Estuarine, Coastal and Shelf Science (2001) 53, 477491

doi:10.1006/ecss.1999.0628, available online at http://www.idealibrary.com on

Integrated Modelling of the Belgian Coastal Zone
E. J. M. Delhez
a
and G. Carabin
b
a
Institut de Mathematique, University of Lie`ge, Sart Tilman B37, B-4000 Lie`ge, Belgium
b
Laboratories of Engineering Geology, Hydrogeology and Geophysical ProspectingLGIH, University of Lie`ge,
Sart Tilman B19, B-4000 Lie`ge, Belgium
Received 11 January 1999 and accepted in revised form 5 May 1999
The management of the water resources in coastal or delta plains asks for an integrated modelling of the water system at
a regional scale. In the SALMON project, detailed descriptions of the groundwater, river and marine domains are
provided by coupling appropriate numerical models of these dierent sub-systems.
The application of this three-fold model to the Scheldt and Belgian Coastal Zone reveals a marked river plume
extending along the Belgian Coast with strong oshore gradients. This plume cannot be explained by the Eulerian
residual transport velocity eld but requires Lagrangian aspects to be taken into account. The computation of the age of
this water mass shows that the dierent parts of the coastal zone are inuenced by a discharge in the Grote Schijn River
after a delay that varies between a few days (for the marine part of the estuary) to about 100 days (for the centre of the
residual gyre). The computation shows also that the groundwater around Antwerp can also be contaminated by a
pollution in the river but the resulting mass loss from the river does not lead to a signicant reduction of the ux to the
coastal zone. 2001 Academic Press
Keywords: integrated modelling; tidal dynamics; groundwater; Scheldt
Introduction
Because of their geographical location, coastal or delta
plains situated at the major river mouths are highly
populated, leading to a very important freshwater
demand. But the anthropogenic activities (urban,
agricultural and industrial) in these regions lead also to
an important degradation of the water quality and of
the whole aquatic ecosystem, threatening the quality
of the freshwater resources as well as the food re-
sources (sheries) and recreational sea and river areas.
There is a strong need for tools to evaluate and
implement cost-eective strategies to control and
decrease pollution of these valuable water resources.
One of the key steps of this research as identied by
the International Geosphere-Biosphere Programme
(IGBP) is the determination, at global and regional
scales, of the uxes of material between land, sea and
atmosphere through the coastal zone, and of the
capacity of coastal systems to transform and store
particulate and dissolved matter (IGBP, 1993).
Identifying the need for an integrated management
of the coastal resources, the SALMON (Sea Air
Land Modelling Operational Network) project was
designed to take benet from the recent advances in
computer technology to analyse the contamination of
groundwater, rivers and oceans as interconnected
problems.
For a long time, the scientic basis for the separate
understanding and modelling of the groundwater,
river and marine systems has indeed been accumu-
lated. Dierent kinds of models with dierent objec-
tives, dierent state variables, dierent time scales and
length scales exist therefore in the dierent scientic
communities studying these systems. The basic aim of
the SALMON project was to build bridges between
these dierent disciplines to develop a single model
describing the environment quality in a whole system
of regional scale including marine, river, groundwater
and atmospheric inputs.
In this paper, we will give a broad overview of the
whole SALMON methodology before focusing on
the results of its application to the Scheldt Basin and
the Belgian Coastal Zone.
The SALMON meta-model
The SALMON project relies on the experience of
three laboratories of the University of Lie`ge (Belgium)
in the mathematical and numerical modelling of the
ocean (GHERGeoHydrodynamics and Environ-
ment Research laboratory), the river (CEMECentre
dEtude et de Modelisation de lEnvironnement) and
the groundwater (LGIHLaboratoire de Geologie de
lIngenieur et dHydrogeologie), to develop a single
02727714/01/100477+15 $35.00/0 2001 Academic Press
model to handle the description of environment qual-
ity in a whole system of regional scale including
marine, river, groundwater and atmospheric inputs. In
this complete three-fold model, uxes of water, nutri-
ents, pollutants, etc., are computed in each sub-model
and transferred continuously from one sub-model to
another at the common boundaries (e.g. the river
discharge in the open ocean).
In practice, this is made possible by taking advan-
tage of the clustered RS/6000 machines in the form of
an IBM SP/2 parallel computer and by using the PVM
software for the exchange of data between the dier-
ent tasks running on the processors. The technical
issues addressed by the SALMON project are: (i) the
design and implementation of the Junction concept,
a tool developed to connect groundwater, river and
ocean models as well as to connect ocean models of
dierent scales in a client-server architecture; (ii) the
denition of the connections between the ground-
water, river and ocean models, identifying the data to
be exchanged and their reprocessing (interpolation in
space and time, aggregation in state space) by the
junction to make them understandable by the receiv-
ing model; (iii) the development and implementation
of a Generalised Domain Decomposition Method
(GDDM) for the parallelization of the numerical
codes of the ocean and groundwater models, taking
advantage of the performance of the SP2 parallel
computer; (iv) the development and implementation
of a graphical server to control the whole application
and provide online visualizations of the results.
More details about these aspects can be found in
Ghiot et al. (1998) and at http://modb.oce.ulg.ac.be/
SALMON.
The methodology of SALMON is meant to be
easily portable from one site to another. This means
that the tools developed to connect models with
dierent scopes, dierent space/time resolutions, dif-
ferent state variables can be used to handle the trans-
fer of information between a wide variety of models
with a minimum of arrangements. The model imple-
mentations are however still specic to a particular
domain. The site chosen for the rst application of
SALMON is the Scheldt Basin and the Belgian
Coastal Zone (Figure 1).
SALMON groundwater model
The domain of application of the groundwater model
covers an area of about 650 km
2
around Antwerp
France
Belgium
Liege
'
Antwerp
The Netherlands
30 km
Oostende
North
Sea
3E 2E 4E 5E 6E
50N
51N
F 1. The Scheldt River network, the groundwater domain (dark grey) and the marine coastal domain (light grey). The
star close to Antwerp indicates the location of the release point of the tracers used in the case study.
478 E. J. M. Delhez and G. Carabin
(Figure 1). The basic geology in this region consists in
a pile of Tertiary and Secondary sedimentary layers
with tabular structures sloping northwards or north-
eastwards. This structure induces a superposition of
conned (or semi-conned) aquifers inter-layered
with semi-pervious layers. The catchment areas of
these aquifers are fully connected with the Quaternary
layers, especially in the alluvial plains. The domain
is bounded by natural boundaries: rivers and
groundwater divides.
The groundwater model, SUFT3D (e.g. Carabin
et al., 1998), simulates the groundwater ow, which
can be density dependent (case of seawater intrusion),
and the transport of dissolved contaminants through
saturated and unsaturated porous media. The state
variables are the pressure head and the concentration
of solutes in the mobile water. The processes taken
into account are the advection, the diusion, the
dispersion, the linear degradation, the adsorption
(with linear, Langmuir or Freundlich isotherm) and
the immobile water eect (which simulates the eect
on transport of a dual porosity). The model addresses
real 3-D cases in both steady and unsteady conditions,
for both conned and unconned (water table)
aquifers.
The numerical code is based on nite elements with
an horizontal size varying between 50 and 250 m and
a thickness from 1 to 100 m (95 415 nodes).
SALMON river model
The river model, PEGASE, is a tool for assessing the
policy of surface water management, by taking into
account the relationships between water discharge,
pollutants load and water quality. The purpose of
PEGASE is to provide a global view of water quality
for a river network, under characteristic meteorologi-
cal conditions: low-ow period, mean ow, well-
established ood regime (e.g. Smitz et al., 1997).
The modelled Scheldt River network is represented
by a series of one-dimensional interconnected
segments representing 120 rivers for a total drainage
area of 17 000 km
2
(Figure 1).
A hydrodynamic sub-model calculates the dis-
charges and the other hydrodynamic parameters
(water velocity, depth, etc.) in dierent reaches from
the morphometric characteristics of the river (slope,
width, presence of dams, etc.) as well as from
measured discharges.
A temperature sub-model calculates the water tem-
perature in the dierent reaches on the basis of the
height above sea level of the river segment, of the
various inputs of warm water and of measured values.
A biological sub-model calculates the development
of plankton biomass (phyto-, zoo- and bacterio-
plankton), the degradation of organic matter, the
production of autochthonous organic matter (primary
production of phytoplankton) and the heterotrophic
activity. The carbon, phosphorus and nitrogen vari-
ables are split according to their physical properties
(prone to sedimentation or not), their chemical prop-
erties (ammonia, phosphates) and their biological
properties (quickly, slowly or non biodegradable).
SALMON marine model
In the scope of the SALMON project, a ne resol-
ution hydrodynamic model of the Belgian coastal zone
has been developed to describe the extension of the
Scheldt plume. The general characteristics of this
model are common to the general form of the GHER
model (e.g. Nihoul et al., 1989; Delhez & Martin,
1992): the model is three-dimensional (using the
terrain following -coordinate system), non-linear and
baroclinic with prognostic salinity and temperature
elds. The robust turbulence closure scheme is based
on the evolution equation for the turbulent kinetic
energy and a parametric, algebraic expression of the
mixing length taking into account both the surface
wind induced activity and the stratication (Nihoul
et al., 1989; Delhez et al., 1999). The model also
includes a drying and ooding algorithm to cope with
the shallow sand banks (a grid point is considered to
be dry when the local water height decreases below a
threshold value of 3 cm).
The discretisation is based on the FBTCS method,
i.e. Forward Backward on Time and Centered Space
(Beckers & Deleersnijder, 1992). The only departures
from this rules arise from the partly implicit evaluation
of the vertical advection and turbulent diusion
terms and the use of a TVD advection scheme with
Superbee limiter (Sweby, 1984).
The model has an horizontal resolution of
500500 m and uses 10 unequally spaced vertical
-levels with enhanced resolution near the surface and
the bottom. Preliminary tests showed that the spread-
ing of the plume was not signicantly aected when
using 15 equally spaced vertical levels. The model
includes more than 120 000 active grid cells. The
mesh is roughly aligned with the Belgian coast for a
better representation of the coastline. This grid allows
also a good description of the river mouth. The
bathymetry, obtained by digitalization of the navi-
gation maps, is shown in Figure 2. The mean depth in
this area is 15 m and the maximum depth is 35 m.
This local hydrodynamic model is forced with 6
hourly ECMWF wind stress, air pressure and heat
Integrated modelling of the Belgian Coastal Zone 479
uxes. More precisely, the heat ux is computed from
the solar radiation, cloud cover, air temperature and
humidity and simulated sea surface temperature. The
precipitation ux is assumed to balance the evapor-
ation. The forcing data at the open boundariesi.e.
water level, temperature and salinityare obtained
from a large scale three-dimensional hydrodynamic
model covering the whole north-western European
continental shelf with a grid size of 10 10i.e.
about 12185 kmand 10 vertical -levels. This
large scale model has the same characteristics as the
local model, is forced with the same meteorological
data and nine principal tidal components at its own
open boundaries. The output of the large scale model
are interpolated in time and space and prescribed
to the local model using a radiation open boundary
condition based on the characteristic method
(Hedstrom, 1979; Red & Cooper, 1987). Note that
the inuence of the local marine model is not taken
into account in the current implementation of the
large scale model.
Interfacing of the models
The computer set-up used to couple the models to
each-other is described in a previous publication
(Ghiot et al., 1998). It is based on the client-server
architecture: the models send and receive to/from a
central application all the information required to
compute the boundary conditions.
The ux of water and dissolved constituents
between the river and the groundwater can take any
direction, i.e. from the aquifer to the river when the
river drains the aquifer or from the aquifer to the river
in the opposite situation. During the initialization of
the model run, an automatic mapping algorithm is
applied to determine the correspondence between the
grid points of the river model and these of the ground-
water model representing the same physical river
segment. Then, the exchange is parameterized as a
function of the dierence between the water level in
the river and the groundwater piezometric head at the
corresponding grid points (Carabin et al., 1998).
The coupling of the river and marine models is
more dicult to implement because the two models
address dierent time scales. As can be seen in Fig-
ures 1 and 2, the local marine model does not cover
the whole Scheldt Estuary but only that part where
environmental conditions can be described as marine
conditions. The tidal signal does however propagate
much further into the river system, i.e. up to
Ghentabout 100 km upstream. As tide is not taken
into account in the PEGASE river model, a 1-D
model of the propagation of the tide and storm surge
into the Scheldt River, a junction river model , was
developed to avoid spurious reections of the tidal
wave at the boundary of the marine model and to get
a better description of the dissipation of the energy.
This model is also used to couple the river and marine
models and compute the corresponding water uxes.
0
40
60
0
10
20
30
40
50
20 60 80 100
0 5 10 15 20 25 30
Depth (m)
Nieuwpoort
Oostende
Zeebrugge
N
Scheldt
River
Westkapelle
S
Distance in km
D
i
s
t
a
n
c
e

i
n

k
m
5100'
230' 300'
F 2. Bathymetry of the hydrodynamic model of the Belgian Coastal Zone. (Axis labels indicate distance in km.) Depth
is in metres and isolines are plotted every 5 metres.
480 E. J. M. Delhez and G. Carabin
This junction river model is based on the continuity
and momentum equations integrated over the cross-
section of the river. It covers a net of six one-
dimensional river branches covering that part of the
Scheldt River subjected to tidal inuence with a
spatial resolution of 2 km (e.g. Regnier et al., 1997).
On the marine side, the junction river model is forced
by the free surface elevation signal produced by the
marine model. At the other sides, the (residual) ow
rates computed by the PEGASE river model are
prescribed. The junction river model provides in turn
the tide+surge+residual ow to be prescribed to the
marine hydrodynamic model. The derivative of the
deviation of the horizontal velocity with respect to its
depth-mean value in the direction normal to the
boundary between the river junction model and the
marine model is assumed to be zero.
To exchange heat, salt and matter between the river
and the marine models, the uxes computed by
PEGASE at its downstream boundary must also be
modulated by the tidal and surge signal before being
prescribed to the local marine hydrodynamic model.
The solution assumes a one way coupling only, i.e.
from the river to the marine domain, and is based on
the hydrodynamic description provided by the junc-
tion river model. For this purpose, a 1-D advection/
dispersion equation is added to the junction river
model. The diusion coecients used in this one-
dimensional model are calibrated against the salinity
distribution in the river and the estuary. The uxes
computed by PEGASE at its downstream boundary
are fed into the junction river model at the corre-
sponding grid point via an articial buer zone that
makes sure that the heat, salt and dissolved matter will
be entrained downstream only in order to remain
compatible with the meaning of the uxes computed
by PEGASE on a daily average basis. Then, a two-way
coupling of the junction river model with the marine
model ensures the exchange of heat, salt and dissolved
constituents.
Hydrodynamics of the Belgian Coastal Zone
As stated previously, the dynamics of the Belgian
Coastal Zone is determined to a large extent by the
strong semi-diurnal tidal signal. At Oostende, the
amplitude of the M
2
and S
2
tides reach 18 and 07 m,
respectively. As a result of the large tidal currents, the
water column remains relatively well-mixed all
through the year. In particular, no permanent haline
stratication can be associated with the plume of the
Scheldt because of the low freshwater discharge
(around 100 m
3
s
1
on average) and the strong
mixing.
The salinity distribution evolves with the tidal
excursion (10 km, i.e. 20 grid points) and the
uctuations of the wind eld at time scales of 5 to 7
days. Northerly winds promote a larger extension of
the plume along the Belgian coast, while southerly
winds have a blocking eect and push the plume
further oshore (e.g. Yang, 1998). During the simu-
lated period of time, i.e. the year 1993, the variations
of the freshwater discharge and of the residual ow
through the Strait of Dover did not have any discern-
ible inuence on the variations of the strength and
extension of the plume. The snapshot of the simulated
salinity distribution shown in Figure 3 is represen-
tative of the recurrent pattern of this plume as it can
be derived also from experimental observations (Fig-
ure 4): the fresh water is swept along in a clockwise
residual gyre o the Belgian coast. When leaving the
Scheldt Estuary, the plume turns left and ows
along the Belgian coast towards Zeebrugge and
Oostende. Further downstream, the plume moves
oshore and is progressively swept along by the out-
ow of the Dover Strait in its way to the North Sea
(e.g. Delhez, 1996a).
It should be stressed that while Figure 4 is illustra-
tive of the overall pattern of the Scheldt Plume, the
spatial gradients associated with the plume are largely
underestimated by such a long term mean picture as
the real plume undergoes large mesoscale excursions.
The situation cannot be described or modelled appro-
priately without taking into account explicitly the
associated oscillations.
To understand the mechanism by which the plume
turns left when leaving the estuary, the horizontal
Eulerian residual transport velocity eld, u
E,0
, was
computed, i.e. the monthly average value of the hori-
zontal transport eld U divided by the corresponding
averaged depth H
0
=KHL
E
:
where the subscript 0 indicates a residual variable and
where the brackets K. . .L
E
are used to denote the
Eulerian average operator.
This velocity eld (Figure 5) shows the inuence of
several forcing functions in this area. The large scale
sea surface slope set up across the whole shelf induces
the general northwards circulation oshore. The local
pressure gradient induced by the salinity dierence
between the freshwater inow and the saltier water
oshore triggers a typical estuarine circulation pattern
with the freshwater owing out of the estuary at the
surface and the intrusion of higher salinity water in
the estuary in the bottom layer. This two-layer
Integrated modelling of the Belgian Coastal Zone 481
circulation is however restricted to the inner estuary
only. The monthly averaged eect of the local
meteorological forcing on the local residual circulation
is small.
Surprisingly, the Eulerian residual transport eld
does not explain the recurrent turning left of the
Scheldt Plume. In both the surface and bottom layers,
the residual circulation along the Belgian coast
appears to ow northwards, in apparent contradiction
with the computed plume extension. To have a better
explanation, the non-linear dynamics of the higher
frequency motions associated with the mesoscale
motions must be taken into account. The large
horizontal tidal excursionand more generally
mesoscale excursion, where the term mesoscale is
understood as referring to time scales ranging from
several hours to several daysand the large mesoscale
velocity gradients associated with the irregular
bathymetry are responsible for a large departure of the
Eulerian residual circulation from the Lagrangian
residual circulation, i.e. the net long-term displace-
ment of water parcels.
The true Lagrangian residual circulation is dicult
to compute and analyse as it depends on the exact
release time of the water parcels used for its compu-
tation and forms therefore a kind of unsteady eld.
The average eect of the mesoscale oscillations
can however be accounted for by the rst order
0 100
60
34.8
10
20
30
40
50
20 40 60 80
34.5
34.0
33.0
32.0
31.5
33.0
33.0
30.0
28.0
26.0
20.0
F 3. Snapshot of the surface salinity distribution on 1 August 1993.
Nieuwpoort
Oostende
Zeebrugge
34
34.6
34.5
34.4
33.5
33
32.5
32
31.5
31
F 4. Salinity distribution derived from long-term measurements (after Mommaerts, 1991).
482 E. J. M. Delhez and G. Carabin
approximation of the Stokes Drift transport velocity
eld v
s
(Delhez, 1996b)
where v
1
is the oscillating mesoscale (mainly
tide+wind induced) velocity vector. Note that there
are no strong mesoscale instabilities of the plume and
that the density-driven part of v
1
remains small. In a
weakly non-linear case, v
1
can be considered as the
dominant term of the correction to be applied to the
Eulerian residual (Feng et al., 1986). It can be clearly
seen in Figure 6 that this velocity eld is of the same
order of magnitude as the Eulerian residual transport
eld but has a completely dierent direction at many
locations. The sum of the Eulerian residual transport
velocity and the rst order approximation of the
Stokes Drift transport velocity, i.e. the rst order
approximation of the Lagrangian residual transport
velocity, shows therefore a completely dierent pic-
ture (Figure 7). This residual velocity eld describes
only part of the complex non-linear dynamics as it
is based on the not fully veried assumption that
the mesoscale velocity gradients experienced by a
water parcel during the mesoscale excursion can be
described by the local gradient of the velocity at the
point where the rst order Lagrangian residual trans-
port velocity eld is dened. It is however sucient to
explain why the plume turns left when leaving the
Scheldt Estuary. It is indeed clear, especially in the
surface layer, that the residual Lagrangian motion
resulting from the mesoscale oscillations is directed
along the Belgian coast, or, to be more precise,
along the principal navigation channel to Antwerp
Harbour. Because of the characteristics of the
bathymetry near Zeebrugge Harbour the residual vel-
ocity turns then oshore following the 10 m depth
contour. Further oshore, the total residual signal is
then similar to the one associated with the large
scale general circulation so that a loop is formed.
More return paths can exist further downstream,
according to the particular meteorological conditions,
and can give an even more complex pattern with
secondary gyres.
In Figure 7, the residual velocity eld seems to be
noisy. This is not the result of a lack of convergence of
the computation or of numerical instabilities (the
Eulerian residual transport velocity is much smoother)
but, as explained above, this reects the large horizon-
tal velocity shear induced by the complex shallow
bathymetry. In the outer estuary, the main ow occurs
along the deeper navigation channel. On the sides of
this channel, complex residual gyres with horizontal
scales of 1 km or less appear also (note that, for clarity,
vector arrows are drawn every two grid points only in
both horizontal directions). On the left side of the
gure, the inuence of the Flemish Banks system
can be seen as a succession of quasi parallel lines
with opposite residual velocity elds along their two
0.40 m s
1
F 5. Surface Eulerian residual transport velocity eld (July 1993).
Integrated modelling of the Belgian Coastal Zone 483
sides as predicted theoretically (Huthnance, 1973;
Zimmerman, 1980).
Case study
Several simulations have been carried out to validate
the dierent components of SALMON and the
coupling of the dierent sub-models as well as to
investigate possible scenarios of pollution and learn
more about the inuence of the hydrodynamics on the
fate of this pollution.
As an example, a constant release of 50 g s
1
of a
conservative tracer was considered in a minor branch
of the river, the Grote Schijn (Figure 1). The
0.40 m s
1
F 6. First order approximation of the surface Stokes drift transport velocity (July, 1993).
0.40 m s
1
F 7. First order approximation of the surface Lagrangian residual transport velocity (July 1993).
484 E. J. M. Delhez and G. Carabin
average ow rate of the river at this point is less than
2 m
3
s
1
. The studied scenario corresponds to the
case of a continuous pollution by the direct discharge
of pollutants in the river through the sewage system or
through outlets of some factory. The passive constitu-
ent is then transported to the sea through the river
system and can possibly inltrate into the ground-
water according to the hydrologic conditions. The
model ran for six months starting on 29 April 1993.
The distribution of the conservative tracer in the
river (Figure 8) reects only the dilution of the release
by lateral inputs of water and the time variations of
the ow and of the inputs. The concentration shows
therefore a maximum at the release point and
decreases rapidly downstream.
The ux of the tracer transferred to the marine
model varies with time (Figure 9) around a mean
value of 50 g s
1
, i.e. the magnitude of the source.
The large variations of the ux are produced by the
variations of the ow rate of the river. The freshwater
ow and the passive tracer ux are indeed positively
correlated in time while the total mass of the tracer in
the river is negatively correlated with the freshwater
ow. All three time series exhibit strong variations at
a time scale of a few days. Larger ow rates induce a
stronger ushing of the river system with a larger ux
of the tracer and a strong decrease of the total mass of
passive tracer in the river.
The strong response of the total mass of the tracer
present in the Grote Shijn River results obviously from
the large residual velocities (25 cm s
1
) observed
and computed in this highly canalized part of the river.
The time required for the tracer to leave the 30 km
river branch is therefore of one or two days only.
While the total mass in the river and the ux of
passive tracer from the river to the marine domain
exhibit strong oscillations with time, the total mass of
the tracer accumulated in the marine coastal zone has
0 30
4
km
k
g

m

3
1
2
3
20
R
e
l
e
a
s
e

p
o
i
n
t
S
c
h
i
l
d
e
B
o
e
c
h
o
u
t
-



K
o
u
d
D
e
u
r
n
e
-



S
c
h
i
j
n
M
e
r
k
s
e
m
10
Grote Schijn: 18-Aug-1993
Flow rate: 1.27 m
3
s
1
F 8. Snapshot of the tracer concentration in the river.
0
1-Sep-93
160
F
r
e
s
h
w
a
t
e
r

f
l
o
w

(
m
3

s

1
)
1-May-93
20
40
60
80
100
120
140
1-Jul-93 1-Jun-93 1-Aug-93
F
l
u
x

o
f

p
a
s
s
i
v
e

t
r
a
c
e
r

t
o

m
a
r
i
n
e

d
o
m
a
i
n

(
g

s

1
)
4.00E + 06
1.40E + 07
2.00E + 06
0.00E + 00
2.00E + 06
4.00E + 06
6.00E + 06
8.00E + 06
1.00E + 07
M
a
s
s

o
f

t
h
e

p
a
s
s
i
v
e

t
r
a
c
e
r

i
n

t
h
e

r
i
v
e
r

(
g
)
1.20E + 07
F 9. Evolution of the freshwater ow (m
3
s
1
), tracer ux from the river to the marine domain (g s
1
) and total mass
of the passive tracer in the Grote Schijn River (g). Thin line: freshwater ow; thick line: tracer ux; triangles: mass of the
tracer in the river.
Integrated modelling of the Belgian Coastal Zone 485
a much smoother behaviour and grows quasi linearly
in time as a result of the larger buer capacity of this
system (Figure 10).
The conservative constituent can be traced in
the Belgian coastal zone where a plume develops
progressively and shows the characteristics described
previously: displacement to the left of the Scheldt
Estuary outow, spreading along the Belgian
coast, movement oshore, mixing with the residual
ow from the Strait of Dover to the North Sea
(Figure 11).
Strong variations of the spatial distribution of this
plume are evident in Figure 11, even after the full
development of the plume (around mid-July), as a
result of the meteorological and tidal oscillations.
To go further into the analysis of the dynamics of
the coastal zone, a second tracer was simulated under
the same conditions, with the same release location
and ux, but assuming this time a radioactive decay
with a half-life equal to 6 months. The comparison of
the concentrations of the conservative and radio-
active tracers makes it possible to compute an age of
the tracer, i.e. the time elapsed since the tracer was
released in the river.
If a radioactive constituent is released at time t
0
with
an initial concentration C
0
, the concentration at any
time t>t
0
will be given by
C
r
=C
0
exp[(tt
0
)] (3)
(where =log 2/180 days
1
is the disintegration rate)
provided that there is no dilution nor diusion of this
constituent. For a conservative tracer, under the
same assumptions, the concentration will remain
unchanged so that the age can be computed as
In a real system, the passive and radioactive con-
stituents undergo the same advection and diusion so
that their ratio at any specic location is still indicative
of their age. (In reality, non-linear advection schemes,
like TVD schemes, result in dierent numerical
advections of passive and radioactive tracers. This
eect can however be neglected provided that the
characteristic time of radioactive decay is larger
than the computed age.) Application of the formula
(Equation 4) will however give a mean age as the
computed (or observed) concentrations result
from the mixture of dierent tracer parcels with
dierent ages. It should be noted that, because
of the diusion and temporal oscillations of the ow,
this mean age is not the time necessary for a single
particle to reach the location examined from its release
point but should rather be interpreted as the time
needed for a drastic change in the characteristics of
the source to aect signicantly the conditions at this
point.
1-Sep-93
M
a
s
s

o
f

t
h
e

p
a
s
s
i
v
e

t
r
a
c
e
r

i
n

t
h
e

r
i
v
e
r
1-May-93 1-Jul-93 1-Jun-93 1-Aug-93
a
n
d

t
h
e

g
r
o
u
n
d
w
a
t
e
r

d
o
m
a
i
n
s

(
g
)
0.00E + 00
7.00E + 08
1.00E + 08
2.00E + 08
3.00E + 08
4.00E + 08
5.00E + 08
6.00E + 08
M
a
s
s

o
f

t
h
e

p
a
s
s
i
v
e

t
r
a
c
e
r

i
n

t
h
e

m
a
r
i
n
e

d
o
m
a
i
n

(
g
)
0.00E + 00
1.40E + 07
2.00E + 06
4.00E + 06
6.00E + 06
8.00E + 06
1.00E + 07
1.20E + 07
F 10. Total mass (g) of the tracer accumulated in the dierent domains. Thin line: groundwater; thick line: coastal
zone; triangles: river.
486 E. J. M. Delhez and G. Carabin
Figure 12 shows the age distribution at dierent
times. After a transient period corresponding to the
development of the plume (this local adaptation time
reects itself the local age of the tracer), a recurrent
pattern emerges. The discharge of the canalized Grote
Schijn occurs downstream Antwerpen and carries one
or two day old tracer only. The 30 km path from this
point to the boundary of the marine model adds about
18 days to the computed age. At the mouth of the
estuary, the age reaches about 40 days. This gure is
the result of the competition between the incoming
ux, tending to reduce the age by bringing new
material, and the advection and diusion of older
material from oshore into the area. The estimated
age is slightly smaller than the usually assumed ush-
ing time of the whole Schledt (e.g. Regnier et al.,
1997) but it should be kept in mind that the outlet of
the Grote Schijn is located some 15 km downstream
Antwerpen and that the current velocity is much
larger in this river than in the Zeeschelde itself. Along
the Belgian coast, the age then increases to 4050
days. This time scale can be understood as the time
required, under average conditions, for a change in
the source behaviour to have a signicant eect on the
Belgian coast. It is the memory of this area with
respect to the source considered. Further oshore,
along the residual gyre, the age increases to 6080
days for the largest part. In the centre of the gyre, the
age can even reach 100 days, but, of course, this is
then associated also with much smaller concentrations
of the tracer.
The transient dynamics of the plume has also a
clear impact on the age distribution. When the plume
is pushed northwards by an appropriate wind event,
the age distribution shows a maximum around the
Dutch coast as older material is advected in this
7-Jun-1993
0.005
0.050
0.115
2-Jul-1993
0.005
0.035
0.110
27-Jul-1993
0.015
0.025
0.110
21-Aug-1993
0.020
0.075
0.135
15-Sep-1993
0.005
0.050
0.135
10-Oct-1993
0.010
0.040
0.100
0.010
0.010
0.005
0.030
0.005
0.005
0.060
0.005
0.020
0.075
0.010
0.020
0.030
0.070
F 11. Tracer plume in the Belgian Coastal Zone: surface concentration (g m
3
).
Integrated modelling of the Belgian Coastal Zone 487
direction. Later, when the plume moves back south-
wards, there is a signicant amount of old material
that is trapped in the centre of the gyre where the age
has then its maximum value as inferred from the
observation and the distribution of the successive
trophic levels in the area (Nihoul & Hecq, 1984).
At the marine side of the estuary, a sharp gradient
usually exists between the young water mass coming
out from the Scheldt and a much older water mass
that has already circulated around the whole loop
before ending up at this location.
At the end of the simulation, nearly 98% of the
mass of the tracer released in the river has reached the
sea (Figure 10). The 2% left are still in the river, for
the most part and only a small part can inltrate into
the groundwater. Usually, the river drains the aquifer
and, even during the summer when the water table is
low, no signicant mass transfer of the tracer into the
groundwater domain can be observed. The very small
mass uxes eectively exchanged during the short
periods of time when the river level increases are
progressively compensated afterwards by a return ux
back to the river, and consequently the computed
tracer plume is very limited in the groundwater
domain.
In the simulation considered here, a pumping well
has been considered near the contaminated river (at a
distance of 230 m) to examine the potential transfer of
tracers from the river into the groundwater domain. It
is observed that a pumping rate as small as
002 m
3
s
1
is enough to induce a nearly continuous
mass ux from the river into the groundwater domain.
As shown in Figure 10, this mass transfer is not
signicant for the river and ocean system budgets. It is
7-Jun-1993
35
30
25
2-Jul-1993
50
27-Jul-1993
50
70
30
21-Aug-1993
50
30
15-Sep-1993
70
30
10-Oct-1993
80
90
30
40
60
70
40
60
40
50
60
40
60
50
100
20
20
15
60
30
20
40
70
80
60
100
80
60
50
60
F 12. Surface age (in days) distribution in the Belgian Coastal Zone computed from the comparison of the conservative
and radioactive tracers.
488 E. J. M. Delhez and G. Carabin
however highly signicant for the groundwater both in
terms of resulting concentrations and persistence of
the contamination in the groundwater.
The computed tracer mass uxes exchanged
between the river and the groundwater domain near
the pumping well are shown in Figure 13 as well as the
simulated uctuations of the water levels in the river
and in the groundwater domain. It can be observed
that the general trend of the computed exchanges is
imposed by the slight variations of piezometric head in
the groundwater domain whereas the short and
intense variations of the exchanged mass uxes are
mainly induced by the water level uctuations in the
river.
1.75
1-Nov-93
3.75
Date
Water level in the river
W
a
t
e
r

l
e
v
e
l

(
m
)
29-Apr-93
2.00
2.25
2.50
2.75
3.00
3.25
3.50
30-May-93
Tracer flux between the river and the groundwater domain
Water level in the pumping well
30-Jun-93 31-Jul-93 31-Aug-93 1-Oct-93
0.2
0.6
0.1
0
0.1
0.2
0.3
0.4
0.5
E
x
c
h
a
n
g
e
d

m
a
s
s

f
l
u
x

o
f

t
r
a
c
e
r

(
g

s

1
)
F 13. Tracer ux between the river and the groundwater domain near the pumping well and water level uctuations
in the river and the groundwater.
0.1
1-Nov-93
0.5
Date
Exchanged between the river and the groundwater domain
M
a
s
s

f
l
u
x

o
f

t
h
e

c
o
n
s
e
r
v
a
t
i
v
e

t
r
a
c
e
r

(
g

s

1
)
29-Apr-93
0
0.1
0.2
0.3
0.4
30-May-93
Extracted from the aquifer at the pumping well
30-Jun-93 31-Jul-93 31-Aug-93 1-Oct-93
F 14. Temporal evolution of the uxes of the conservative tracer exchanged between the groundwater and the river and
extracted at the pumping well.
Integrated modelling of the Belgian Coastal Zone 489
Figure 14 shows the computed time evolution of the
uxes of the conservative tracer through the area
inuenced by the pumping well. (Fluxes from the
river into the groundwater domain are considered as
positive values.) The tracer mass extracted from the
aquifer at the pumping well is also shown on this
gure. One can observe that a delay of about 40 days
is necessary before the tracer can be detected in the
pumping well. The extracted mass ux of the tracer
does not exceed 012 g s
1
, i.e. 024% of the injected
ux in the river. This value is the maximum of the
total mass ux of tracer which denitively goes out of
the river into the aquifer in the pumping well inuence
zone.
From the injection point in the river to the pumping
well in the aquifer, a total transfer time varying
between 84 and 93 days is calculated using the ratio
between the concentrations of conservative and radio-
active tracers. Then a mean velocity in the aquifer can
be deduced as varying between 25 and 3 m d
1
.
These values are in range of those deduced from the
classic application of the Darcys law for a mean
gradient of 06%, hydraulic conductivity of about
10
4
m s
1
and eective porosity of 002. It must be
noted that this small value of the eective porosity
explains the signicant transfer of the tracer computed
during the considered period of time. The results may
be transposed to other conditions or locations by
simple dimensional arguments; for instance, the con-
clusions would remain unchanged in the case of
eective porosity of 02 provided that the space scale
and dispersivity values are divided by 10 (the pumping
well would then be located at about 23 m from the
river).
Concluding remarks
The SALMON model provides an appropriate frame-
work for the integrated modelling of the groundwater,
the river and the coastal sea and for the assessment of
their common response to meteorological conditions
and human inuences. As such it forms the ideal tool
to address the complexity and intricate aspects of the
water system in coastal plains.
The scientic advantages of the simultaneous mod-
elling of the groundwater, the river and the coastal sea
are particularly obvious in a management perspective.
On the one hand, the integrated modelling provides
a way to analyse all the eects of management policies
or environmental hazards. It is indeed usual to
approach local problems at a local scale, disregarding
the larger scale impacts. The river eutrophication
issue, for instance, is often addressed locally by appar-
ently appropriate techniques but without any consid-
eration for the possible inuences of these measures
downstream (where large amounts of dead organic
matter may ask for a better solution) or for the
possible memory eect of the groundwater. The cur-
rent approach allows all those aspects to be taken into
account simultaneously.
On the other hand, the integrated modelling
enables also the testing of scenariosreecting man-
agement decisions or the inuence of global
changesin a much more realistic way. The dynamics
of local models, like the model of the Belgian coast, is
indeed driven, to a large extent, by its boundary
conditions. To investigate the eects of some
scenarios, it is therefore of prime importance that the
boundary conditions reect also the same scenario. In
the approach developed in SALMON, the changes
introduced in the scenario will be reected in all the
dierent models for a more accurate prediction of the
global inuences.
The methodology developed in the SALMON
project is meant to be easily portable from one site to
the other. Although the mathematical and numerical
models have to be adapted every time a new site is
considered, the uxes of information to be exchanged
between the dierent models and the appropriate
handling of the dierent time scales and length scales
addressed by these models have been identied and
implemented in a general fashion in SALMON. The
tool is therefore ready to address more specic
problems in a global way.
The rst application of the SALMON meta-model
to the Scheldt Basin and the Belgian Coastal Zone has
highlighted the interactions between the groundwater,
the river and the marine domains. The dierent time
scales of interactions have been determined from
numerical experiments by following a conservative
tracer owing from one domain to the other.
It is now planned to implement a detailed descrip-
tion of the complex biochemical interactions taking
place in the estuary and in the coastal sea, to extend
the scope of the SALMON model and describe the
fate of organic and inorganic nutrients without any
modication to the architecture of SALMON. It will
then be possible, for instance, to have a global test of
the application of the European Directive EC/91/271,
concerning the urban waste water treatment or
to examine the local and remote inuences of
denitrication and dephosphatation post-treatments.
Acknowledgements
The authors are indebted to IBM for its support
within IBM ERPSALMON. Thanks are also
expressed to the Belgian Geological Survey (BGS), to
490 E. J. M. Delhez and G. Carabin
the Flemish Administration for Environment, Water
and Infrastructure (AMINAL) and to the Center for
Nuclear Power in Mol (CEN/SCK) for their collab-
oration in the data collection needed to apply the
groundwater model to the chosen domain.
The authors also wish to thank the anonymous
referees for their fruitful comments.
References
Beckers, J. M. & Deleersnijder, E. 1992 Stability of a FBTCS
scheme applied to the propagation of shallow-water inertia-
gravity waves on various space grids. Journal of Computational
Physics 108, 95104.
Carabin, G., Dassargues, A. & Brouye`re, S. 1998 3D ow and
transport groundwater modelling including river interactions.
Comp. Meth. in Water Resources XII Vol. 1, 569576.
Delhez, E. J. M. & Martin, G. 1992 Preliminary results of 3-D
baroclinic numerical models of the mesoscale and macroscale
circulations on the North-Western European Continental Shelf.
Journal of Marine Systems 3, 423440.
Delhez, E. J. M. 1996a Reconnaissance of the general circulation of
the North-Western European Continental Shelf. Earth Science
Reviews 41, 329.
Delhez, E. J. M. 1996b On the residual advection of passive
constituents. Journal of Marine Systems 8, 147169.
Delhez, E. J. M., Gregoire, M., Nihoul, J. C. J. & Beckers, J.-M.
1999 Dissection of the GHER turbulence closure scheme.
Journal of Marine Systems 21, 379397.
Feng, S., Cheng, R. T. & Xi, P. 1986 On tide induced Lagrangian
residual transport, 2. Residual transport with application in South
San Francisco Bay, California. Water Resource Research 22, 1635
1646.
Ghiot, C., Bauler, P., Beckers, J.-M., Delhez, E. J. M., Carabin, G.,
Dassargues, A., Delie`ge, J.-F. & Everbecq, E. 1998 Sea Air Land
modelling operational network. In Hydroinformatics 98 (Babovic,
V. & Larsen, L. C., eds). Balkema, Rotterdam, pp. 371377.
Hedstrom, G. W. 1979 Nonreecting Boundary Conditions for
Non-Linear Hyperbolic Systems. Journal of Computational Physics
30, 222237.
Huthnance, J. 1973 Tidal current asymmetries over the Norfolk
Sandbanks. Estuarine, Coastal and Shelf Science 1, 8999.
IGBP, 1993 The LOICZ Science Plan. IBBP Report No 25.
Stockholm, Sweden, 50 pp.
Mommaerts, J. P. 1991 Distribution and trends of contaminants in
seawater and sediments of the Southern Bight of the North Sea
and the Western Scheldt Estuary. Technical Report, Management
Unit of the Mathematical Models of the North Sea, MUMM,
Brussels.
Nihoul, J. C. J. & Hecq, J.-H. 1984 Inuence of the residual
circulation on the physico-chemical characteristics of water
masses and the dynamics of ecosystems in the Belgian coastal
zone. Continental Shelf Research 3, 167174.
Nihoul, J. C. J., Deleersnijder, E. & Djenidi, S. 1989 Modelling the
general circulation of shelf seas by 3D k- models. Earth Science
Reviews 26, 163189.
Regnier, P., Wollast, R. & Steefel, C. I. 1997 Long-term uxes of
reactive species in macrotidal estuaries: estimates from a fully
transient multicomponent reaction-transport model. Marine
Chemistry 58, 127145.
Red, L. P. & Cooper, C. K. 1987 A study of various open
boundary conditions for wind-forced barotropic numerical ocean
models. In Three-dimensional Models of Marine and Estuarine
Dynamics (Nihoul J. C. J. & Jamart, B., eds). Elsevier,
Amsterdam, pp. 305335.
Smitz, J., Everbecq, E., Deliege, J. F., Descy, J. P. & Vanderborght,
J. P. 1997 PEGASEPlanication et Gestion de lASsainisse-
ment des Eaux. Tribune de lEau, CEBEDOC, Belgium.
Sweby, P. K. 1984 High resolution schemes using ux limiters for
hyperbolic conservation laws. SIAM Journal of Numerical Analysis
21, 9951011.
Yang, L. 1998 Modelling of Hydrodynamic Processes in the Belgian
Coastal Zone. Ph D. Thesis, Katholieke Universiteit Leuven,
Faculteit Wetenschap, 214 pp.
Zimmerman, J. T. F. 1980 Vorticity transfer by tidal currents over
and irregular topography. Journal of Marine Research 38, 601
630.
Integrated modelling of the Belgian Coastal Zone 491