Marine Pollution Bulletin 61 (2010) 124131

Contents lists available at ScienceDirect

Marine Pollution Bulletin

journal homepage: www.elsevier.com/locate/marpolbul

A modelling assessment of contaminant distributions in the Severn Estuary

N. Murdoch a,*, P.J.C. Jonas a, R.A. Falconer b, B. Lin b
a
b

Environment Agency, Manley House, Kestrel Way, Exeter, Devon EX2 7LQ, UK
School of Engineering, Cardiff University, The Parade, Cardiff CF24 3AA, UK

a r t i c l e

i n f o

Keywords:
Severn
Water quality
Metals
Numerical modelling
Regulation

a b s t r a c t
The regulatory requirements imposed by the Habitats Directive (EU 93/43/EEC) require the Environment
Agency for England and Wales (EA) to review consented discharges and determine whether they are compliant with Environmental Quality Standards (EQS). Since the EQS are annual averages, model predictions, and sample comparisons, should be made on an annual average basis. Advection and dispersion
of metal contaminants in the Severn Estuary were computed using a coupled 1-D and 2-D hydrodynamic-water quality model. The external inputs of dissolved copper, arsenic, mercury and chromium
to the model were from 66 industrial discharges and sewage treatment works and 30 rivers. The annual
average predicted concentrations were compared with the annual average dissolved metal concentrations from the 2004 and 2005 monitoring programme, and any discrepancy used to identify the role of
additional processes, mainly involving the sediments. This ability to separate anthropogenic inputs from
internal estuarine processes contributes to a better understanding of the functioning of the estuary and
hence an improved management capability. The paper discusses the approach in designing scenarios and
characterising uncertainty, when decision-making in the regulatory context.
2009 Elsevier Ltd. All rights reserved.

1. Introduction
Water quality in the Severn Estuary is subject to inuences from
many sources. The major rivers, including the Severn, Wye, Bristol
Avon, Usk, Taff and Parret with a combined annual mean ow of
295 m3 s1 deliver metal contaminants from their catchments into
the estuary (Fig. 1). An additional 24 smaller rivers with a combined annual mean ow of 125 m3 s1 add to the contaminant
load. Direct discharges, with a combined annual mean ow of
16 m3 s1 from industry and sewage treatment works, complete
the inventory of external loadings to the estuary. Generally, rivers
supply a greater proportion of the nutrient load, whilst dissolved
metals appear to be dominated by the direct discharges (Jonas
and Millward, this volume). Metals exist either bound to suspended particulate material (SPM) or as dissolved species, with
their relative proportions being characterised by a partition coefcient, which depends to a large extent on estuarine chemistry
(Turner and Millward, 2002). Dissolved and particulate concentrations of metals are also inuenced by hydrodynamic processes of
dilution, advection and dispersion. Although recent research suggests that metal concentrations in sediments from the estuary
are declining (Duquesne et al., 2006; Langston et al., 2007, this volume; Langston and Millward, 2008), in some cases they may be
acting as an additional internal source of or sink for metals (Jonas
and Millward, this volume).
* Corresponding author. Tel.: +44 1392 352345; fax: +44 1392 352371.
E-mail address: Neil.Murdoch@Environment-Agency.Gov.UK (N. Murdoch).
0025-326X/$ - see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.marpolbul.2009.12.019

The hydrodynamics and water circulation of the Severn Estuary

and Bristol Channel are the main agents for the dispersion and
dilution of contaminants. Previous modelling studies of contaminants in the Severn have examined the distributions of cadmium
(Radford et al., 1981) and caesium-137 (Uncles, 1979; Rattray
and Uncles, 1983). In a study aimed at predicting the distributions
of caesium-137, Rattray and Uncles (1983) developed a 1-dimensional (D) model that showed that the dispersion characteristics
for salinity and a non-conservative tracer were similar and this
could be used to predict estuarine concentrations from dened
sources. Radford et al. (1981) developed a 1-D tidally averaged
model which was used to estimate the advective and dispersive
properties of the estuary. Estimates of the known cadmium loads
to the estuary, at the time, were then used to predict estuarine concentrations of dissolved cadmium. Comparison of the computer
simulations with concentrations of dissolved cadmium in the estuary (i.e. 30 axial transects) showed the crucial importance of establishing accurate and precise values for the magnitudes of the input
loads and hence the need for consistently monitored sources was
emphasised.
There has been signicant development of various numerical
models (Uncles, this volume) for the system, for example a 3-D
model, used to predict tidal and residual currents in the Severn
Estuary, was developed by Wolf (1987). Recently Yang et al.
(2008) applied a 2-D model to the prediction of enteric bacteria
distributions in the Severn Estuary. A conceptual model representing the transport of bacteria due to sediment movement was
developed, including deposition and suspension and the effect of

N. Murdoch et al. / Marine Pollution Bulletin 61 (2010) 124131

125

Fig. 1. The domain of the Severn Estuary covered by the model. N, Sample sites; d, discharge points.

sediment on the transport and decay of enteric bacteria. The basis

of the code developed by Yang et al. (2008) was the DIVAST model
(Lin and Falconer, 1997, 2001; Falconer et al., 2001) which is a coupled hydrodynamic-water quality model that has been widely applied in the UK and overseas. For example, a 2-D version has been
used to model trace metal concentrations in the Mersey estuary
(Wu et al., 2005). Also, Delaney et al. (2007) have incorporated DIVAST into a GIS expert system for habitat management in estuarine
systems. A 3-D version of the model has previously been applied to
the Severn Estuary (Lin and Falconer, 2001). Numerical modelling
is becoming more important in the decision-making process,
although Jones et al. (2002) provided a critical review of the role
of models in estuarine management.
Here we used the established DIVAST model with a view to examining the fate of discharge and river loadings mainly with respect to
advection and dispersion processes, in a similar way to Radford et al.
(1981). This is a cautious approach in that sediment resuspension
processes, including sedimentwater interactions, are not included
in the model. However, this is not to say that they are not relevant
or that the model is incapable of representing them, rather it is considered essential, from an estuary management perspective, to separate out the contributions to dissolved concentrations from the
discharges directly from any estuarine geochemical processes. This
separation is achieved by comparing the model predictions with
measurements of the dissolved concentrations of metals in the estuary. This approach allows identication of potential geochemical
processes that require further investigation.
The aim of the present study is to evaluate the impact on the
estuarine concentrations of the dissolved metal releases from the
currently monitored rivers and discharges. The unique contribution of this work is that for the rst time the combined effect of
a large number of discharges and rivers impacting on the Severn
are examined, which would be very difcult, if not impossible,
without the aid of a numerical model. The results of this study
could then be used to assess whether the estuarine concentrations
resulting from the discharges are compliant with the Environmental Quality Standards (EQS) as required by the Habitats Directive.

2. Modelling approach
The model domain has an upper boundary at the tidal limit of
Maisemore Weir and a lower seaward boundary given by a line between Hartland Point, on the north coast of Devon, and Stackpole
Point in Wales (Fig. 1). The major rivers are listed in Table 1 along
with their ow statistics. In the model, the discharge locations
(Fig. 1) were placed below the low water mark to ensure continuity
of discharge. Also shown are estuarine sample points and the sampling is described in Section 5.
The model is a 2-D, depth-averaged type in the lower estuary
and is dynamically linked to a 1-D model up-estuary, with an
overlap area, located approximately between stations 12 and
13 (Fig. 1). This was set up to facilitate the exchange of variables
between the 1-D and 2-D model regions. At the seaward boundary, tidal level data produced using the Proudman Oceanographic
Laboratory model was used to specify the boundary condition.
River Severn and lateral river inows were as described in the
various simulations below. The resolution of the 2-D grid is
600 metres.

2.1. Hydrodynamic equations

In modelling estuarine and riverine processes, the modelling
domain often covers areas of different physical characteristics,
e.g. large water basins with a 2-D or 3-D ow structures and narrow meandering channels with a predominately 1-D ow structure. For many engineering problems, these physical features are
prevalent in estuaries and rivers; hence the use of a combined 2D and 1-D model has been used in this study to predict accurately
the hydrodynamic and water quality processes.
The hydrodynamic model used to predict the water elevations
and velocity elds in coastal, estuarine and riverine waters initially
involves the solution of the governing equations of uid ow. The
2-D hydrodynamic equations are generally based on the depthintegrated 3-D Reynolds equations for incompressible and unstea-

126

N. Murdoch et al. / Marine Pollution Bulletin 61 (2010) 124131

Table 1
River ow statistics.
River

Easting

Northing

Mean Flow
(m3 s1)

Q95 Flow
(m3 s1)

Q55 Flow
(m3 s1)

Severn
Wye
Usk
Avon
Taff
Parrett
Taw_Torridge
Nedd
Tawe
Ogwar
Ebbw
Rhymney
Brue
Afan
Ely
EastWest Lyn
Frome
Axe
Little Avon
Congresbury Yeo
Keng
Avill River
Doniford Stream
Aller-Horner Water
Washford River
Land Yeo
Portbury Ditch
Banwell
Kilve Stream
Pill River
Total

381548
354231
331798
350115
318218
329130
246482
271881
266598
286123
331480
322282
329428
274556
318583
272291
375173
330852
366257
336494
277919
300883
309059
289210
306997
338862
347817
334900
314335
302706

219350
190223
183633
178583
172672
146844
131180
192432
191653
175787
183805
177474
147527
188667
172672
149678
210497
158536
200314
166748
183473
144247
143213
148512
143524
170310
177420
166682
144453
143520

106.4
90.0
33.3
22.8
22.8
20.1
39.1
13.2
11.6
10.5
8.3
6.5
5.7
5.2
4.8
3.8
2.9
2.3
1.6
1.5
1.2
1.2
1.2
1.2
1.0
0.5
0.4
0.3
0.3
0.3
419.9

19.5
11.3
4.8
4.0
3.7
3.6
2.9
1.6
1.7
1.2
1.3
0.8
0.9
1.0
0.4
0.3
0.5
0.4
0.3
0.3
0.3
0.2
0.2
0.2
0.1
0.1
0.1
0.0
0.0
0.0
61.8

68.2
49.6
19.4
14.4
13.9
12.7
6.5
7.2
6.8
5.5
4.9
3.5
3.4
3.3
2.1
1.8
1.8
1.5
1.0
1.0
0.9
0.8
0.8
0.8
0.6
0.3
0.3
0.2
0.2
0.2
233.7

dy turbulent ows, with the effects of the earths rotation, bottom

friction and wind shear being included to give in the x-direction:

sectional area; R A=P, hydraulic radius and P, wetted perimeter

of the cross-section.

@f @qx @qy

0
@t @x
@y

2.2. Contaminant Transport

@qx
@Uqx
@Vqx
@f sxw sxb

b
b
fqy  gH
@x
@t
@x
@y
q
q
"
#
2
2
@ U @ U
eH

@x2 @y2

Contaminant transport may be described by the following two

dimensional advection dispersion equation:

where H f h, total water column depth; f, water elevation above

(or below) datum; h, water depth below datum; U; V, depth-averaged velocity components in x; y directions; qx ; UH; qy VH, unit
width discharge components (or depth-integrated velocities) in
x; y directions; b, momentum correction factor; f , Coriolis parameter; g, gravitational acceleration; sxw , surface wind shear stress components in x direction; sxb , bed shear stress components in x
directions; and e, depth average eddy viscosity.
A similar equation can be developed for the y-direction as to
that given for the x direction (i.e. Eq. (2)). Likewise for the 1-D river
ow equation assumed that the St. Venant hypotheses were valid
and the corresponding conservation of mass and momentum equations for the 1-D ows were represented as:

@fR @Q R

0
@t
@x

@Q R
@
Q2
b R

@x
@t
A

!
gA

@fR
Q jQ j
g R2 R 0
@x
C z AR

where T, top width of the channel; fR , water elevation above (or below) datum; Q R , discharge; b, momentum correction factor due to
the non-uniform velocity over the cross-section; A, wetted cross-





@
@
@
@
@C
@
@C
HC qx C qy C 
HDx

HDy
@t
@x
@y
@x
@x
@y
@y
advection

kloss C sources

dispersion

where C is the contaminant concentration, Dx and Dy are the respective x and y dispersion coefcients, k is a loss term. The sources term
in Eq. (5) represents the external sources arising from the rivers and
discharges but does not indicate suspended sediment sources (or
sinks). For metals, a more sophisticated treatment based on phase
partitioning may be developed (Wu et al., 2005), although in the
present study this was not included, as mentioned above. The loss
term is generic and is not intended to indicate a specic loss mechanism and is discussed further in Section 3.
2.3. Model spin-up
Model spin-up is the time taken for modelled estuarine concentrations to reach equilibrium and is related to the estuarine water
retention time. For the Severn Estuary this has been estimated to
be between 90 and 200 days. Fig. 2 shows the time series of model
concentrations (for a conservative determinant, with constant uvial inputs) in the upper estuary. The springneap cycles are
clearly evident and the model begins to, but does not completely,
equilibrate after seven springneap cycles. Thus, a spin-up period
of 90 days was used and only dissolved metal concentrations after
this were taken as representative. This period is similar to the esti-

N. Murdoch et al. / Marine Pollution Bulletin 61 (2010) 124131

127

Fig. 2. Time series prole of concentration of a tracer in the upper estuary

indicating the time needed to spin-up the model to achieve equilibrium.

mated ushing time for the estuary derived by Uncles (1984, this
volume).

3. Validation and model tting

3.1. Hydrodynamic validation
The model was validated for the key hydrodynamic parameters.
Tidal range and current data obtained from the Admiralty
Chart No. 1179 were rst used to calibrate the model. A comparison between the model predicted and observed tidal ranges at
the calibration sites show that the degree of agreement generally
being very good at almost all of the sites. The difference between
the predicted tidal ranges and the observations were generally less
than 0.3 m. The model was then used to predict the water level and
velocity data collected in the summer of 2001 at four sites, two
each on the Welsh and English coastlines. For the hydrodynamic
calibration, the agreement between model predicted and measured data has been generally very good for the four sites. Fig. 3
shows an example comparison between modelled and observed
current speed at an offshore location near Minehead, on the English coastline. It can be seen that the model prediction agrees well
with the measured maximum speed and its phase. By way of illustration, Fig. 4 shows velocity vectors from the model at two different phases of the tide.

Fig. 4. Example model output showing the velocity vectors and water levels.

3.2. Water quality model tting

Water quality was modelled using the advection dispersion Eq.
(5), above. The adjustable parameters are the dispersion constants
and the generic loss term. The dispersion parameters were established in a previous study (Lin and Falconer, 2001) and, therefore,
only the generic loss term was adjusted. The main use of the loss

Fig. 3. Validation of model velocities.

128

N. Murdoch et al. / Marine Pollution Bulletin 61 (2010) 124131

term, in this study, is to compensate for inadequate ushing of a

contaminant out of the estuary by the model.
Ideally a conservative tracer would have been available for validation and often salinity is used for this purpose. Because of the
long spin-up, which is much increased for advection from the seaward boundary, the salinity prole did not stabilise even after a
real time simulation of one year. Simulations beyond 12 months
were considered too time consuming to be practical. This could
have been circumvented by initialisation of the salinity eld, but
this was not available at the time of the study. Therefore, another
determinant, oxidised nitrogen, was used as data were available for
all the inputs, i.e. discharges and rivers, and the estuarine sample
points.
The suitability of oxidised nitrogen for model calibration depends onto what extent it is acting conservatively. Conservative
behaviour implies that the de-nitrication rate is zero and it will
not contribute to the loss term (Eq. (5)) which may then be unambiguously interpreted as compensating for inadequate ushing.
De-nitrication losses have been much discussed in the literature.
From an assessment of a number of estuaries Nixon et al. (1996)
suggested that net de-nitrication losses were proportional to
the logarithm of the residence time, with losses less than 20% for
retention times of 1 month or less, rising to 40% for 10 month
(300 days) retention times. There was variability in the data and
losses as low as 10% were indicated for 10 month residence time.
In a study of the Humber Estuary, Barnes and Owens (1999) reported 25% nitrogen loss for a 6 week residence time. In a more recent study on the Humber (Tappin et al., 2003) found that denitrication was small and all consequently all nitrate added to
the estuary was lost to the sea. Neilsen et al. (1995) reported loses
as low as 3% for a short residence time estuary in Denmark and
they also suggested that net losses of > 40% may be over estimated.
Examination of the variation of the observed estuarine concentrations of oxidised nitrogen estuarine with salinity should reveal
the presence of source and loss processes. The EA collected water
samples for dissolved metal analysis (and other analytes), at near
high water, during axial transects of the Severn Estuary, involving
the same xed sampling stations, in August, September and
November 2004 and March, June, August and September 2005 (Jonas and Millward, this volume). Fig. 5a shows that the annual average concentrations of oxidised nitrogen, obtained from the axial
transects, vary inversely with salinity. The approximate linear relationship suggests that losses processes are not playing a signicant
role and the nitrogen appears to be behaving approximately conservatively. Considering the empirical evidence from Fig. 5a it
seems that the nitrogen losses over 100 day water retention time
are low and, consistent with the literature, a lower end value of
less than 15% is estimated.
The model predictions for estuarine concentrations of oxidised
nitrogen were compared with the sample data. For the comparison,
the model was run with constant input values for river and discharge ows and concentrations chosen so as to generate the annual average response within the estuary; the method for
selecting these input values is described in Section 4, below. The
model was run for 90 days spin-up and then over a springneap
cycle with the prediction data being taking from the springneap
period The results were compared with the annual averages of
the corresponding sample data and both are shown in Fig. 5b, as
a function of distance down-estuary from the tidal limit. The
dashed line is the annual average predicted by the model over
the spring to neap tidal period. The upper envelop indicates the
maximum concentrations during the period, at low water, whilst
the lower line shows the minima at high water. Since the water
samples were generally taken as near high water as practical the
appropriate comparator is the lower model line. This line was tted by adjusting the loss constant k.

Fig. 5. (a) Variation of the annual average concentration of oxidised nitrogen with
annual average salinity obtained from a total of seven axial transects of the estuary
in 2004 and 2005; (b) comparison of annual average modelled and sampled
oxidised nitrogen as a function of distance in the estuary, where - - - - - is the
annual average predicted oxidised nitrogen concentrations over a springneap
simulation period;  is the upper limit corresponding to the distribution at low
water; -  -  -  is the lower limit corresponding to the distribution at high water.

The value of k used in the computations corresponded to an efolding time of 50 days, i.e. the loss term in Eq. (5) where the concentration is reduced by 63% in 50 days and 87% in a 100 days. This
is far in excess of the estimated loss by de-nitrication which is
15% over 100 days. Therefore the loss term is primarily compensating for inadequate model ushing and to a lesser extent reecting
de-nitrication. In the subsequent simulations for the dissolved
metals below, the same value of k is used. As part of this k is not
attributable to hydrodynamic processes (i.e. to de-nitrication)
there is an uncertainty in the model water quality predictions of
about 15% arising from the advection dispersion representation.
In summary the model has been set up to dilute, advect and disperse contaminant inputs throughout the estuary with an uncertainty of 15% for annual average predictions.
4. Model scenario design
The context in which this study originated was regulatory and it
is important to reect on the signicance for modelling. Regulatory
organisations use EQS to control contaminant levels in the environment and hence to control discharges to the environment. The EQS
are expressed in statistical form as a percentile or annual average;
and sometimes as a maximum. Appropriate percentile standards
will control short term impacts whilst annual averages address
the long term problem. From a regulatory perspective modelling
is simply a calculation to determine whether or not there is compliance with a particular EQS.

N. Murdoch et al. / Marine Pollution Bulletin 61 (2010) 124131

Table 2
Percentiles used to derive river and discharge ows and concentrations.
Input

Percentile

Flow statistic

River ow
Discharge ow
River concentration
Discharge concentration

45
60
73
73

Q55
Q40

This focus on EQS has inuenced the approach to modelling. In

freshwater modelling in England and Wales, the regulatory paradigm is statistical modelling using Monte-Carlo techniques (Crabtree et al., 2006; Daldorph et al., 2007; Warn, 2003). This allows
a direct comparison between model predictions and the EQS, since
they are both expressed in statistical form. The approach adopted
in the marine modelling eld has been different, due in part to
the greater complexity implicit in numerical codes (e.g. nite difference/nite element) where the emphasis has been decidedly
deterministic and process based. Models have been used in the
regulatory context, but more to provide illustrations surrounding
specic events than to couple to the EQS concept. To some extent
this has limited the application of marine models in this area.
Matching marine model output to an EQS is not straightforward
and is often avoided. This is because EQS statistics, whether percentiles or annual means, (usually) apply over an annual time period. By contrast marine modelling has to some extent pursued
greater sophistication in process representation which entails
longer runs times and so annual runs are often not undertaken outside of the research environment. The result is a tendency for episodic event simulation and no link is established between the
model outputs and an EQS.
Because of long run times scenarios are constructed with xed
freshwater and discharge inputs and run over a springneap cycle
rather than a full year. This approach is adopted here. The problem
of linking the EQS to the xed inputs is addressed using the justied design evaluation method (Murdoch, 2001). The approach
identies percentiles of the input variables (ow and concentrations) which drive an annual average response in the estuary. Here
river and discharge ow and concentration data are represented by
lognormal distributions and the identied percentiles are used to
select the xed input values from these distributions. Table 2 lists
the percentiles so derived and used in this study. There is uncertainty in scenario denition. This arises from the sampling errors
characterising the discharge and river ow and quality distributions and in identifying the percentiles to x the input values.
The errors here were estimated at around 20%. Having calibrated
the model, considered the uncertainties and selected the inputs
the model set up process was complete.
5. Results and discussion
Model runs were performed for dissolved copper, arsenic, mercury and chromium, with the seaward boundary concentrations set
at zero for each metal. The model was spun up for 90 days and then
a spring to neap cycle run from which the axial concentrations
were obtained. All discharge and river ow concentrations were
at xed values derived as described in Section 4. The advection
dispersion Eq. (5), with the loss term mainly compensating for
inadequate model ushing was employed. Therefore, the model
predictions are primarily the result of (conservative) hydrodynamic processes with a combined uncertainty of 30% arising from
the discussion in Sections 3 and 4. The model predictions were
compared with the annual average concentrations of the dissolved
metals calculated from seven axial transects, described in
Section 3.2.

129

The results are shown in Fig. 6 where the predictions of the annual average metal concentrations are compared with those observed in the monitoring programme. The prole of the mean
suspended particulate matter concentrations (SPM) is also shown
for interpretative purposes.
Copper: The observed concentrations of dissolved copper are
highly variable, particularly in the upper estuary, and decrease
down-estuary. Part of the variability arises because in August
2005, the dissolved copper concentrations exceeded the EQS from
the freshwater reaches to about 70 down-estuary (Jonas and Millward, this volume). Consequently, the model ts to the observations vary down-estuary (Fig. 6a), such that up to 20 km downestuary the model gives higher values than the observed concentrations. This is followed by a region between 20 and 60 where
there is better agreement between the predictions and the observations. After about 60 km down-estuary, the model predictions
are higher than the observed concentrations. The variable agreement between the predictions and observations suggests other
processes, involving metal exchanges between the dissolved and
particulate phases, including net uxes of dissolved copper from
the sediment porewaters.
Arsenic: Dissolved arsenic shows relatively high, and most variable, concentrations in the region from the freshwater reaches to
about 20 km down-estuary. Beyond 20 km the dissolved arsenic
concentrations decline and are less variable. The model predictions
indicate arsenic concentrations that are consistently lower than
the observations (Fig. 6b), even allowing for modelling uncertainty.
Whilst the observed sample concentrations are, in part, the result
from the discharges and river inputs it is clear that another source,
or sources, are present in the estuary. The sediments may be a potential reservoir of arsenic, and, copper having accumulated inputs
over several decades and it has the lowest partition coefcient,
KD  104 L kg1, of the metals studied in the Severn Estuary (Jonas
and Millward, this volume).
Mercury: Concentrations of dissolved mercury were at or near
the detection limit of 0.01 lg L1 throughout the system (Fig. 6c).
Only in the freshwater reaches was there evidence of higher dissolved mercury concentrations, albeit with considerable variability. In the upper estuary, the model clearly over-estimates the
concentrations and it is only after 80 km that the predictions agree
with the observations. This suggests other processes may be
important, including uptake of mercury onto particles, even after
considering the modelling uncertainty. Within the Severn Estuary,
Jonas and Millward (this volume) show that mercury has the highest KD (approximately 107 L kg1) of the metals investigated. Thus,
the low observed concentrations generally arise because of the
high SPM loads (Fig. 6e), particularly within the main estuarine
turbidity maximum (Kirby, this volume). With abundant SPM
and an elevated KD it is highly likely that dissolved mercury adsorbs onto SPM, with the sediments potentially acting as a sink
for the metal. Particle-water interactions are not taken account
of in the model which is why it gives signicantly higher dissolved
mercury concentrations than the observations.
Chromium: Concentrations of dissolved chromium are also low
and close to the detection limit of 0.5 lg l1 (Fig. 6d). The model
is persistently over-estimating dissolved chromium, particularly
in the upper estuary. The behaviour of chromium is similar to that
of mercury, in that it has a relatively high partition coefcient (Jonas and Millward, this volume), and is likely adsorbed from solution in the turbid regions of the estuary (Fig. 6e). As is the case
with mercury, it appears that the sediments are acting as a sink
for chromium.
In summary, comparing the proles of the concentrations of
dissolved metals with the SPM concentrations, the clearest indication of the inuence of the high SPM load are for mercury and chromium, for both these elements the main SPM maximum identies

130

N. Murdoch et al. / Marine Pollution Bulletin 61 (2010) 124131

Fig. 6. Comparison of the annual average modelled and the annual average (1r) sampled dissolved metals and the annual average SPM as a function of distance in the
estuary, where - - - - - is the annual average predicted dissolved metal concentration over a springneap simulation period;  is the upper concentration limit
corresponding to the distribution at low water and -  -  -  is the lower concentration limit corresponding to the distribution at high water. (a) copper; (b) arsenic; (c)
mercury; (d) chromium; and (e) SPM.

with the largest discrepancy between the model over-prediction

and the observations. This is consistent with the high partition
coefcients for these metals which resulted in lower dissolved
concentrations.
From the regulatory point of view it is reassuring that both the
observed and model predictions are within the values for the individual EQSs (Table 3), with the exception of copper in August 2005.
The sample data were mainly compliant with the EQS (86% of the
total discharge load was in dissolved phase). The most interesting
aspects here are in fact the discrepancies. By separating out the
inuences of the discharges the role played by the sediments begins to emerge. It is not claimed here that a full understanding
has been developed but the model has revealed potential sedimentary inuences that need to be claried by further research. This is

Table 3
Water quality contaminants and annual average Environmental Quality Standards.
Determinant

EQS, lg l1

Copper (dissolved)
Arsenic (total)
Mercury (dissolved)
Chromium (dissolved)

5
25
0.3
15

relevant to the long term prognosis with or without a Severn Barrage; that is whether, in the long-tem, the estuarine sediments will
act as a source or sink of dissolved metals.

N. Murdoch et al. / Marine Pollution Bulletin 61 (2010) 124131

6. Summary and conclusion

The fate of external loadings of four dissolved metals from a
large number of discharges and rivers into the Severn Estuary
has been investigated. The combined application of a numerical
modelling and sampling provides a powerful tool to disentangle
the role of suspended sediment sources and sinks from the external loadings. Mercury and chromium appear to be removed to
the particulate phase and, therefore, the sediments act, potentially,
as a sink. Copper and arsenic have relatively low partition coefcients and appear more geochemically mobile than mercury and
chromium, with proportionally more metal in the dissolved phase.
Since particulate copper and arsenic are labile they are potentially
desorbed into anoxic porewaters and are, therefore, available for
remobilisation into the overlying water column. For copper and arsenic, it appears that the sediments are acting as a source of dissolved metal.
The focus on annual averages has been driven by regulatory
concerns over EQS. From the regulatory perspective the model
and sample results suggest compliance with the EQS, which are annual averages; this is not to meant to imply that individual sample
concentrations will always be less than EQS but only that on an annual averaged basis is this the case.
The implications for the modelling community from this study
are that greater attention to simulation outputs being compatible
with EQS will help the regulator. Further clarity on the modelling
uncertainty and the long term contaminant budget (i.e. input/output loads) are important requirements.
As far as the Severn is concerned further research on seasonal
and annual load budgets and the interplay of external and internal
loadings would help us to better understand the long term evolution of the system. This paper offers a beginning on this.
Acknowledgements
The Environment Agency (South West) acknowledges MarCon
Computations International for developing the GIS interface under
contract. The views expressed here are not necessarily those of the
Environment Agency.
References
Barnes, J., Owens, N.J.P., 1999. Denitrication and nitrous oxide concentrations in
the Humber Estuary, UK and adjacent coastal zones. Marine Pollution Bulletin
37, 247260.
Crabtree, B., Seward, A.J., Thompson, L., 2006. A case study of regional catchment
water quality modelling to identify pollution control requirements. Water
Science and Technology 53, 4754.
Daldorph, P.W.G., Spraggs, G.E., Lees, M.J., Wheater, H.S., Chapra, S.C., 2007.
Integrated lake and catchment phosphorus model a eutrophication

131

management tool. II: Application to Rutland Water. Water and Environment

Journal 15, 182189.
Delaney, C., Berry, A., Hartnett, M., Simmons, P., 2007. A description of a
sophisticated GIS based expert system for use in habitat management in
dynamic estuarine environments. In: Proceedings of the 8th International
Symposium on GIS and Computer Mapping for Coastal Zone Management, vol.
2, pp. 413423.
Duquesne, S., Newton, L.C., Giusti, L., Marriott, S.B., Stark, H.J., Bird, D.J., 2006.
Evidence for declining levels of heavy-metals in the Severn Estuary and Bristol
Channel, U.K. and their spatial distribution in sediments. Environmental
Pollution 143, 187196.
Falconer, R.A., Lin, B., Wu, Y., Harris, E., 2001. DIVAST Reference Manual. Manual
Environmental Water Management Research Centre, Cardiff University.
Jones, P.D., Tyler, A.O., Wither, A.W., 2002. Decision-support systems: do they have
a future in estuarine management? Estuarine. Coastal and Shelf Science 55,
9931008.
Langston, W.J., Millward, G.E., 2008. Review of Contaminant Levels in the Severn
Estuary SAC, SPA, Plymouth Marine Science Partnership, July 2008, p. 95.
Langston, W.J., Chesman, B.S., Burt, G.R., Campbell, M., Manning, A.J., Jonas, P.J.C.,
2007. The Severn Estuary: Sediments, Contaminants and Biota. Marine
Biological Association of the UK Occasional Publication, No. 19, p. 176.
Lin, B., Falconer, R.A., 1997. Tidal ow and transport modelling using the ULTIMATE
QUICKEST scheme. Journal of Hydraulic Engineering ASCE 123, 303314.
Lin, B., Falconer, R.A., 2001. Numerical modelling of 3-D tidal currents and water
quality indicators in the Bristol Channel. In: Proceedings of Institution of Civil
Engineers, Water and Maritime Engineering, vol. 148, pp. 155166.
Murdoch, N., 2001. Justied design evaluation for percentile standards.
Environmental Modelling and Software 16, 725738.
Neilsen, K., Nielsen, L.P., Rasmussen, P., 1995. Estuarine nitrogen retention
independently estimated by denitrication rate and mass balance methods: a
study of Norsminde Fjord, Denmark. Marine Ecology Progress Series 119, 275
283.
Nixon, S.W., Ammerman, J.W., Atkinson, L.P., Berounsky, V.M., Billen, G., Boicourt,
W.C., Boynton, W.R., Church, T.M., DiToro, D.M., Elmgren, R., Garber, J.H., Giblin,
A.E., Jahnke, R.A., Owens, N.J.P., Pilson, M.E.Q., Seitzinger, S.P., 1996. The fate of
nitrogen and phosphorus at the land-sea margin of the North Atlantic Ocean.
Biogeochemistry 35, 141180.
Radford, P.J., Uncles, R.J., Morris, A.W., 1981. Simulating the impact of technological
change on dissolved cadmium distribution in the Severn estuary. Water
Research 15, 10451052.
Rattray, M., Uncles, R.J., 1983. On the predictability of the 137Cs distribution in the
Severn Estuary, Estuarine. Coastal and Shelf Science 16, 475487.
Tappin, A.D., Harris, J.R.W., Uncles, R.J., 2003. The uxes and transformations of
suspended particles, carbon and nitrogen in the Humber estuarine system (UK)
from 1994 to 1996: results from an integrated observation and modelling study.
The Science of the Total Environment 314316, 665713.
Turner, A., Millward, G.E., 2002. Suspended particles: their role in estuarine
biochemical cycles. Estuarine, Coastal and Shelf Science 55, 857883.
Uncles, R.J., 1979. A comparison of the axial distributions of salt and 137Cs in the
Severn Estuary during August 1974. Estuarine and Coastal Marine Science 9,
585594.
Uncles, R.J., 1984. Hydrodynamics of the Bristol Channel. Marine Pollution Bulletin
15, 4753.
Warn, T., 2003. SIMCAT 9: A Guide and Reference for Users, Environment Agency.
Wolf, J., 1987. A 3-D model of the Severn Estuary. In: Nihoul, J.C.J., Jamart, B.M.
(Eds.), Three-Dimensional Models of Marine and Estuarine Dynamics. Elsevier,
Amsterdam, pp. 609624.
Wu, Y., Falconer, R.A., Lin, B., 2005. Modelling trace metal concentration
distributions in estuarine waters, Estuarine. Coastal and Shelf Science 64,
699709.
Yang, L., Lin, B., Falconer, R.A., 2008. Modelling enteric bacteria levels in coastal and
estuarine waters. In: Proceedings of Institution of Civil Engineers, Engineering
and Computational Mechanics, vol. 161 (EM4), pp. 179186 (December 2008).