AQUACULTURE ENVIRONMENT INTERACTIONS
Aquacult Environ Interact
Vol. 3: 275–291, 2013
doi: 10.3354/aei00068
Published online June 4
INTRODUCTION
Aquaculture, and in particular sea-cage fish farm-
ing, is a significant primary industry that is under -
going rapid expansion worldwide. The immediate
and obvious environmental impacts associated with
finfish farming are well documented (e.g. Gowen
& Brad bury 1987, Brooks et al. 2002, Brooks &
Mahnken 2003, Kalantzi & Karakassis 2006). Seabed
effects tend to be localised and are typically routinely
monitored, with the results used to regulate the
intensity of the aquaculture activity (Wilson et al.
2009). Depositional models are a useful tool for both
predicting and managing seabed effects, as they
combine physical and behavioural properties of
water and particles with farm configuration and pro-
duction parameters to predict the distribution and
intensity of waste products (Cromey et al. 2002a). In
New Zealand, as in many other Southern Hemi-
sphere countries, caged fish-farming is a developing
© The authors 2013. Open Access under Creative Commons by
Attribution Licence. Use, distribution and reproduction are un -
restricted. Authors and original publication must be credited.
Publisher: Inter-Research · www.int-res.com
*Email: nigel.keeley@cawthron.org.nz
Predictive depositional modelling (DEPOMOD) of
the interactive effect of current flow and resuspen-
sion on ecological impacts beneath salmon farms
N. B. Keeley
1,2,
*
, C. J. Cromey
3
, E. O. Goodwin
1
, M. T. Gibbs
4
, C. M. Macleod
2
1
Cawthron Institute, Nelson 7010, New Zealand
2
Institute of Marine & Antarctic Science (IMAS), University of Tasmania, Private Bag 86, Hobart, Tasmania 7001, Australia
3
SV ‘Whanake’, Falkland Islands, South Atlantic Ocean
4
AECOM, Fortitude Valley, Brisbane, Queensland 4006, Australia
ABSTRACT: Sediment resuspension is an important factor in controlling the impact of any
localised point source impacts such as salmon farms; at high-flow (dispersive) sites, resuspension
can significantly reduce potential effects. Depositional modelling (DEPOMOD) is widely used to
predict localised seabed impacts and includes an optional flow-related resuspension module. This
study examined the observed impacts at 5 farms with contrasting flow regimes to evaluate the role
of modelled resuspension dynamics in determining impacts. When resuspension was included in
the model, net particle export (i.e. no significant net downward flux of organic material) was pre-
dicted at the most dispersive sites. However, significant seabed effects were observed, suggesting
that although the model outputs were theoretically plausible, they were inconsistent with the
observational data. When the model was run without resuspension, the results were consistent
with the field survey data. This retrospective validation allows a more realistic estimation of the
depositional flux required, suggesting that approximately twice the flux was needed to induce an
effect level at the dispersive sites equivalent to that at the non-dispersive sites. Moderate enrich-
ment was associated with a flux of ~0.4 and ~1 kg m
−2
yr
−1
, whilst highly enriched conditions
occurred in response to 6 and 13 kg m
−2
yr
−1
, for low and dispersive sites, respectively. This study
shows that the association between current flow, sediment resuspension and ecological impacts is
more complex than presently encapsulated within DEPOMOD. Consequently, where depositional
models are employed at dispersive sites, validation data should be obtained to ensure that the
impacts are accurately predicted.
KEY WORDS: Aquaculture · Benthic · Biodeposition · Enrichment · Dispersive · Depositional
modelling · DEPOMOD · Marlborough Sounds
O
PEN
PEN
A
CCESS
CCESS
Aquacult Environ Interact 3: 275–291, 2013
industry, and accurately predicting impacts and
ensuring that farms are properly situated are critical
steps in the planning and permitting process.
The numerical algorithms that describe the physi-
cal processes underpinning the advection, dispersion
and accretion of particles in most deposition models
are valid across a wide range of environments, pro-
vided the model boundary conditions are adequately
described. DEPOMOD (Cromey et al. 2002a) is prob-
ably the most established and widely used deposi-
tional model for the purposes of predicting salmon
farm effects, largely because it has been proven in a
wide range of environments and is considered to be
robust and credible (SEPA 2005, ASC 2012). Some
of the key input parameters that are required, such
as observations of current dynamics, bathymetry and
basic farming practice information (e.g. cage layout,
feed characteristics and input rates), are relatively
easy to obtain, whilst others can be more difficult
to quantify (e.g. feed wastage, critical erosion thresh-
olds). In these latter cases, default data can be
employed as long as the model is not overly sensitive
to these parameters. As a result, it is possible to trans-
fer a depositional model that has been developed in
one environment to another region, often with only
minor alterations. For example, although DEPOMOD
was developed for salmon farming in cool temper-
ate systems, it has been applied successfully to cod
farming (CODMOD, Cromey et al. 2009), and to both
warm-temperate culture of sea bream and bass
(i.e. MERAMOD, Cromey et al. 2012) and more
recently tropical fish-culture (i.e. TROPOMOD, C. J.
Cromey pers. obs.). The validation process for these
new applications was relatively straightforward and
only required site-specific data and the inclusion of a
few new processes (e.g. wild fish feeding), indicating
that the physical components were on the whole
comparable and transferable.
Although the primary components of the models
are generally transferable, the relationship between
depositional flux and ecological response can be
strongly influenced by physical environmental prop-
erties, and is therefore site-specific. Sediment type
(i.e. sand versus mud; Kalantzi & Karakassis 2006,
Papageorgiou et al. 2010) and flow regime (Macleod
et al. 2007, Mayor & Solan 2011, Keeley et al. 2013)
will each influence ecological responses. Dispersive
sites (i.e. with strong currents) will respond charac-
teristically differently to organic enrichment and are
potentially more resilient to benthic effects (Frid &
Mercer 1989, Borja et al. 2009, Keeley et al. 2013),
with the total seabed area measurably affected
by farming—hereafter termed the ‘footprint’—often
being noticeably larger and more diffuse (Keeley et
al. 2013). Nevertheless, strong biological responses
can occur at dispersive sites (Chamberlain & Stucchi
2007), as evidenced by very high macrofaunal abun-
dances and biomass in the immediate vi cinity of the
cages (Keeley et al. 2012a). These differences can
largely be attributed to the stronger currents, which
increase initial particle dispersal (Cromey et al.
2002b), and provide an increased oxygen supply
buffering against near-bottom anoxia (Findlay &
Watling 1997). Presumably, greater resuspension
also plays an important role, re-entraining and re-
distributing particles post-settlement and thereby
limiting excessive organic accumulation and related
ecological effects (Keeley et al. 2013). However, the
validity of including resuspension in depositional
models remains in question, as its inclusion can
strongly influence the results, and the optimum
critical velocity threshold (vr) to use is debatable
(Chamberlain & Stucchi 2007).
The ability to clearly and quantitatively link pre-
dictions of depositional flux to predictions of ecologi-
cal effects would greatly increase the usefulness of
depositional models. Connecting the mathematical
theory and the ecology is essential if the models are
to be used for managing farms in relation to benthic
effects, i.e. setting maximal and optimal feed levels
and/or fine-scale positioning of cages. Studies have
been conducted with respect to relatively unique and
sensitive communities such as maerl beds (Sanz-
Lázaro et al. 2011) and seagrass habitats (Apostolaki
et al. 2007, Holmer et al. 2008), or assessing lower
tolerance thresholds, where impacts are initially ob -
served (Hargrave 1994, Findlay & Watling 1997,
Chamberlain & Stucchi 2007, Cromey et al. 2012).
These studies suggest that ecological effects can be
observed across a broad range of depositional flux
levels spanning 2 orders of magnitude (i.e. between
0.1 and 10 kg solids m
−2
yr
−1
), and the results are dif-
ficult to compare due to differences in the enrich-
ment criteria or ecological thresholds that have been
adopted. Additionally, efforts to relate deposition to
benthic responses empirically have focussed on a rel-
atively limited suite of biological indicators, e.g. total
macrofaunal abundance, the infaunal trophic index
(ITI; Cromey et al. 2002a), biomass, Shannon-Wiener
diversity (H’) and the biological fraction index
(Cromey et al. 2012). However, relationships with
other biotic indices that can be more effective for
discerning benthic enrichment status are yet to be
established (e.g. the AZTI marine biotic index:
AMBI, Multivariate-AMBI and the benthic quality
index: BQI; Borja et al. 2009, Keeley et al. 2012a).
276
Keeley et al.: Predicting salmon farm deposition and effects
Hence, the main aim of this study was to utilise a
long-term benthic monitoring dataset to develop
empirical models that can be used to convert
between predicted flux and observed effects for dis-
persive and non-dispersive sites, and in doing so
contribute to our understanding of the role of re -
suspension. As a component of this study, it was also
necessary to evaluate the strength of the link be -
tween model predictions and observed responses by
examining the fine-scale differences between the
overall size, shape and intensity in the predicted and
observed depositional footprints.
MATERIALS AND METHODS
Study sites and environmental data
We used data obtained from an annual compliance
monitoring programme over 12 yr (1998 to 2010) at
6 salmon farms located within the Marlborough
Sounds, New Zealand (Fig. 1). The farms were situ-
ated at comparable depths (27 to 40 m) and spanned a
range of ages (1 to 25 yr of operation, Table 1). Four of
these farms (A to D) had mean current velocities be-
low 9 cm s
−1
at 20 m water depth (approximately mid-
water), and these are hereafter re ferred to as ‘non-
dispersive’ sites, whereas the other 2 (E and F) had
mean current velocities in excess of 15 cm s
−1
and are
referred to as ‘dispersive’ sites. All of the sites were
situated over unconsolidated sediments; the non-dis-
persive sites tended to be sandy mud (55 to 91%
mud), and the dispersive sites were muddy sand (28
to 32% mud; Table 1). All of the sites had, at some
point, displayed strong enrichment gradients with
proximity to the farms (Keeley et al. 2012a, 2013). The
analyses presented here were conducted on a deliber-
ately broad range of scenarios, whereby the years that
were used for each farm were selected to span a wide
cross-section of total annual feed inputs and therefore
presumably, associated levels of impact (Table 1).
Sediment samples were collected from directly
beneath cages, and at stations along an enrichment
gradient extending away from the cages (25 to
250 m), as well as at reference stations. Macrofauna
were sampled using replicate (n = 2, 3 or 5, depend-
ing on year of survey) Perspex sediment corers
(13 cm diameter, 0.013 m
2
) deployed to a depth of 10
cm. Core contents were sieved to 0.5 mm, and the
retained fauna was identified to the lowest practical
taxonomic level and enumerated, enabling calcula-
tion of a variety of community composition statistics
and biotic indices: N (total abundance), S (number of
taxa), H ’ (Shannon-Wiener diversity), AMBI (Borja et
al. 2000) and BQI (Rosenberg et al. 2004). The sur-
face 3 cm of smaller sediment cores (7 cm diameter)
was collected for analysis of grain size and total
organic matter (OM). Sediments were oven-dried to
constant weight at 105°C, and size class fractions
from silt-clay through to gravel were analysed gravi-
metrically. Percentage OM (%OM) was calculated as
the % weight loss of dried samples after ashing at
550°C for 2 h (modified after Luczak et al. 1997).
Redox potential (Eh
NEH
, mV) and total free sulphide
(TFS, µM) were also routinely measured after 2008.
Re dox was measured directly from the grab (at 1 cm
depth) using a Thermo Scientific combination Redox/
ORP electrode. TFS was sampled with a cut-off 5 cm
3
plastic syringe driven vertically into the surface sedi-
ments (0−4.5 cm depth interval), and the TFS con-
tents were extracted and quantified following the
methods of Wildish et al. (1999).
Bathymetry and hydrography
Bathymetry was established for each site, and the
xyz data were gridded to the desired size and resolu-
277
Fig. 1. Location of study sites in Marlborough Sounds, New
Zealand
Aquacult Environ Interact 3: 275–291, 2013
tion using Surfer v9 for incorporation
into DEPOMOD. Model grid sizes were
set such that they would comfortably
encompass the whole initial depositional
footprint (grid areas ranged from 0.23
km
2
for Farm C to 1.1 km
2
for Farm A).
Water currents were measured using
Acoustic Doppler Current Profilers
(ADCP, Sontek, 500 kHz) every 15, 30 or
45 min intervals over 25 to 42 d. ADCPs
were bottom-mounted within approxi-
mately 30 m from the cage edge and
sampled the water column in 2 or 3 m
depth bins (with a 1 m blanking dis-
tance). Water current data were con-
verted to hourly averaged bins, and the
5 depth bins that evenly spanned the full
water column at each site (i.e. from near
surface to near bottom) were selected for
use in the models (Table 1).
Model parameters
DEPOMOD was selected because it
is widely used and published, and was
designed specifically for managing fish
farm wastes (Cromey et al. 1998, Thet-
meyer et al. 2003, Cromey & Black 2005,
Cook et al. 2006, Magill et al. 2006);
moreover, a number of the processes in
DEPOMOD have already been validated
against field measurements (Cromey et
al. 2002a, Chamberlain & Stucchi 2007).
It is also used as a regulatory tool in
Scotland for discharge consents of in-
feed chemotherapeutants, and in setting
biomass limits (SEPA 2005), and it is the
model that is recommended for predict-
ing sea bed effects by the Aquaculture
Stewardship Council (ASC 2012).
Standard feed wastage (F
wasted
) of 3%
was used for all sites and all years in the
absence of any reliable historical esti -
mations. This level was selected be cause
it represents a compromise between the
level of 5% shown to support predictions
in other studies (Cromey et al. 2009,
2012), and the level most recently deter-
mined in local studies (<1%, Cairney &
Morrisey 2011). Three percent is also
the level currently recommended by the
Scottish Environmental Protection Agency
278
Site attributes Units Farm A Farm B Farm C Farm D Farm E Farm F
Year of survey ‘04, ‘06, ’09, ’10 ‘05, ‘08, ‘10 ‘03, ‘05, ‘09 ’10 ‘05, ’08, ‘09, ‘10 ‘08, ‘09, ‘10
Farm age yr 19, 21, 24, 25 16, 19, 21 14, 16, 20 1
a
13, 16, 17, 18 1, 2, 3
Corresponding feed levels kt yr
−1
1.9, 3.3, 2.2, 2.0 2.2, 2.0, 1.9 2.1, 2.6, 3.1 3.3 2.3, 4.1, 3.9, 4.7 2.8, 3.1, 3.5
Site depth range m 34−35 37−39 28−30 32−35 27−31 30−40
Mean current speed m: cm s
−1
Depth 1 (near-surface) 1: 3.6 (30.0) 1: 8.6 (35.9) 1: 11.9 (59.1) 1: 3.4 (16.1) 1: 18.7 (62.8) 2: 20.4 (87)
Depth 2 7: 4.0 (21.6) 9: 3.7 (46.1) 7: 8.2 (34.3) 8: 3.0 (9.3) 7: 16.7 (59.2) 10: 20.2 (85)
Depth 3 (mid-water) 15: 3.7 (17.5) 16: 6.0 (34.6) 14: 8.2 (29.9) 16: 3.0 (10.1) 14: 14.4 (53.8) 18: 19.9 (117)
Depth 4 22: 3.1 (12.9) 26: 9.7 (30.4) 21: 8.5 (30) 24: 3.2 (11) 21: 13.9 (42.4) 28: 19.7 (129)
Depth 5 (near-bottom) 30: 3.5 (14.2) 34: 3.6 (13.5) 28: 9.1 (29.1) 32: 3.2 (10.9) 26: 15.9 (49.8) 36: 19.5 (79)
ADCP sample bin size m 4 4 1 1 1 2
ADCP sampling interval min 45 45 30 30 30 15
Deployment month January March July February August October
Flow category Low Low Low−mod Low High High
Sampling stations (distance m 0 (×2), 50, 0 (×2), 50, 0 (×2), 50 (×2), 0 (×2), 25, 50, 75, 0 (×2), 50, 0 (×2), 50 (×2),
from cages) 150, 250, Ref 150, 250, Ref 100, Ref 100, 150, 200, Ref 100, Ref 100 (×2), 150 (×2),
200 (×2), 250 (×2), Ref
Natural sediment properties
Sediment mud content % 80 (69−84) 55 (34−73) 78 (69−85) 91 (84−95) 28 (21−38) 32 (29−37)
%Organic matter % w/w 5.2 (4.8−5.8) 5.0 (2.8−7) 4.9 (4.5−5.8) 5.5 (4.4−6.5) 3.1 (2.5−3.7) 3.3 (2.5−4.2)
No. taxa No. core
−1
22 (18−28) 18 (17−19) 20 (16−23) 21 (11−26) 35 (27−48) 39 (31−42)
Macrofauna abundance No. core
−1
107 (76−147) 72 (52−92) 78 (37−128) 54 (18−72) 218 (152−285) 231 (102−278)
a
Farm had been reinstated for 1 yr at time of monitoring after 8 yr of recovering since being fallowed in 2001
Table 1. Summary of farm configurations, historical use and physical attributes used in models, and of natural (reference site, Ref) sediment characteristics for each site.
Current speeds for each depth bin (in m; left of colon) are mean values, with maximum speed in brackets. For sampling stations, ‘×2’ denotes 2 separate sampling stations
for the given position. Sediment properties show mean values and ranges (min−max) from the reference sites that were sampled during the selected surveys for each farm.
ADCP: Acoustic Doppler Current Profiler
Keeley et al.: Predicting salmon farm deposition and effects
(SEPA) for regulatory modelling of fish farms in Scot-
land (Annex H in SEPA 2005). Feed digestibility (F
dig
)
and water content (F
w
) were set at 85 and 9%,
respectively, which are the DEPOMOD defaults
based on technical data provided by feed manufac-
turers (Cromey et al. 2012) and were used in the
absence of farm and time-specific estimates. All
other model parameters were consistent with exist-
ing salmon farm waste modelling methodologies
(Cromey et al. 2002a,b) and the SEPA Annex H reg-
ulatory farm modelling standards (SEPA 2005), and
remained constant in the tested model scenarios
(Table 2). As the model does not allow the settling
velocity of particles to change through the growing
cycle, the values used for feed and faeces repre-
sented those that would be encountered during the
period of highest waste output from the farm (maxi-
mum standing biomass), which is when the fish are at
pre-harvest size.
Feed input data were based on total feed used per
farm per month and spread evenly across all cages.
In practice, 1 or 2 cages may be empty for short
periods of time as a result of operational require-
ments, but this resolution of spatial and temporal
information was not available and would be imprac-
tical to include in the model. However, this repre-
sents a potential source of variability in the outputs,
which was accounted for by taking the average
result from multiple scenarios. The farm manage-
ment conditions for each scenario (i.e. number of
cages, net depths, overall size and position of
farm and monitoring stations) were determined
from information collected during annual monitor-
ing surveys (e.g. GPS fixes of farm corners), histori-
cal aerial and satellite images, and discussions with
farm operators. The standard farm configurations
involved square cages with a net depth of 20 m
arranged in adjoining clusters, either 1 or 2 cages
wide and 4 to 8 cages long.
Depositional flux was predicted for 110 benthic
sampling locations, representing 18 different histori-
cal farming arrangements, encompassing all 6 study
farms (Farms A to F) over 8 yr (2003 to 2010, Table 1).
Results were obtained for 4 different feed levels
based on the average reported feed use for the 1, 3,
6 and 12 mo immediately prior to the environmental
data being collected. Four critical resuspension ve -
locities were contrasted within each average feed
use period: (1) without resuspension, or with resus-
pension based on critical velocity thresholds (vr) of:
(2) 9.5 cm s
−1
(model default), (3) 12 cm s
−1
and (4)
15 cm s
−1
. Thus, 16 model runs were conducted for
each of the 18 different farming scenarios, giving a
total of 288 runs. Matlab™ code was developed to
enable batch processing of model runs.
Relating predicted flux to observed enrichment stage
Environmental condition was determined using
established ecological indicators: N, S, H ’, AMBI and
BQI in combination with physico-chemical variables
(%OM, redox, TFS). All variables were also unified
following the methods of Keeley et al. (2012a,b) to
obtain an indication of overall enrichment stage (ES),
a bounded continuous variable that places the results
on a scale between ES1 = ‘pristine’ to ES7 = azoic/
anoxic. Generalised additive modelling was then
used to establish the relationship between predicted
flux and observed ecological responses, as shown by
ES and each the individual indicator variables.
Prior to analysis, both predicted flux and ES values
were log-transformed to improve data normality and
reduce heteroscedasticity. The necessity to construct
flow-specific models was checked by testing the
significance of Flow as a fixed factor (high/low) using
linear models in R (R Development Core Team
2011). In all cases, Flow was highly significant (p <
0.0001). The optimum linear model was then identi-
fied by fitting 4 different polynomials (of order 1 to
4) and then selecting the model with the smallest
Akaike Information Criterion (AIC) value. If the AIC
values of 2 models were within 2 units (and could
therefore be considered equivalent, Burnham &
Anderson 2002), then the simplest model was chosen
in preference. The best-fit polynomials were solved
for x (or ES) to obtain estimates of the average flux
associated with ES3 (i.e. ES = 3) and ES5 (i.e. ES = 5),
and the standard errors of the coefficients were used
to calculate the associated 95% pointwise confidence
bounds (hereafter referred to as confidence intervals,
CI). ES3 was selected to represent the outer bound-
279
Input variable Setting
Feed wastage 3%
Water content of feed pellet 9%
Digestibility 85%
Settling velocity of feed pellet 0.095 m s
−1
Settling velocity of faecal pellet 0.032 m s
−1
Random walk model
k
x
, k
y
0.10 m
2
s
−1
k
z
0.001 m
2
s
−1
Table 2. Default model settings that were applied consis-
tently throughout the modelling. k
x
, k
y
and k
z
are horizontal
and vertical dispersion coefficients