability in the model. Whilst temperature dominates the water column structure over most of the region, salinity is relatively important in the vicinity of the Bristol Channel and at the margins of stratified water in St. George’s Channel (Brown et al., 2003). In

St. George’s Channel, the predicted frontal region did not extend far enough north into the southern Irish Sea. This was clearly illustrated by a comparison of the predicted thermal structure with high spatial resolution data from a towed undulating CTD. Such data are the template against which to assess the ability of numerical models to predict the detailed structure of the water column. In this paper, it has been shown that the model ably predicted the gross features of the observed temperature field as revealed by relatively coarse resolution traditional CTD data. However, it was unable to adequately resolve the strong vertical temperature gradients detailed by the high resolution data sets due to insufficient vertical resolution, a factor which is unfortunately limited by available computer power. The thermocline is the transition region between the nutrient-poor, well-lit surface layer and the darker, nutrient-rich deeper water, and as such plays an important role in determining the biological properties of the water column. Observations in the western English Channel (Sharples et al., 2001) show narrow subsurface chlorophyll maxima situated in the lower half of the thermocline with a vertical thickness of about 5 m. The growth of these phytoplankton layers requires sufficient light and nutrients, with a thermocline situated above the compensation depth, and a flux of nutrients into the base of the thermocline through periodic tidally generated turbulent mixing. These observations suggest that modelling of ecosystem processes in stratified shelf seas needs to be underpinned by a hydrodynamic model of sufficient vertical resolution and complexity to adequately resolve both the depth and gradient of the thermocline and the turbulent mixing processes. It is therefore important that hydrodynamic models of the European shelf be tested against high resolution data sets, as described in this paper, to ensure that the physics used as the basis for ecosystem models is adequately represented. The strength and location of geostrophic flows, which dominate the transport pathways during the summer months in the European shelf seas, are dependent on the position and gradient of bottom density fronts. Thus inaccuracies in the predicted

ARTICLE IN PRESS E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

frontal regions impact on the Lagrangian transport routes. Observations of the Lagrangian transport in the Celtic Sea region in 1998, obtained using satellite tracked drifters, revealed an essentially cyclonic circulation pattern, following the contours of bottom density. Using the predicted flow fields to drive a particle tracking model, this broad transport pattern was successfully predicted, although transport velocities were reduced due to the inability of the model to fully resolve the fine core of the frontal jets. In St. George’s Channel and at the northern margins of the Celtic Deep where the bottom fronts were poorly predicted, the model did not reproduce the observed transport. Historically, shelf models used for the prediction of contaminant transport and as tools for the management of fisheries have used a simplistic approach, generally treating the water as vertically well mixed, thus allowing the use of two-dimensional models, and with no consideration of baroclinic flows. Whilst this approach is valid during winter when the shelf seas are well mixed and transport is predominantly wind driven, during the summer months regions of the northwest European shelf stratify and jet-like frontal flows develop. Using this model to assess the relative contribution of tides, winds and thermally generated density stratification to flows along the frontal region of St. George’s Channel, density was predicted to contribute 91% to the net westward flux. This clearly demonstrates the importance of including the baroclinic component in models of the northwest European shelf to develop an accurate understanding of contaminant transport pathways and for the informed management of fisheries.

Acknowledgements AVHRR data were obtained from the NASA Physical Oceanography Distributed Active Archive Center at the Jet Propulsion Laboratory, California Institute of Technology. The bathymetry used for this work was funded by the Department of Transport and the Regions and the Ministry of Agriculture Fisheries and Food

35

(Brown et al., 1999b). Atmospheric forcing was supplied by the European Centre for Mediumrange Weather Forecasting (ECMWF) through the British Atmospheric Data Centre. Tide gauge and current meter data were obtained from the British Oceanographic Data Centre. This research was supported by the Department for Environment, Food and Rural Affairs (DEFRA) through contracts AE1021, AE1214 and AE1225. The authors are grateful to two referees for their helpful comments.

References Aldridge, J.N., Davies, A.M., 1993. A high-resolution threedimensional hydrodynamic tidal model of the Eastern Irish Sea. Journal of Physical Oceanography 23 (2), 207–224. Asselin, R., 1972. Frequency filters for time integration. Monthly Weather Review 100, 487–490. Bauer, J., Fischer, J., Leach, H., Woods, J.D., 1985. SEAROVER Data Report 1-North Atlantic Summer 1981 NOA’81. Berichte aus dem Institut fur . Meereskunde an der ChristianAlbrechts-Universitat Kiel, 143, 155pp. Blumberg, A.F., Mellor, G.L., 1987. A description of a threedimensional coastal ocean circulation model. In: Heaps, N.S. (Ed.), Three Dimensional Coastal Ocean Models, Coastal and Estuarine Sciences, Vol. 4. American Geophysical Union, Washington, 208pp. Brown, J., Hill, A.E., Fernand, L., Bennett, D.B., Nichols, J.H., 1995. A physical retention mechanism for Nephrops norvegicus larvae. ICES, C. M., 1995/K:31 Ref.C (Mimeo). Brown, J., Brander, K.M., Fernand, L., Hill, A.E., 1996. Scanfish: a high performance towed undulator. Sea Technology 37, 23–27. Brown, J., Hill, A.E., Fernand, L., Horsburgh, K.J., 1999a. Observations of a seasonal jet-like circulation at the Central North Sea cold pool margin. Estuarine Coastal and Shelf Science 48, 343–355. Brown, J., Joyce, A.E., Aldridge, J.N., Young, E.F., Fernand, L., Gurbutt, P.A., 1999b. Further identification and acquisition of bathymetric data for Irish Sea modelling. DETR research contract CW0753. Brown, J., Carillo, L., Fernand, L., Horsburgh, K.J., Hill, A.E., Young, E.F., 2003. Observations of the physical structure and seasonal jet-like circulation of the Celtic sea and St. George’s channel of the Irish sea. Continental Shelf Research 23 (6), 533–561. Davies, A.M., Jones, J.E., 1992. A three dimensional model of the M2, S2, N2, K1 and O1 tides in the Celtic and Irish Seas. Progress in Oceanography 29, 197–234. Dickey-Collas, M., Brown, J., Fernand, L., Hill, E.D., Horsburgh, K.J., Garvine, R.W., 1997. Does the Western

ARTICLE IN PRESS 36

E.F. Young et al. / Continental Shelf Research 24 (2004) 13–36

Irish Sea gyre influence the distribution of pelagic juvenile fish? Journal of Fish Biology 51a, 206–229. Evans, S.R., Middleton, J.F., 1998. A regional model of shelf circulation near Bass Strait: a new upwelling mechanism. Journal of Physical Oceanography 28, 1439–1457. Fernand, L., 1999. High resolution measurements of the velocity and thermohaline structure of the Western Irish Sea gyre. Ph.D. Thesis, University of Wales, Bangor, 93pp. Flather, R.A., Proctor, R., Wolf, J., 1991. Oceanographic forecast models. In: Farmer, D.G., Rycroft, M.J. (Eds.), Computer Modelling in the Environmental Sciences. IMA Conference Series 28, Clarendon Press, Oxford, pp. 15–30. Galperin, B., Mellor, G.L., 1990. A time-dependent, threedimensional model of the Delaware Bay and river system. Estuarine Coastal and Shelf Science 31, 231–281. Gill, A.E., 1982. Atmosphere-Ocean Dynamics. Academic Press, New York, 662pp. Hill, A.E., Durazo, R., Smeed, D.A., 1994. Observations of a cyclonic gyre in the Western Irish Sea. Continental Shelf Research 14, 470–490. Hill, A.E., Brown, J., Fernand, L., 1996. The western Irish Sea gyre: a retention mechanism for the Norway lobster (nephrops norvegicus)? Oceanologica Acta 19, 357–368. Hill, A.E., Brown, J., Fernand, L., 1997. The Summer Gyre in the Western Irish Sea: shelf sea paradigms and management implications. Estuarine, Coastal and Shelf Science 44 (A), 83–95. Holt, J.T., James, I.D., 1999. A simulation of the Southern North Sea in comparison with measurements from the North Sea Project. Part 1: temperature. Continental Shelf Research 19, 1087–1112. Holt, J.T., James, I.D., 2001. An s coordinate density evolving model of the northwest European continental shelf 1, model description and density structure. Journal of Geophysical Research 106 (C7), 14015–14034. Horsburgh, K.J., 1999. Observations and modelling of the Western Irish Sea gyre. Ph.D. Thesis, University of Wales, Bangor, 173pp. Horsburgh, K.J., Hill, A.E., 2003. A three-dimensional model of density-driven circulation in the Irish Sea. Journal of Physical Oceanography 33, 343–365. Horsburgh, K.J., Hill, A.E., Brown, J., Fernand, L., Garvine, R.W., Angelico, M.M.P., 2000. Seasonal evolution of the cold pool gyre in the western Irish Sea. Progress in Oceanography 46, 1–58. Jones, J.E., Davies, A.M., 1996. A high-resolution, threedimensional model of the M2, M4, M6, S2, N2, K1 and O1 tides in the Eastern Irish Sea. Estuarine, Coastal and Shelf Science 42, 311–346. Large, W.G., Pond, S., 1981. Open ocean momentum flux measurements in moderate to strong winds. Journal of Physical Oceanography 11, 324–336. Mellor, G.L., Yamada, T., 1982. Development of a turbulence closure model for geophysical fluid problems. Review of Geophysics and Space Physics 20, 851–875.

Mesinger, F., Arakawa, A., 1976. Numerical Methods Used in Atmospheric Models. GARP Publication Series, 17, World Meteorological Organisation, Geneva, 64pp. Oey, L.Y., Mellor, G.L., Hires, R.I., 1985. A three-dimensional simulation of the Hudson–Raritran estuary. Journal of Physical Oceanography 15, 1676–1709. Pingree, R.D., 1980. Physical oceaonography of the Celtic Sea and English Channel. In: Banner, F.T, Collins, W.B., Massie, K.S. (Eds.), The North-West European Shelf Seas: the Sea Bed and the Sea in Motion. II. Physical and Chemical Oceanography and Physical Resources. Elsevier, Amsterdam, Oxford, New York, pp. 415–465. Pingree, R.D., Griffiths, D.K., 1987. Tidal friction for semidiurnal tides. Continental Shelf Research 7 (10), 1181–1209. Reed, R.K., 1976. On estimations of net long-wave radiation from the ocean. Journal of Geophysical Research 81, 5793–5794. Robinson, I.S., 1979. The tidal dynamics of the Irish and Celtic Seas. Geophysical Journal Royal Astronomical Society 56, 159–197. Sharples, J., Moore, C.M., Rippeth, T.P., Holligan, P.M., Hydes, D.J., Fisher, N.R., Simpson, J.H., 2001. Phytoplankton distribution and survival in the thermocline. Limnology and Oceanography 46 (3), 486–496. Simpson, J.H., Hunter, J.R., 1974. Fronts in the Irish Sea. Nature 250, 404–406. Sinha, B., Pingree, R.D., 1997. The principal lunar semidiurnal tide and its harmonics: baseline solutions for M2 and M4 constituents on the North-West European continental shelf. Continental Shelf Research 17 (11), 1321–1365. Smagorinsky, J., 1963. General circulation experiments with the primitive equations: 1. the basic experiment. Monthly Weather Review 91, 99–164. Smolarkiewicz, P.K., 1983. A fully multidimensional positive definite advection transport algorithm with small implicit diffusion. Journal of Computational Physics 54, 325–362. Xing, J., Davies, A.M., 2001. A three-dimensional baroclinic model of the Irish Sea: formation of the thermal fronts and associated circulation. Journal of Physical Oceanography 31, 94–114. Yang, H., Weisberg, R.H., 1999. Response of the West Florida shelf circulation to climatological wind stress forcing. Journal of Geophysical Research 104 (C3), 5301–5320. Young, E.F., 2002. Tidal validation of a three-dimensional, primitive equation model for the Irish and Celtic Seas region. CEFAS internal report, 46pp. Young, E.F., Aldridge, J.N., Brown, J., 2000. Development and validation of a three-dimensional curvilinear model for the study of fluxes through the North Channel of the Irish Sea. Continental Shelf Research 20, 997–1035. Zavatarelli, M., Mellor, G.L., 1995. A numerical study of the Mediterranean Sea circulation. Journal of Physical Oceanography 25, 1384–1414.