R

MbM

ROa

b

c

a

ARRA

KSDBMA

1

t2ssaomla2bfs

h0

Ecological Modelling 361 (2017) 122–134

Contents lists available at ScienceDirect

Ecological Modelling

j ourna l h omepa ge: www.elsev ier .com/ locate /eco lmodel

esearch Paper

ACAROMOD: A tool to model particulate waste dispersion andenthic impact from offshore sea-cage aquaculture in theacaronesian region

odrigo Riera a,∗, Óscar Pérez a, Chris Cromey b, Myriam Rodríguez a, Eva Ramos a,mar Álvarez a, Julián Domínguez a, Óscar Monterroso a, Fernando Tuya c

Centro de Investigaciones Medioambientales del Atlántico (CIMA), SC de Tenerife, Canary Islands, SpainFreelanceIU ECOAQUA, Grupo en Biodiversidad y Conservación, Universidad de Las Palmas de Gran Canaria, Las Palmas de Gran Canaria, Canary Islands, Spain

r t i c l e i n f o

rticle history:eceived 15 May 2017eceived in revised form 4 August 2017ccepted 8 August 2017

eywords:ea-cage aquacultureispersion modelenthic impact

a b s t r a c t

Uneaten feeding pellets and fish released faeces cause the most severe impact on the benthos beneathaquaculture offshore sea-cages. A modelling tool, ‘MACAROMOD’, composed of particulate waste dis-persion and benthic response, was developed to predict the environmental disturbances of offshoresea-bream (Sparus aurata), sea-bass (Dicentrarchus labrax) and meagre (Argyrosomus regius) aquacul-ture in the Macaronesian region (oceanic archipelagos in the north-eastern Atlantic). MACAROMOD wastested at 8 sites (7 farms in the Canary Islands and 1 farm in Madeira), hence covering a high variabilityin oceanographic and environmental conditions. In general, a low percentage of lost pellets was found(3%), while a high rate of pellets were consumed by wild fishes (97%). Considering all studied sites, signif-

2

acaronesian region

tlantic ocean

icant correlations were shown between observed and predicted solid fluxes (R = 0.89), and also betweensolid fluxes and the depositional footprint on the benthos, by taken advantage of observed and predictedvalues of the ecological status AMBI index (R2 = 0.6966). A flux threshold of 12 kg solids m−2 yr−1 waspredicted as a boundary from which ecological degradation occurs for the study region. MACAROMOD istherefore a valid tool to improve planning and monitoring Macaronesian aquaculture.

© 2017 Elsevier B.V. All rights reserved.

. Introduction

Global production of fish from aquaculture has grown substan-ially in the past decades, reaching 97.2 million tonnes in 2013 (FAO,013). Aquaculture is the fastest growing animal food producingector and currently accounts for 33.7% of the global food fish con-umption (FAO, 2013). However, the negative consequences thatquaculture may have on natural systems are an increasing causef environmental concern (Holmer, 2010). Sea-cage aquacultureay change the physical and chemical environmental conditions,

eading to changes in the density, species richness, and overallbundances of benthic organisms (Edgar et al., 2010; Machias et al.,004; Riera et al., 2013). There are many forms of wastes produced

y marine fish cage aquaculture; however, particulate waste, in theorm of uneaten feeding pellets and fish released faeces, is con-idered the primary source of ecological impact on the benthic

∗ Corresponding author.E-mail address: rodrigo@cimacanarias.com (R. Riera).

ttp://dx.doi.org/10.1016/j.ecolmodel.2017.08.006304-3800/© 2017 Elsevier B.V. All rights reserved.

community (Beveridge, 2004; Riera et al., 2011, 2014; Vezzulli et al.,2003). This material, which generally settles on the seabed nearthe cages, may exceed the carrying capacity of the environment(Kalantzi and Karakassis, 2006; Papageorgiou et al., 2010; Telferet al., 2009).

Worldwide efforts are underway to develop more sustainablefarming techniques (Troell et al., 2003), ensuring conservation ofcoastal ecosystems in areas with aquaculture development. Cur-rent technologies, i.e. modelling, provide powerful and reliablemanagement tools to integrate aquaculture into a broader con-text of integrated coastal zone management (ICZM). During thelast two decades, sea-cage aquaculture particulate waste mod-elling has rapidly developed from several research-based models(Gowen et al., 1989; Findlay and Watling, 1994; Hevia et al., 1996;Pérez et al., 2002) to well-established and professionally-used tools(Cromey et al., 2002a; Stigebrandt et al., 2004). Nowadays, these

models are cost-effective tools, which are been widely used to assistin the prediction of impacts (e.g. SEPA, 2005). Regulatory authori-ties are increasingly turning to predictive models to make informeddecisions when licensing new marine fish farms and granting con-

Model

se2(oohwi

pmTarcSmffttficoiiMteGr

dio(wcedwgtbsa(caaermd

2

2

sAsS

R. Riera et al. / Ecological

ents to discharge waste (Chamberlain and Stucchi, 2007; Dudleyt al., 2000; Henderson et al., 2001; Read et al., 2001; Dapueto et al.,015). However, these tools have not yet been widely used becausei) a calibration and a posterior validation is necessary dependingn different variables, such as local-scale oceanographic conditionsr culture characteristics, and (ii) a minor fraction of these modelsave successfully linked solids accumulation with benthic impacts,hich in turn provides an estimation of possible environmental

mpacts derived from fish farm activity.Among the aquaculture particulate waste dispersal models cou-

led with benthic impact estimation, DEPOMOD (Depositionalodelling) (Cromey et al., 2002a) is renowned for its extensive use.

his model is well balanced in terms of interface, data requirementsnd performance, and is routinely used by the SEPA (Scottish Envi-onment Protection Agency), as a modelling tool to set dischargeonsents based on maximum farmed biomass for salmon farms incotland. This model has been lately adapted for Atlantic cod (Gadusorhua L.) farming in N Atlantic Ocean (Cromey et al., 2009) and

or sea-bream (Sparus aurata) and sea-bass (Dicentrarchus labrax)arming in the E Mediterranean Sea (Cromey et al., 2012). However,hese models have been built for specific geographical areas, andhus have been parameterized with data to suit environments andsh species from particular regions; these tools are of limited appli-ability outside their boundaries (Pérez et al., 2014). This is the casef the Macaronesian region, which comprises several archipelagos

n the NE Atlantic Ocean off the coast of Europe and Africa, includ-ng the Azores, Madeira, the Canary Islands and Cape Verde (Fig. 1).

aritime and environmental characteristics remain similar amonghe archipelagos, volcanic origin (Longhurst, 2007) and highly influ-nced by the Azores Current. This current flows eastward from theulf Current to supply the eastern Atlantic subtropical boundary

egion.The aim of the present study was to develop and vali-

ate a model capable of predicting the dispersion and benthicmpact of particulate waste from offshore sea-cage aquaculturef sea-bass, sea-bream and meagre in the Macaronesian regionhenceforth MACAROMOD). It is well recognized that algorithms,hich describe the advection, dispersion and accretion of parti-

les in most deposition models are valid across a wide range ofnvironments, if the model boundary conditions are adequatelyescribed (Keeley et al., 2013). Thus, it is possible to transfer theell-established and tested depositional model DEPOMOD to other

eographical regions with minor alterations. On the contrary, rela-ionships between depositional fluxes and ecological responses cane strongly influenced by physical environmental properties, i.e.and vs mud, flow regimes, local macrofaunal assemblages, whichre site-specific (Keeley et al., 2013). Thus, it was necessary toi) obtain data related to growing (species-specific food and fae-es settling velocities, percentage of uneaten food, among others)nd environmental conditions (e.g. currents, dispersion coefficientsnd food consumption by wild fishes) needed for model param-terization, (ii) derive a benthic response using a semi-empiricalelationship between modelled fluxes and a benthic index, and (iii)

odel validation, so MACAROMOD may be used as a tool to help inecisions for planning and monitoring Macaronesian aquaculture.

. Material and methods

.1. Study sites

This study was conducted around eight sea-cage fish farms,

even at the Canary Islands and one in Madeira (Table 1). Thezores and Cape Verde archipelagos were not included, becauseea-cage aquaculture was not developed at the time of this study.ite notation is given as: A B c, where A is the archipelago name

ling 361 (2017) 122–134 123

(C = Canaries and M = Madeira), B is the island name (GC = GranCanaria, M = Madeira and T = Tenerife) and c is the site referencenumber (1–8). This anonymous site reference system is used, asthe data were collected under a confidentiality agreement. Foursites were used to derive a benthic response relationship undervarying particulate waste flux rates (kg m−2 year−1) and four siteswere used for model validation (particle tracking and benthicresponse). Data used to derive benthic response relationships (sitesC T 1, C T 2, C T 3 and C T 4) was obtained from annual compilationmonitoring programs during 5 years (2006–2010). Data for modelvalidation was obtained from sampling field surveys during 2012(sites C T 5, C T 6 and C GC 7) and 2013 (M M 8).

Coastal environmental conditions greatly varied among thestudied aquaculture facilities, encompassing most of the coastalecosystems in the Macaronesian region. Thus, this implies thatthe model output is realistic, considering the high coastal andoceanographic variability in the study area. Continuous currentsthroughout the entire year are present at all studied aquaculturefacilities, ranging from <5 cm s−1 on the surface and >20 cm s−1 onthe bottom layer (Riera et al., 2015). The benthic habitats are mainlycomprised by sandy unvegetated seabeds, with sparse Cymodoceanodosa and Caulerpa prolifera meadows on the surroundings of theaquaculture facilities (>500 m from cages). Sediments were mainlycomposed by fine and medium-sized sands, with scarce content ofvery fine-grained sediments, e.g. silt/clay. No subtidal rocky sub-strate is present beneath the farms.

2.2. Model structure description

The model general structure is that initially developed for DEPO-MOD (Cromey et al., 2002a) and further adapted to Mediterraneanconditions for MERAMOD (Cromey et al., 2012). The model isassembled by 4 modules: (i) grid generation, (ii) particle track-ing, (iii) re-suspension, and (iv) benthic faunal response (benthicimpact) (Fig. 2). A grid containing information on depth, sea cagesand sampling stations positions for the area of interest is initiallyrequired. In contrast to the original model (DEPOMOD) and laterversions, MACAROMOD offers a larger grid domain of 999 × 999cells. This is useful for covering extensive areas and, consequently,additive effects from adjacent farms may be quantified withoutdecreasing model resolution. Given wastage rates of fish food andfaeces from the bioenergetics model, hydrodynamic data, indi-vidual settling velocity for wasted food and faeces for differentculture species and wild fish waste consumption, initial depositionof particles on the seabed are predicted with the particle-trackingcomponent. An important innovation in this module, based on theresults of this study, is the possibility to determine cage parti-cles starting position, both in the horizontal and vertical axis. Are-suspension model (Cromey et al., 2002b) then redistributes par-ticles according to near-bed current flow fields to predict net solidsaccumulated. Finally, an impact assessment is provided by cor-relating model predictions (e.g. solids fluxes as g m−2 y−1) withthe benthic faunal indicator of environmental impact AMBI (AZTI’sMarine Biotic Index) (Borja et al., 2000).

Sensitivity analysis is an important procedure for modeldevelopment, used to increase confidence in the model and its pre-dictions, by studying how the variation in the output of a model canbe apportioned to different sources of variation, and how the givenmodel depends upon the information. In this study, model sensi-tivity analysis was not performed because underlying equations

and algorithms are equal to those already tested for DEPOMOD andMERAMOD, hence, its sensitivity to parameters changes is alreadyknown and taken into consideration in the model parameterization,adaptation and validation procedures.

124 R. Riera et al. / Ecological Modelling 361 (2017) 122–134

Fig. 1. Macaronesian region showing all archipelagos.

Table 1Characteristics of the fish farms for the model parametrization and validation.

Use Farm Farmed species Total farmbiomass (t)b

Water depth (m) Current speed(bottom, mid andsurface (cm s−1)

Grain size Organic Matter(%)

Benthicresponserelationship

C T 1 Sea bass 375 25–30 Mean: 7.9/9.8/34.8 Very Fine sand 0.70Sea bream

C T 2 Sea bass 447 25–30 Max: 21.0/29.3/87.4 Very Fine sand 0.75Sea bream

C T 3 Sea bass 280 25–30 Mean: 6.3/6.9/13.5 Very course sand 0.41Sea bream

C T 4 Sea bass 357 25–30 Max: 25.2/32.4/63.7 Fine sand 0.57Sea bream

Model parametrizationand validation

C T 5 Sea bass 572 30–35 Mean: 6.4/7.1/11.8 Fine sand 0.96Sea bream Max: 21.4/21.1/45.6Meagre

C T 6 Sea bream 326 25–30 Mean: 18.4/21.4/34.0 Course sand 0.93Max: 57.1/63.4/119.9

C GC 7 Sea bass 525 25–30 Mean: 10.2/11.0/16.4 Fine sand 0.57Sea bream Max: 39.4/46.9/66.7

M M 8a Sea bream 240 20 Mean: 5.9/6.1/5.8 Fine sand 2.69Max: 41.6/52.3/48.3

a Farm M M 8 was also used for model benthic response relationship.b Biomass refers to the average number of tonnes throughout the study period that remains constant, except in C T 5, C T 5 and C T 5 (data from 2012) and M M 8 (data

from 2013).

R. Riera et al. / Ecological Modelling 361 (2017) 122–134 125

iagram

2

flBw5abp

s(cf

92bsrb

Fig. 2. Conceptual d

.3. General model set-up

Bathymetry for sites located in the Canary Islands was obtainedrom the existing governmental database (digitalized 1 m iso-ines). For Madeira, data was obtained from vector charts (GarminlueChart Atlantic 9.5). For each site, a Triangulated Irregular Net-ork (TIN) was created, later converted to an ASCII file with a

× 5 m cell size and finally exported to MACAROMOD to create bathymetry grid (5 m cell resolution). This resolution was used,ecause it provides maximum spatial detail without losing modelrediction capabilities (Cromey et al., 2012).

Sensitivity analysis performed in DEPOMOD and MERAMODhowed that sedimentation is influenced by surrounding cagesCromey et al., 2012), as well as the cage above; thus, individualage dimension, depth and positioning (coordinates) were obtainedor each cage within a site.

Feeding pellet digestibility and water content were set at 85 and% (see Magill et al., 2006 for sea bass and sea bream; Pérez et al.,014 for meagre), respectively, based on technical data providedy manufacturers and were used in the absence of farm and time-

pecific estimates. For sites used to derive the benthic responseelationship, averaged feed input data were used for the year beforeenthic sampling started. Weekly averaged data were obtained for

of MACAROMOD.

each farm via farmers questionnaires; occasionally, only one cagedata was obtained. Used percentages of uneaten feeding pellets andpellets consumed by wild fish were those calculated in the presentstudy. For sites used for model validation, real-time feed input datafor each cage within a farm (e.g. number of rations, kg feed, pel-lets size a brand, empty cages were accounted for where thesewere recorded), were obtained from farmers at the time of eachfield survey. Used feeding pellets percentages and uneaten pelletsconsumed by wild fish were those calculated for each site in thisstudy.

Settling velocity of uneaten feed pellets and faecal materialare key parameters for the model prediction accuracy. Feedingpellets and faeces settling velocities for sea-bream and sea-basswere those estimated by Magill et al. (2006). Feed settling datawere specified depending on the pellet diameter and brand, andwhen this information was not available, a generalised relation-ship for feed settling velocity was used (8.5 ± 4.7 cm s−1). Faecalsettling velocities for sea-bream were 0.4, 1.5, 2.5 and 3.0 cm s−1

with mass percentages of 24, 45, 18 and 13%, respectively, andfor sea-bass were 0.4, 1.4, 2.5, 3.6 and 4.6 cm s−1 with mass per-

centages of 6, 9, 20, 38 and 27%, respectively. For meagre, valueswere those presented in Pérez et al. (2014). Used feed settlingdata for 9 mm and 12 mm pellets were 9.83 ± 0.17 cm s−1 and

1 Model

9du13

atdwtpc

DIdttracd(ttc

yatt(f

Ar

2

2e

tehutwtirtrTeiplaptt

S

26 R. Riera et al. / Ecological

.67 ± 0.28 cm s−1, respectively. When no information on pelletsiameter was available, a mean value of 9.75 ± 0.24 cm s−1 wassed. Faecal settling velocities for meagre were 0.25, 0.70, 1.21,.69, 2.24 and 2.54 cm s−1 with mass percentages of 32.2, 5.3, 14.9,5.9, 7.4 and 4.3%, respectively.

The starting position of pellets and faeces, both in the horizontalnd in the vertical axis, were different. For pellets, horizontal dis-ribution was homogeneous among the whole cage except from aistance of 2 m at the cage border. The vertical starting distributionas from the surface. For faeces, the horizontal and vertical dis-

ributions were random within the cages. In all simulations, totalarticle numbers were optimised (>5 × 105) and trajectories recal-ulated every 6 s.

Current data was obtained by deployment of an Acousticoppler Current Profiler (ADCP, model FlowQuest 600, LinkQuest

nc.), close to the fish farm and for a period ≥ 30 days. Speed andirection readings were taken for the entire water column (2 mhick layers) with a 30 and 15 min intervals for sites used to derivehe benthic response relationship and sites for model validation,espectively. For each interval and layer, 100 readings were madend the mean values recorded. From all layers in which the waterolumn was divided, the model used five: (i) surface layer; (ii) mid-le net layer; (iii) button net layer; (iv) mid-water column layer andv) bottom layer. Dispersion coefficients for sites used to derivehe benthic response relationship and for model validation werehe mean values and particulate values for each site, respectively,alculated in this study.

To account for vertical cage movement due to tide changes, aearly mean tidal height value was used for sites intended to derive

benthic response relationship and for the benthic module valida-ion. For sites used for particle tracking module validation, a meanidal height value was calculated for the period that sediment trapssee below) were deployed (7–9 days). Tidal values were obtainedrom the nearest port database.

Model predictions (irregularly spaced data) were exported torcGIS 10.2 (Esri Inc.) for contouring via kriging interpolation algo-ithm (linear variogram) and for map representation.

.4. Model parameterization and adaptation

.4.1. Uneaten feeding pellets and fish faeces lost to thenvironment

Fish cage particulate waste dispersion models are sensitiveo the input of uneaten feeding pellets, fish faeces lost to thenvironment and their starting position within the cage, bothorizontally and vertically. Reported amounts (percentages) ofneaten pellets in marine fish cage farming are normally estima-ions based on direct observations. These estimations include aide range of values mainly according to cultured species (salmon,

rout, sea-bass, sea-bream, among others), regions/countries, feed-ng regimes (number of rations per day, at libitum or predefinedations, daily feeding or only during working days, etc.), feedingechnique (manually, semiautomatic or automatic), marine envi-onment peculiarities (dispersive o non-dispersive), among others.herefore, reported amounts in the literature cannot be acknowl-dged as adequate or not (when compared with others calculatedn different conditions), but as suitable or unsuitable for a particularlace and farming conditions. On the contrary, estimations of faeces

ost to the environment are normally calculated using a mass bal-nce, based on empirical data on digestibility mainly for differentellet types and cultured fish species (Brigolin et al., 2010). Hence,

hese values are unlikely to change from place to place, providedhe pellet type and species are similar.

The starting position of particles varies from model to model.implistic models use a single point source, located at the cage cen-

ling 361 (2017) 122–134

tre either at the surface (Gowen et al., 1994) or at the bottom (Pérezet al., 2002). This is also the case when using general hydrodynamicmodels coupled with a Lagrangian particle module (i.e. MOHID orMIKE), which are not specifically developed for fish cage aquacul-ture and, hence, their correct parameterization is normally limitedto a single point source (Pérez et al., 2014). Most elaborated aqua-culture waste dispersion models include a random initial particledistribution in the cage (Cromey et al., 2002a, 2009, 2012).

Three groups of n = 2 PVC sediment traps were used to incor-porate more realistic data on the amounts of wasted material andparticles starting position, which in turn will greatly improve MAC-AROMOD performance. These sediments traps were built usingthe well-accepted criterion of length: diameter ratio of at least5:1 (Wassmann et al., 1991) and internal diameter of 15.3 cm.They were deployed directly outside a cage net per farm (avoid-ing wild fish consumption), following the main current direction.Traps were located at three net depths: top, middle and bottomof the net, respectively. Initially, all sediment traps were deployedclosed (<25 m apart). The experiment started when the 3 traps perfarm were opened, which collected all material (uneaten pelletsand faeces) during 24 h, corresponding to a daily feeding cycle.These data were used to determine the total particulate materiallost to the environment, despite it was not possible to differen-tiate between food and faeces. To determine the ratio betweenlost food and fish faeces as contributors to particulate material, 3sediment traps were opened only when feeding was concluded,therefore exclusively collecting faecal material. Mass differencesbetween each set of traps provided an indication of each amount(percentage).

Traps were retrieved and the contents settled after 15–20 min;then excess of water was siphoned off with care to not disturb thecontents. The contents of sediment traps were analysed for totaldry solids (TDS). Contents were filtered through coarse glass fibrefilters with a vacuum apparatus (Whatman GF/A, pore size 1.6 �m,� = 15 cm) and then dried at 60 ◦C overnight. Samples were cooldown in a desiccator before weighting (accuracy: 0.001 g). Posteri-orly, collected material was scaled for each sediment trap accordingto net size and current rose directions and frequencies, to be ableto extrapolate traps data to a cage scale. In addition, precise data onfeed inputs (kg food and feeding time) and stocking biomass wasnecessary. Results are expressed as percentages of uneaten foodand faeces lost to the environment. In Madeira, this experimentwas not carried out due to logistic problems. Calculated percent-ages of uneaten pellets lost to the environment were used for modelvalidation and to provide a standardised value for users in casesite-specific data were not available. In addition, mass differencesbetween surface, mid and bottom net sediment traps, were usedto determine pellets and faeces starting positions included in themodel.

2.4.2. Uneaten pellets lost to the environment and consumed bywild fishes

Model performance is sensitive to percentages of uneaten foodconsumed by wild fish populations around sea-cages, because wildfish might reduce undesired effects on benthos resulting fromdeposition of particulate waste, as they are able to consume sig-nificant amounts of the waste feed before sedimentation occurs(Uglem et al., 2014). The magnitude to which waste food pelletsderived from a farm are consumed by wild fish largely depend onthe biomass of wild fish around cages and the species composi-tion of the assemblages (Dempster et al., 2005). To this extent,MACAROMOD parameterization requires of knowledge concerning

utilization of waste feed by wild fish in the Macaronesian region,to increase model performance.

Benthic studies in the Macaronesian region, in particular inthe Canary Islands, have indicated that cage aquaculture impacts

Model

afitpttpaswc

sfdf3wtb

mpwc

wirctfcbd1tfsfi

2

tMetsasdcvd

mbbftGTkdte

R. Riera et al. / Ecological

re negligible, i.e. no food pellets have been found during recenteld surveys (Riera et al., 2013, 2014, 2015). By direct observa-

ions, it is well known that pelagic fish actively feed over wastedellets, removing most material. Yet, the amount of remaining set-led material and its fate are unknown. As for modelling purposes,here is no need to differentiate between wasted food consumed byelagic or benthic fish, only the total amount consumed is required,nd thus, only benthic trials were undertaken. These experimentsolely focused on the quantification of wasted pellets removed byild fish, nor in the community structure of fish aggregates around

age farms.To find out whether pellets not consumed by wild fishes while

inking are consumed by benthivorous fishes, a simple ground-eeding experiment was carried out. After a regular feeding event,ivers placed 100 feed pellets (same size and brand as those to

eed fish in the corresponding cage and farm) inside a frame of0 × 20 cm on the seabed. At each aquaculture facility, two cagesere selected to run this experiment and six plots were established,

wo beneath the cage, two halfway between the centre and the cageorder, and two at the border.

The experiments were filmed by remote-operated cameras toinimize interference with fish. After one hour, the number of

ellets inside the frames was counted by divers. If all pelletsere eaten, the experiment was consider concluded, otherwise the

ounting was repeated till a period of 3 h.In order to qualitatively identify fish assemblages around cages,

hich in turn will help results interpretation, counts of wild fishesn the water column were conducted by scuba divers. Five-minuteapid visual counts (Kingsford and Battershill, 1998) (n = 6) werearried out at the centre of each aquaculture facility, at the edge ofhe aquaculture lease (“influence” stations) and sites far from fisharms (>500 m) (“control” farms). Each count covered a volume ofa. 11,250 m3 (15 m wide × 15 m deep × 50 m long), and was madey two divers. The first diver concentrated on estimating the abun-ance of the dominant species, which were counted in groups of, 2–5, 6–10, 11–30, 31–50, 51–100, 101–200, 201–500 and >500o minimize error (Harmelin-Vivien et al., 1985). The second diverollowed slightly behind the first and specifically looked for crypticpecies and smaller individuals that may have been missed by therst diver. Counts of both divers were summed.

.4.3. Dispersion coefficientsMarine hydrodynamics at the Macaronesian region differ rela-

ive to Scotland and the Mediterranean Sea, where DEPOMOD andERAMOD models were developed, respectively. The scarce pres-

nce of bays and naturally sheltered sites, competition with theourism industry for the same coastal space, in conjunction withtrong local currents (mean minimum and maximum values of 5nd 35 cm s−1, unpublished data from 57 deployments) and con-tant winds, is forcing aquaculture industry to exposed and veryynamic sites in the Macaronesian region. Therefore, dispersionoefficients were calculated for model validation and also to pro-ide standardised values for MACAROMOD users if site-specificata are not available.

Dispersion coefficients needed for the turbulence Random Walkodel were determined using a similar method as that presented

y Cromey et al. (2012) and Pérez et al. (2014). A group of six num-ered drifting buoys was deployed close to each aquaculture facilityor a period of 12 h (i.e. a neap-spring cycle). Buoys position wereaken every 15 min, by an operator in a small boat using a handheldPS receiver with EGNOS capabilities to provide accuracy < 1.5 m.he horizontal components of the dispersion coefficients (kx and

y) were calculated according to Yanagi et al. (1982). The verticalispersion coefficient (kz) was not measured and set, by default,o a standard value of 0.001 m−2s−1 (Cromey et al., 2009). Cromeyt al. (2002a) found that DEPOMOD was insensitive to kz, as the set-

ling 361 (2017) 122–134 127

tling velocity of the waste particles is high in relation to the verticalsteps calculated by the random walk model.

2.4.4. Quantification of benthic impactDifferent models incorporate a variety of environmental indexes

(i.e. N, S, Hı́ or BQI), according to particular culturing and envi-ronmental conditions, as these relations are typically site-specific(Keeley et al., 2013). Therefore, to couple MACAROMODı́s partic-ulate waste dispersal module with the benthic impact module, asemi-empirical relationship was derived to relate model flux pre-dictions (kg solids m−2 year−1) with observed ecological status.In this study, the tolerance of the benthos to organic enrichmentwas evaluated using the AZTI’s Marine Biotic Index (AMBI) (Borjaet al., 2000). AMBI classifies benthic species (infauna) into ecolog-ical groups based on their sensitivity/tolerance to pollution. AMBIquantitatively varies from 0 to 7: scores <1.2 indicate undisturbedconditions, 1.2–3.3 slightly disturbed conditions, 3.3–5.0 moder-ately disturbed conditions, 5.0-6.0 heavily disturbed conditions,and >6 extremely disturbed conditions. The AMBI index offersa site-specific disturbance or pollution classification (Grall andGlémarec, 1997). This index has been used for the determination ofthe ecological quality status (EcoQ) within the context of the Euro-pean Water Framework Directive (WFD) (Borja et al., 2003, 2004).EcoQ is established on the basis of physico-chemical and biologi-cal variables (Borja et al., 2004). In the benthic EcoQ determination,three parameters are proposed by the WFD: diversity, species abun-dance and the presence/absence of indicator and stress-sensitivespecies. The latter is represented by the AMBI (Borja et al., 2003).

For all sites (C T 1, C T 2, C T 3 C T 4), 11 sampling stations wereavailable with reliable AMBI data during the 4 years annual com-pliance monitoring programs, which in total provided 44 records.Infauna data used to derive this index was collected using core sam-ples. Cores (20 cm inner diameter) were pushed into the sedimentto a depth of 20 cm; this size is within the standard range used tostudy macrofaunal assemblages (e.g. Heilskov et al., 2006; Gilletet al., 2007). Three replicates per sampling station were randomlycollected for faunistic determinations. Faunal samples were pre-served in a 10% seawater formaldehyde solution, and subsequentlydecanted through a 0.5 mm mesh sieve. The fraction remaining wasseparated into different taxonomic groups under a binocular micro-scope, and preserved in 70% ethanol. Macrofaunal specimens werethen determined to species level, whenever possible, by meansof a binocular microscope, or a Leica DMLB microscope equippedwith Nomarski interference. General modelling procedures wereutilized for flux predictions at each site. Calculated solids accumula-tion (kg m−2 yr−1) was obtained at each individual sampling stationand for each year.

Semi-empirical relationships between the average bio-deposition rates and the observed benthic community descriptorswere derived by the adjustment of regression models (lineal,exponential, logarithm or polynomial). A selection of the best fit(regression line) of seven regression models was conducted andthe fourth-degree polynomial regression was selected, consideringthe higher-adjusted and predicted R-squared values, as well asthe p-values for the predictors. Curve fitting was used by theleast-square criterion, choosing the best coefficients to minimizethe sum of square residuals from data.

2.5. Model validation

The procedure of model validation, both for particle trackingand benthic responses, followed that proposed by Mesple et al.

(1996) and Portilla and Tett (2007). The methodology is based onthe comparison between observed and simulated data, using a sim-ple linear regression. According to these authors, model predictionsare placed on the x-axis and observed values on the y-axis. The line

128 R. Riera et al. / Ecological Modelling 361 (2017) 122–134

Table 2Mean percentages of particulate solids from sediment traps.

Sediment trap Individually% Weighted%

Faeces Food Faeces Food

Top 61.5 38.5 25.8 38.5Mid-water 64.5 35.5 35.5 46.2

obc

fio1tcieiI

rMivbawgcntdtwqttap

3

3

3e

wswftrsbenapmb

Table 3Abundances of wild fishes aggregations around fish farms.

Species Abundance (n◦) Abundance (%)

Boops boops 33,510 93.2Sparus aurata 1338 3.7Scomber colias 700 1.9Atherina presbyter 100 0.3Canthigaster capistratus 43 0.1Bothus podasmaderensis 37 0.1Sphyraena viridensis 33 0.1Sphoeroides marmoratus 28 0.1Pseudocaranx dentex 22 0.1Spondyliosoma cantharus 21 0.1Dasyatis pastinaca 17 0.0Taeniura grabata 17 0.0Myliobatis aquila 15 0.0Trachinus draco 14 0.0Abudefduf luridus 12 0.0Sparisoma cretense 11 0.0Xyrichtys novacula 9 0.0Mullus surmuletus 7 0.0Pagellus erythrinus 7 0.0Seriola dumerilii 6 0.0Serranus atricauda 5 0.0Squatina squatina 5 0.0Thalassoma pavo 5 0.0Gymnura altavela 4 0.0Pagrus pagrus 4 0.0

Bottom 87.5 12.5 38.7 15.4Mean 71.2 28.8

f best fit was found by applying a major axis regression (MAR),ecause it has been proved to be a more robust choice than the mostommon and widely used ordinary least squares method (OLS).

Models were assessed with the linear regression of Yi = �0 + �1Xior observed (Yi) and predicted (Xi) values, where �0 and �1 are thentercept and the slope, respectively. A studentı́s t-test was carriedut for R2, �0 and �1 to test for significant differences from 1, 0 and, respectively (Jusup et al., 2009). Models were grouped followinghe classification of Tett (2007) and Jusup et al. (2009), with fourategories: (i) Very good; (ii) Good; (iii) Fair; and (iv) Poor. Concern-ng the benthic response model, curves were fitted minimizing forach comparison a measure of goodness of fit: �(O−E)2/E where Os the indicator observed value, and E the indicator modelled value.n addition, R2 was also maximized (Cromey et al., 2012).

Data used for plotting predicted particle tracking and benthicesponse values (x-axis) was obtained from the model outputs.

odelling was undertaken with detailed information for each site,.e. cage properties, farmed species, feed inputs and faecal settlingelocities. Data used for plotting observed particle tracking andenthic response values (y-axis) was collected using sediment trapsnd core samples, respectively. The location of sampling stationsas identically for both, positioned at 0, 20, 40 and 60 m in a lon-

itudinal and 0, 20 and 40 m in a perpendicular transects. Threeontrol stations were located at >500 m away, following a longitudi-al transect, with similar depth and grain size composition relativeo fish farms. Within each farm, stations at 0 m were establishedirectly beneath the cage placed in the farm perimeter closesto the shore. Before sediment traps deployment, benthic samplesere collected using the same methodology formerly described for

uantification of benthic impact through the AMBI index. Sedimentraps, deployed for a period of 7-9 days, were used to determineotal dry solids flux (g m−2). Traps construction and total dry solidnalysis methodology is that previously explained for determiningercentages of wasted material.

. Results

.1. Model parameterization and adaptation

.1.1. Uneaten feeding pellets and fish faeces lost to thenvironment

No significant differences in the amount of particulate solidsere found between traps placed at the same depth at different

ites (Kruskal-Wallis test, H = 3.67, p = 0.12); therefore, mean valuesere calculated and used to derive percentages of pellets and lost

aeces (Table 2). From all particulate solids captured by sedimentraps, mean percentages of food and faeces were ca. 29 and 71%,espectively. Faecal material escaping from the net top was slightlymaller (ca. 26%) than those from the middle of the net and the netottom, ca. 35 and 39%, respectively; these last differences are at thedge of significance (Kruskal-Wallis test, H = 6.45, p = 0.052). Sig-ificant differences were then only found between the net top trap

nd the net middle and bottom traps (Kruskal-Wallis test, H = 7.89,

= 0.034). Uneaten food percentages measured in the net top andiddle, ca. 35 and 39%, respectively, were significantly different to

ottom values (ca. 15%) (Kruskal-Wallis test, H = 8.81, p = 0.029).

Dasyatis centroura 3 0.0Synodus saurus 1 0.0TOTAL 35,974 100

These results denote a starting position vertical pattern, bothfor faecal particles and uneaten pellets, which were incorporatedinto MACAROMOD as a subroutine (Fig. 3). The starting positionof particles in the vertical axe included five options, although the“random between two predetermined depths” option was used forall models in this study, and is therefore the recommended choice.For faeces, this distance was set from the net middle to the bottom,while for uneaten pellets it was adjusted from the net surface tothe middle. The model also incorporates three options for particlesstarting position in the horizontal axe. Based on the feeding tech-nique used in the survey sites, hand feeding avoiding spread of foodnear the cage borders, the “random within a distance from the cageborder” (2 m) option was used for all modelling scenarios.

Based on the amount of uneaten food material collected in traps,provided food (kg of pellets), estimated existing biomass and cur-rent data measured during the 24 h experimental time, estimatedmean values of percentage of wasted food was 3.5% (3.5%, 3.2% and3.8% for sites C T 5, C T 6 and C GC 7, respectively). For site M M 8(farm in Madeira), this value was set to 5.5% based on providedhusbandry data and seabed visual inspection.

3.1.2. Uneaten consumed by wild fishesSea-cage fish farms attracted a variety of fish species (Table 3). A

total of 35,974 fish were counted at all farms (8 sites x 5 replicates),belonging to 27 species. The bogue (Boops boops) was the mostabundant species (33,510 fish), accounting for ca. 93% of the over-all abundance. Escaped sea-bream (Sparus aurata) was the secondmost abundant species (ca. 4% of the overall abundance) (Table 3).Fish aggregations are concentrated around cages, being scarce atsites not directly affected by the presence of farms, i.e. “influence”and “control” stations.

In the Canaries (7 aquaculture farms), most of the lost food pel-lets (95–100%) were consumed by fish aggregations. In Madeira(1 aquaculture facility), however, the percentage of food pellets

consumed by wild fish was 75%, since fish aggregations werenot abundant here. In turn, significant differences were foundbetween the Canarian and the Madeiran lease (Mann-Whitneytest, U = 12.9, p = 0.023). Despite this variability, the mean value of

R. Riera et al. / Ecological Modelling 361 (2017) 122–134 129

Fig. 3. Options of particles staring position included in the model. Vertical axe: a) top, b) middle, c) bottom, d) random and e) random between two predetermined depths.Horizontal axe: f) centre, g) random and h) random within a predetermined distance from the cage border.

Table 4Statistical tests using the regression equation for observed and predicted flux values for the particle tracking study at each fish cage and considering all cages (Global). Boldletters denote significant differences.

Cage R2 Sxy F B1 (slope) Class

F p t p

Los Cristianos 0.89 27.36 33.70 0.004 5.81 0.004 Very goodCaletillas 0.90 4.65 35.35 0.004 5.95 0.004 Very goodMelenara 0.85 14.53 23.27

Ensenada Da Abra 0.58 6.73 5.63

Global 0.89 20.89 170.75

F

piptpfsi(

3

0ddk

ig. 4. Relationship between predicted depositional fluxes and the AMBI index.

ellet consumption from the Canary Islands (97%) was includedn the MACAROMOD modelling tool. No significant differences inellet consumption were found with regard to plot position (cen-re, middle and border of the net) (Kruskal-Wallis test, H = 2.45,

= 0.12). Even within each farm, no consistent differences wereound between farms (Kruskal-Wallis test, H = 1.12, p = 0.33). Nopatial variability was observed among studied aquaculture facil-ties, considering the plot positions (centre, middle and border)Kruskal-Wallis test, H = 0.97, p = 0.37).

.1.3. Dispersion coefficientsMinimum and maximum values for kx and ky were

2 −1 2 −1

.041–0.917 m s and 0.178–0.892 m s , respectively. Meanispersion coefficients values, recommended when site-specificata are not available, were: kx = 0.476 ± 0.379 m2 s−1 andy = 0.524 ± 0.290 m2 s−1.

0.008 4.82 0.008 Very good0.077 2.37 0.077 Good<0.001 13.07 <0.001 Very good

3.1.4. Quantification of benthic impactThe relationship between the predicted depo-

sitional flux and the AMBI index was reasonablywell established through a fourth-degree polynomialregression (R2 = 0.6966). The regression equation was:AMBI = −0.0015x4 + 0.0567x3 −0.6511 ×2 + 2.4565x −0.607 (Fig. 4).AMBI showed highly significant differences depending on the flux(One-way ANOVA, F = 24.16, p < 0.0001). Moderate flux (2–3 kgm2 year−1) inputs provided the best ecological status (Fig. 4);however, this correlation was negative under high flux values(ca. >12 kg m2 year−1) occur in the environment, with an evidentimpoverishment of the ecological status.

3.2. Model validation

3.2.1. Particle trackingComparisons between observed and predicted fluxes showed

high correlation values (R2 > 0.85) at all fish cages, with the excep-tion of M M 8 (R2 = 0.58). When considering all cages, the modelperformed even better (R2 = 0.89) (Fig. 5), as confirmed by the sta-tistical test (one-way ANOVA, F = 170.75, p < 0.001). Thus, modelpredictive capacity was classified as “Very good” (Table 4).

3.2.2. Benthic impactObserved and predicted AMBI values were highly correlated

(R2 > 0.70) at all fish cages, ranging from 0.70 to 0.75. When con-

sidering all cages as a whole, the correlation value increased to0.74 (Fig. 6), showing significant differences (One-way ANOVA,F = 12.11, p = 0.013) (model predictive capacity classed as “Verygood”) (Table 5).

130 R. Riera et al. / Ecological Modelling 361 (2017) 122–134

Fig. 5. Observed and predicted fluxes (g m−2) for sites (C T 5, C T 6, C GC 7 and M M 8) and pooled sites.

Table 5Statistical tests using the regression equation for observed and predicted AMBI values for the benthic impact study at each fish cage and considering all cages (Global). Boldletters denote significant differences.

Cage R2 Sxy F B1 (slope) Class

F p t p

Los Cristianos 0.75 9.87 12.35 0.012 4.05 0.007 Very goodCaletillas 0.70 8.88 10.21 0.018 3.89 0.012 Very goodMelenara 0.74 12.13 12.10 0.013 4.01 0.009 Very goodEnsenada Da Abra 0.75 10.01 12.33 0.012 4.07 0.006 Very goodGlobal 0.74 9.52 12.11 0.013 4.03 0.008 Very good

R. Riera et al. / Ecological Modelling 361 (2017) 122–134 131

(C T 5

4

4e

tu

Fig. 6. Observed and predicted AMBI for sites

. Discussion

.1. Uneaten feeding pellets and fish faeces lost to thenvironment

The amount of faeces lost to the environment was about 2.5imes greater than uneaten food, being therefore the main partic-late solids input in cage farming at the Macaronesian region. As a

, C T 6, C GC 7 and M M 8) and pooled sites.

result, faeces seem to be the main contributor to possible environ-mental perturbations on the benthos beneath aquaculture facilities.Nevertheless, this is not necessary true because, although faecalmaterial flux is much higher, settling velocities are lower relativeto unconsumed pellets, what increase dispersion. On the contrary,

uneaten pellets settling velocities are higher, leading to a fastersedimentation in a smaller area, which in turn may result in a moresevere benthic affection.

1 Model

v(aaoCs(tmf2vmgrCcc(

4w

scTfi‘Afireokitewtpdttfbcfi

c7obof

4

iHveo

32 R. Riera et al. / Ecological

Food wastage rates are difficult to quantify, due to a largeariation in husbandry practices and hydrodynamic conditionsdispersive or non-dispersive) between sites/locations/regions. Inddition, technical improvements in physical properties of pelletsnd in feeding techniques progressively have reduced their impactsn the benthos. Gowen et al. (1989) assumed a 20% loss, whileromey et al. (2002a,b) reported values of 12% from sediment traptudies in the UK. Later studies lowered these percentages to 10%Pérez et al., 2002), 5% (Cromey et al., 2009, 2012) and more recentlyo 3% (Keeley et al., 2013). A 3% is also the level currently recom-

ended by the Scottish Environmental Protection Agency (SEPA)or regulatory modelling of fish farms in Scotland (Annex H in SEPA005). Cairney and Morrisey (2011), in a local study, reported aalue as low as <1%. In the present study, we consider 3.5% as aean value that is consistent with previous works from other bio-

eographic regions; however, higher percentages (>5%) have beenecorded in other areas with similar fish culture procedures (e.g.romey et al., 2009). A plausible explanation is the oceanographiconditions in the Macaronesian region, with continuous coastalurrents that disperse uneaten pellets to the surrounding areasRiera et al., 2015).

.2. Uneaten pellets lost to the environment and consumed byild fishes

Fish farms attract a variety of fish species by providing a solidtructure and a constant food supply via released pellets from sea-ages (Dempster et al., 2006; Riera et al., 2014; Tuya et al., 2006).he magnitude of attracting large multi-species schools of pelagicsh is such, that sea-cage fish farms has been reported to act as

super-FADs’ (FAD: fish aggregation device) (Dempster et al., 2002). range of studies have focused on understanding the way thesesh assemblages change through a variety of spatial and tempo-al scales (Boyra et al., 2004; Fernandez-Jover et al., 2008; Uglemt al., 2014). Despite these efforts, little is known about the amountf uneaten food consumed by wild populations around cages, aey factor for modelling farm solid fluxes and subsequent benthic

mpacts. This rate is a site/regional related process and, therefore,he use of generalised values may not be suitable for all mod-lling situations. Some authors reported values as low as 33% ofasted feed (Dempster et al., 2011; Sanchez-Jerez et al., 2011). On

he contrary, Felsing et al. (2005) and Vita et al. (2004) registeredercentages up to 60% and 80%, respectively. The first particulateispersion model that incorporated a wild fish waste removal rou-ine was MERAMOD, although no specific studies were undertakeno estimate a wasted feed percentage (Cromey et al., 2012). Theormer authors suggested that 5 − 50% of wasted feed is consumedy wild fishes, although direct observations during field samplingonfirmed that, in most situations, pellets were consumed by wildshes (Thetmeyer et al., 2003).

In the present study, fish aggregations played a pivotal role,onsuming most of uneaten pellets (ca. 97% in the Canaries and5% in Madeira). These percentages may be explained by the olig-trophic nature of Macaronesian waters, underpinning low fishiomass in coastal environments. Hence, the organic output fromffshore cages constitutes a high nutritional and easily accessibleood source for fish aggregations.

.3. Dispersion coefficients

Dispersion coefficients measured in Mediterranean sites var-ed from <0.01 m2s−1 to >0.4 m2s−1 for the most dispersive site.

owever, unless site-specific data are available, the worst-casealue of 0.1 m2s−1 was recommended. In Scotland, regulatory mod-ls apply a standardised horizontal dispersion coefficient (kx, ky)f 0.1 m2s−1 unless site-specific data are provided (SEPA, 2003).

ling 361 (2017) 122–134

Both models used a value of 0.001 m2s−1 for the vertical dispersioncoefficient (kz). In this study, Macaronesian horizontal dispersioncoefficients (0.041–0.917 m2 s−1) are higher than those used in theMediterranean and Scotland, due to the high hydrodynamics inMacaronesian aquaculture sites.

4.4. Quantification of benthic impact

The relationship between solid fluxes and the AMBI indexhas been developed and validated for MACAROMOD. Low organicinputs underpin an increase of AMBI due to the oligotrophic con-dition of Macaronesian coastal ecosystems. However, high organicinputs (>12 kg solids m−2 yr−1) produce significant decreases in theecological status.

This relationship (flux-AMBI) has been solely developed andvalidated for MACAROMOD. Hence, it is not suitable to use thispredictive result for other dispersion models, developed in areasunder different aquaculture husbandry practices, oceanographicand ecological conditions.

4.5. Model validation

High correlation rates were found between observed and pre-dicted solid fluxes, therefore demonstrating that the model wasadequately validated. The same occurred between observed andpredicted values of AMBI; thus, predictions are reliable along therange of this ecological status index. Both results (solid fluxesand AMBI) were statistically confirmed at 5% of significance level.Hence, the model validation can be classified as “very good”, basedon the results of previous calculations. This classification belongs toresults where R2 is significantly different from zero and the regres-sion line does not differ significantly from the line with slope oneand intercept zero (Portilla and Tett, 2007; Jusup et al., 2009).

4.6. Model performance and limitations

Feed pellet composition is species-specific, thus their digestibil-ity and faeces characteristics have a key role on settling velocityrates (Vassallo et al., 2006). For these reasons, the use of non-specific data for waste dispersion modelling leads to considerableinaccuracies (Pérez et al., 2014). MACAROMOD has been validatedfor a high number of offshore fish farms in the Macaronesianregion, covering all oceanographic conditions from coastal systemswhere aquaculture may be developed. This model produces rea-sonable predictions under different hydrodynamic conditions anddepths. MACAROMOD generates good predictions of fluxes (faecesand uneaten pellets), including various particle tracking techniquesthat approximate accurately to natural conditions in the field. Thepercentages of lost pellets are also quantified considering differentscenarios of husbandry technique, e.g. manual or automatized. Thismodel also reaches a satisfactory level of validation considering therole of wild fishes around cages, which are important to diminishorganic footprint on surrounding areas in the Macaronesian region.The optimisation of model performance resulted in 2.5% of pelletswastage and 97.5% of this is consumed by wild fishes.

The effects of depositional fluxes on the seabed are predicted byMACAROMOD; thus, future impacts can be predicted prior to a fieldsurvey. The deposition footprint is predicted by the indicator AMBI,which has been previously used as a good predictor of environmen-tal status (Borja et al., 2000). The predicted results (flux-AMBI) areconsistent with previous field observations (Riera et al., 2015), witha threshold of organic input from which a sharp impoverishment

of ecological status.

The present modelling tool, MACAROMOD, can be used topredict solid fluxes and subsequent benthic response from off-shore sea-bream (Sparus aurata), sea-bass (Dicentrarchus labrax)

Model

aroM(etchmi

A

PsmM

R

B

B

B

B

B

B

C

C

C

C

C

C

D

D

D

D

D

R. Riera et al. / Ecological

nd meagre (Argyrosomus regius) sea-cages in the Macaronesianegion. The model accuracy greatly depends on the magnitudef fluxes and benthic indicators, i.e. AMBI. All data integratingACAROMOD need to be accomplished by daily husbandry data

number of cultured tonnes, fish sizes, pellet quantity, among oth-rs). MACAROMOD has been tested in unvegetated sandy seabedshat constitute the most suitable ecosystem to install offshore aqua-ulture cages in the Macaronesian region; hence, MACAROMODas not considered to be used in rocky substrates and/or seagrasseadows. Particle flocculation or disaggregation was not included

n MACAROMOD.

cknowledgements

The present study was funded by the Transnational Cooperationrogram MAC 2007–2013 (PCTMAC), Research Project “Disper-ion of organic matter from aquaculture farms: development of aathematical model to assure environmental sustainability” (code:AC/3/C136).

eferences

everidge, M., 2004. Cage Aquaculture, third edition. Blackwell Publishing Ltd.,Oxford, UK.

orja, A., Franco, J., Perez, V., 2000. A Marine Biotic Index to establish the ecologicalquality of soft-bottom benthos within European estuarine and coastalenvironments. Mar. Pollut. Bull. 40, 1100–1114.

orja, A., Muxika, I., Franco, J., 2003. The application of a Marine Biotic Index todifferent impact sources affecting soft-bottom benthic communities alongEuropean coasts. Mar. Pollut. Bull. 46, 835–845.

orja, A., Franco, J., Muxika, I., 2004. The biotic indices and the Water FrameworkDirective: the required consensus in the new benthic monitoring tools(correspondence). Mar. Pollut. Bull. 48, 405–408.

oyra, A., Sanchez-Jerez, P., Tuya, F., Espino, F., Haroun, R., 2004. Attraction of wildcoastal fishes to an Atlantic subtropical cage fish farms Gran Canaria, CanaryIslands. Env. Biol. Fish. 70, 393–401.

rigolin, D., Pastres, R., Tomassetti, P., Porrello, S., 2010. Modelling the biomassyield and the impact of seabream mariculture in the Adriatic and TyrrhenianSeas (Italy). Aquac. Int. 18, 149–163.

airney, D., Morrisey, D., 2011. Estimation of Feed Loss from Two Salmon CageSites in Queen Charlotte Sound. NIWA Client Report n◦ NEL2011-026. NationalInstitute of Water and Atmospheric Research, Port Nelson.

hamberlain, J., Stucchi, D., 2007. Simulating the effects of parameter uncertaintyon waste model predictions of marine finfish aquaculture. Aquaculture 272,296–311.

romey, C.J., Nickell, T.D., Black, K.D., 2002a. DEPOMOD-modelling the depositionand biological effects of waste solids from marine cage farms. Aquaculture 214,211–239.

romey, C.J., Nickell, T.D., Black, K.D., Provost, P.G., Griffiths, C.R., 2002b. Validationof a fish farm water resuspension model by use of a particulate tracerdischarged from a point source in a coastal environment. Estuaries 25,916–929.

romey, C.J., Nickell, T.D., Treasurer, J., Black, K.D., Inall, M., 2009. Modelling theimpact of cod (Gadus morhua L.) farming in the marineenvironment-CODMOD. Aquaculture 289, 42–53.

romey, C.J., Thetmeyer, H., Lampadariou, N., Black, K.D., Kögeler, J., Karakassis, I.,2012. MERAMOD: predicting the deposition and benthic impact of aquaculturein the eastern Mediterranean Sea. Aquac. Environ. Interact. 2, 157–176.

apueto, G., Massa, F., Costa, S., Cimoli, L., Olivari, E., Chiantore, M., Federici, B.,Povero, P., 2015. A spatial multi-criteria evaluation for site selection of offshoremarine fish farm in the Ligurian Sea, Italy. Ocean Coast. Manage. 116, 64–77.

empster, T., Sanchez-Jerez, P., Bayle-Sempere, J.T., Giménez-Casalduero, F., Valle,C., 2002. Attraction of wild fish to seacage fish farms in the south-westernMediterranean Sea: spatial and short-term temporal variability. Mar. Ecol.Prog. Ser. 242, 237–252.

empster, T., Fernandez-Jover, D., Sanchez-Jerez, P., Tuya, F., Bayle-Sempere, J.,Boyra, A., Haroun, R.J., 2005. Vertical variability of wild fish assemblagesaround sea-cage fish farms: implications for management. Mar. Ecol. Prog. Ser.304, 15–29.

empster, T., Fernández-Jover, D., Sánchez-Jerez, P., Tuya, F., Bayle-Sempere, J.,Boyra, A., Haroun, R.J., 2006. Vertical variability of wild fish assemblagesaround sea-cage fish farms: implications for management. Mar. Ecol. Prog. Ser.

304, 15–29.

empster, T., Sanchez-Jerez, P., Fernandez-Jover, D., Bayle-Sempere, J., Nilsen, R.,Bjørn, P.A., 2011. Proxy measures of fitness suggest coastal fish farms can act aspopulation sources and not ecological traps for wild gadoid fish. PLoS One 6,e15646.

ling 361 (2017) 122–134 133

Dudley, R.W., Panchang, V.G., Newell, C.R., 2000. Application of a comprehensivemodeling strategy for the management of net-pen aquaculture wastetransport. Aquaculture 187, 319–349.

Edgar, G., Davey, A., Shepherd, C., 2010. Application of biotic and abiotic indicatorsfor detecting benthic impacts of marine salmonid farming among coastalregions of Tasmania. Aquaculture 307, 212–218.

FAO, 2013. Statistics and Information Service FishStatJ: Universal Software forFishery Statistical Time Series. Fisheries and Aquaculture Department.

Felsing, M., Glencross, B., Telfer, T., 2005. Preliminary study on the effects ofexclusion of wild fauna from aquaculture cages in a shallow marineenvironment. Aquaculture 243, 159–174.

Fernandez-Jover, D., Sanchez-Jerez, P., Bayle-Sempere, J.T., Valle, C., Dempster, T.,2008. Seasonal patterns and diets of wild fish assemblages associated withMediterranean coastal fish farms. ICES J. Mar. Sci. 65, 1153–1160.

Findlay, R.H., Watling, L., 1994. Towards a process level model to predict the effectof salmon net-pen aquaculture on benthos. In: Hargrave, B.T. (Ed.), ModellingBenthic Impacts of Organic Enrichment from Marine Aquaculture. CanadianTechnical Report of Fisheries and Aquatic Sciences, Canada, pp. 47–48.

Gillet, D.J., Fedrerick-Holland, A., Sanger, D.M., 2007. On the ecology ofoligochaetes: monthly variation of community composition and environmentalcharacteristics in two South Carolina tidal creeks. Est. Coast. 30, 238–252.

Gowen, R.J., Bradbury, N.B., Brown, J.R., 1989. The use of simple models in assessingtwo of the interactions between fish farming and the marine environment. In:De Pauw, N., Jaspers, E., Wilkins, N. (Eds.), Aquaculture-A Biotechnology inProgress. European Aquaculture Society, Belgium, pp. 1071–1080.

Gowen, R.J., Smyth, D., Silvert, W., 1994. Modelling the spatial distribution andloading of organic fish farm waste to the seabed. In: Hargrave, B.T. (Ed.),Modelling Benthic Impacts of Organic Enrichment from Marine Aquaculture.Canadian Technical Report of Fisheries and Aquatic Sciences, Canada, pp.19–30.

Grall, J., Glémarec, M., 1997. Using biotic indices to estimate macrobenthiccommunity perturbations in the Bay of Brest. Est. Coast. Shelf Sci. 44, 43–53.

Harmelin-Vivien, M.L., Harmelin-Vivien, J.G., Chauvet, C., et al., 1985. Evaluationvisuelle des peuplements et populations de poissons: Méthodes et problèmes.Terre Vie 40, 467–539.

Heilskov, A.C., Alperin, M., Holmer, M., 2006. Benthic fauna bio-irrigation effects onnutrient regeneration in fish farm sediments. J. Exp. Mar. Biol. Ecol. 339,204–225.

Henderson, A., Gamito, S., Karakassis, I., Pederson, P., Smaal, A., 2001. Use ofhydrodynamic and benthic models for managing environmental impacts ofmarine aquaculture. J. Appl. Ichthyol. 17, 163–172.

Hevia, M., Rosenthal, H., Gowen, R.J., 1996. Modelling benthic deposition under fishcages. J. App. Icht. 12, 71–73.

Holmer, M., 2010. Environmental issues of fish farming in offshore waters:perspectives, concerns and research needs. Aquac. Environ. Interact. 1, 57–70.

Jusup, M., Klanhscek, J., Petricioli, D., Legovic, T., 2009. Predictingaquaculture-derived benthic organic enrichment: model validation. Ecol. Mod.220, 2407–2414.

Kalantzi, L., Karakassis, I., 2006. Benthic impacts of fish farming: meta-analysis ofcommunity and geochemical data. Mar. Pollut. Bull. 52, 484–493.

Keeley, N.B., Cromey, C.J., Goodwin, E.O., Gibbs, M.T., Macleod, C.M., 2013.Predictive depositional modelling (DEPOMOD) of the interactive effect ofcurrent flow and resuspension on ecological impacts beneath salmon farms.Aquac. Environ. Interact. 3, 275–291.

Kingsford, M., Battershill, C., 1998. Studying Marine Temperate Environments: AHandbook for Ecologists. Canterbury University Press Christchurch, NewZealand, 335 pp.

Longhurst, A., 2007. Ecological Geography of The Sea, second edition. AcademicPress, San Diego, 575 p.

Machias, A., Karakassis, I., Labropoulou, M., Somarakis, S., Papadopoulou, K.N.,Papaconstantinou, C., 2004. Changes in wild fish assemblages after theestablishment of a fish farming zone in an oligotrophic marine environment.Est. Coast. Shelf Sci. 60, 771–779.

Magill, S.H., Thetmeyer, H., Cromey, C.J., 2006. Settling velocity of faecal pellets ofgilthead sea bream (Sparus aurata L.) and sea bass (Dicentrarchus labrax L.) andsensitivity analysis using measured data in a deposition model. Aquaculture251, 295–305.

Mesple, F., Troussellier, M., Casellas, C., Legendre, P., 1996. Evaluation of simplestatistical criteria to qualify a simulation. Ecol. Mod. 88, 9–18.

Pérez, O.M., Telfer, T.C., Beveridge, M.C., Ross, L.G., 2002. Geographical InformationSystems (GIS) as a simple tool for modelling waste distribution under marinefish cages. Est. Coast. Shelf Sci. 54, 761–768.

Pérez, O., Almansa, E., Riera, R., Rodriguez, M., Ramos, E., Costa, J., Monterroso, O.,2014. Food and faeces settling velocities of meagre (Argyrosomus regius) and itsapplication for modelling waste dispersion from sea cage aquaculture.Aquaculture 420, 171–179.

Papageorgiou, N., Kalantzi, I., Karakassis, I., 2010. Effects of fish farming on thebiological and geochemical properties of muddy and sandy sediments in theMediterranean Sea. Mar. Environ. Res. 69, 326–336.

Portilla, E., Tett, P., 2007. Assessing Goodness of Fit for ECASA Models (ECASAInternal Paper). Napier University, Edinburgh, Scotland.

Read, P.A., Fernandes, T.F., Miller, K.L., 2001. The derivation of scientiic guidelinesfor best environmental practice for the monitoring and regulation of marineaquaculture in Europe. J. Appl. Ichthyol. 17, 146–152.

1 Model

R

R

R

R

S

S

S

S

T

34 R. Riera et al. / Ecological

iera, R., Monterroso, O., Rodríguez, M., Ramos, E., Pérez, O., 2011. Biotic indexesreveal the impact of harbour enlargement on benthic fauna. Chem. Ecol. 27,311–326.

iera, R., Sacramento, A., Pérez, Ó., Monterroso, Ó., Ramos, E., Rodríguez, M.,Almansa, E., 2013. Effects of organic enrichment on macrofauna communitystructure: an experimental approach. Braz. J. Ocean. 61 (4), 223–229.

iera, R., Pérez, O., Rodríguez, M., Ramos, E., Monterroso, O., Sacramento, A., 2014.Are assemblages of the fireworm Hermodice carunculata enhanced insediments beneath off-shore fish cages in the Canarian archipelago? ActaOcean. Sin. 33, 108–111.

iera, R., Becerro, M.A., Ramos, E., Monterroso, O., Rodríguez, M., 2015. Response ofdifferent benthic habitats to off-shore fish cages. Aquac. Res. 46, 1490–1500.

EPA, 2003. Marine Fish Farming Manual (annex H). Scottish EnvironmentProtection Agency, Stirling, Scotland www.sepa.org.uk/aquaculture.

EPA (Scottish Environmental Protection Agency), 2005. Regulation andmonitoring of marine cage fish farming in Scotland. In: Annex H: Method forModelling In-Fed Antiparasitics and Benthic Effects. SEPA, Stirling.

anchez-Jerez, P., Fernandez-Jover, D., Uglem, I., Arechavala-Lopez, P., Dempster,T., Bayle-Sempere, J.T., Valle, C., Izquierdo, D., Bjørn, P.A., Nilsen, R., 2011.Coastal fish farms as fish aggregation devices (FADs). In: Bortone, S.A., Pereira,F., Brandini, G., Fabi, G., Otake, S. (Eds.), Artificial Reefs in Fishery Management.CRC Press, Taylor & Francis Group, FL, USA, pp. 187–208.

tigebrandt, A., Aure, J., Ervik, A., Kupta Hansen, P., 2004. Regulating theenvironment impact of intensive marine fish farming III: a model forestimation of the holding capacity in the Modelling ongrowing fish farm

monitoring system. Aquaculture 234, 239–261.

elfer, T.C., Atkin, H., Corner, R.A., 2009. Review of Environmental ImpactAssessment and Monitoring in Aquaculture in Europe and North America.Environmental Impact Assessment and Monitoring in Aquaculture. FAO FishAquac Tech Pap No. 527. UN FAO, Rome, pp. 285–394.

ling 361 (2017) 122–134

Tett, P., 2007. ECASA Toolbox, Available online at:www.ecasatoolbox.org/uk/the-toolbox/eia-species/models/model-reliability.

Thetmeyer, H., Pavlidis, A., Cromey, C., 2003. Development of MonitoringGuidelines and Modelling Tools for Environmental Effects from MediterraneanAquaculture - Newsletter 3: Interactions Between Wild and Farmed Fish, 7 pp.

Troell, M., Halling, C., Neori, A., Chopin, T., Buschmann, A.H., Kautsky, N., Yarish, C.,2003. Integrated mariculture: asking the right questions. Aquaculture 226,69–90.

Tuya, F., Sánchez-Jerez, P., Dempster, T., Boyra, A., Haroun, R.J., 2006. Changes indemersal wild fish aggregations beneath a sea-cage fish farm after thecessation of farming. J. Fish Biol. 69, 682–697.

Uglem, I., Karlsen, Ø., Sanchez-Jerez, P., Sæther, B., 2014. Impacts of wild fishesattracted to open-cage salmonid farms in Norway. Aquac. Env. Interact. 6,91–103.

Vassallo, P., Doglioli, A.M., Rinaldi, F., Beiso, I., 2006. Determination of physicalbehaviour of feed pellets in Mediterranean water. Aquac. Res. 37, 119–126.

Vezzulli, L., Marrale, D., Moreno, M., Fabiano, M., 2003. Sediment organic matterand meiofauna community response to long-term fish-farm impact in theLigurian Sea (Western Mediterranean). Chem. Ecol. 19, 431–440.

Vita, R., Marín, A., Madrid, J.A., Jiménez-Brinquis, B., Cesar, A., Marín-Guirao, L.,2004. Effects of wild fishes on waste exportation from a Mediterranean fishfarm. Mar. Ecol. Prog. Ser. 277, 253–261.

Wassmann, P., Heiskanen, A.S., Lindahl, O., 1991. Sediment trap studies in theNordic countries. In: Sediment traps in marine ecological research andmonitoring, Vol 2. Proceedings of the symposium. Kristineberg Marine

Biological Station, Fiskebäckskil, 21–25 November 1990. In: Wassmann, P.,Heiskanen, A.S., Lindahl, O. (eds.) Sediments trap studies in Nordic countries.NurmiPrint OY, Nurmijärvi.

Yanagi, T., Murashita, K., Higuchi, H., 1982. Horizontal turbulent diffusivity in thesea. Deep-Sea Res. 29, 217–226.