Lny Mercier , Jacques Panli , Christelle Paillon a, Awa Ndiaye , David Mouillot ,
SYM, PlUMR 5119 UM2-CNRS-IRD-IFREMER-UM1 ECOSYM, IR
a r t i c l e i n f o
Article history:Received 5 July 2010Accepted 13 February 2011Available online 18 February 2011
population dynamics (Rochet and Trenkel, 2003). As overshingmodies life history traits linked to individual growth, such as ageat rst maturity (Ali et al., 2003), assessing growth parameters isalso vital for the denition of strategies for sustainable sheries(Rochet and Trenkel, 2003). During the last century, otolithometry
unparalleled accuracy (Campana et al., 2000).Many published works therefore successfully used otoliths to
assess growth in exploited sh species (e.g. Reibish, 1899; Campanaand Neilson, 1982; Gordoa and Moli, 1997; Tomas and Panli, 2000;Pajuelo et al., 2006). However, in most cases, population samplingwhen assessing age and growth is based on commercial catches andonly the growth function of Von Bertalanffy (VBGF) is tted to theageelengthdata to infer growthparameters. This generates biases inthe estimation of population growth parameters, mainly because
* Corresponding author.
Contents lists availab
Estuarine, Coastal a
Estuarine, Coastal and Shelf Science 92 (2011) 534e545E-mail address: email@example.com (A.M. Darnaude).of juveniles and adult sh is then advised to obtain objective estimations of growth parameters whensampling cannot be corrected towards older sh.
2011 Elsevier Ltd. All rights reserved.
Accurate knowledge of sh growth parameters is extremelyimportant for population management since it allows prediction ofgrowth rate evolution, identication of cohorts and evaluation of
has emerged as the most useful tool in sh stock and age assess-ment (Panli et al., 2002). As otoliths continuously growthroughout sh life, with no evidence of resorption (Campana,1999), the daily and seasonal (annual) increments recorded intheir structure allow retrospective assessment of sh growth withVon Bertalanffy growth functionmulti-model inferenceAIC weightsMediterranean SeaGulf of Lions0272-7714/$ e see front matter 2011 Elsevier Ltd.doi:10.1016/j.ecss.2011.02.001D B.P. 1386, 18524 Dakar, Senegal
a b s t r a c t
Accurate knowledge of sh age and growth is crucial for species conservation and management ofexploited marine stocks. In exploited species, age estimation based on otolith reading is routinely usedfor building growth curves that are used to implement shery management models. However, theuniversal t of the von Bertalanffy growth function (VBGF) on data from commercial landings can lead touncertainty in growth parameter inference, preventing accurate comparison of growth-based historytraits between sh populations. In the present paper, we used a comprehensive annual sample of wildgilthead seabream (Sparus aurata L.) in the Gulf of Lions (France, NW Mediterranean) to test a method-ology improving growth modelling for exploited sh populations. After validating the timing for otolithannual increment formation for all life stages, a comprehensive set of growth models (including VBGF)were tted to the obtained ageelength data, used as a whole or sub-divided between group 0 individualsand those coming from commercial landings (ages 1e6). Comparisons in growth model accuracy basedon Akaike Information Criterion allowed assessment of the best model for each dataset and, when nomodel correctly tted the data, a multi-model inference (MMI) based on model averaging was carriedout. The results provided evidence that growth parameters inferred with VBGF must be used with highcaution. Hence, VBGF turned to be among the less accurate for growth prediction irrespective of thedataset and its t to the whole population, the juvenile or the adult datasets provided different growthparameters. The best models for growth prediction were the Tanaka model, for group 0 juveniles, and theMMI, for the older sh, conrming that growth differs substantially between juveniles and adults. Allasymptotic models failed to correctly describe the growth of adult S. aurata, probably because of the poorrepresentation of old individuals in the dataset. Multi-model inference associated with separate analysisaUMR 5119 UM2-CNRS-IRD-IFREMER-UM1 ECObace Eugne Bataillon, 34095 Montpellier Cedex 5, FranceAudrey M. Darnaude a,*Otolith reading and multi-model inferenand growth in the gilthead seabream Sp
journal homepage: wwwAll rights reserved.e for improved estimation of agerus aurata (L.)
le at ScienceDirect
nd Shelf Science
juveniles are often included in the sample although VBGF badlydescribes young sh growth (Gamito, 1998; Charnov, 2008).Improved sh growth modelling is thus urgently needed to produceaccurate growth curves that encompass the shape for growth at alllife stages. In particular, condent assessment of early sh growth,based on appropriate models for young stages, is of primaryimportance, not only for sheries management strategies(Katsanevakis, 2006) but also in an ecological perspective (Dioufet al., 2009). Fish lifetime migrations between habitats, forexample, can be assessed by combining growth measurements andmicrochemical analysis on the otoliths (Babaluk et al., 1997;Campana et al., 1999; Fodrie and Herzka, 2008). However, thisnecessitates precise timetables of otolith growth, both at the annualand seasonal scale, in order to correlate the position of the areaanalysed in theotolith to a given calendardate and its environmentalcharacteristics (Campana, 1999; Campana and Thorrold, 2001).
In the present work, we compared the accuracy in ageelengthprediction of various growth models (including the traditionalVBGF) for different life stages of sh, using comprehensive annualdata gathered from juveniles and adults of one of the most valuableMediterranean sh species, the gilthead seabream Sparus aurata (L.,1758). Despite the high commercial value of S. aurata, manyfundamental aspects of its biology are still poorly known. Previouswork on the species mainly focused on aquaculture production (e.g.Gordin and Zohar, 1978; Kentouri et al., 1994; Brusl-Sicard andFourcault, 1997), and studies on wild populations are still scarce(Arias, 1980; Kraljevic et al., 1998; Chaoui et al., 2006; Mehanna,2007). As for most exploited sh, most growth studies in thespecies were based on data from commercial landings (e.g.Lasserre, 1974; Arias, 1976; Kraljevic and Dulcic, 1997), and oftenincluded age class 0 sh (Bentuvia, 1979; Emre et al., 2009) for
which the application of VBGF was shown to be inadequate(Gamito,1998). In the present study, a comprehensive set of growthmodels was tested in addition to VBGF and the best model choicewas made objectively for each life stage using an informationtheory approach (Burnham and Anderson, 2002). This modernapproach, based on the likelihood theory and widely used acrossbiological elds of research (Johnson and Omland, 2004), is anemerging tool for mathematical expression of the mean populationgrowth in sh (Katsanevakis and Maravelias, 2008). When the datasupport more than a single model, it allows inference of robustgrowth curves by averaging inside the condent set of models(Diouf et al., 2009). This should allow re-assessment of growthparameters for both S. aurata juveniles and adults with adequatemodels and accurate estimations of the growth potential of thespecies in aquaculture and in the wild. This is especially needed,both at the local scale and over the whole repartition area of thespecies (Lasserre and Labourg, 1974), especially in the context ofglobal change, now that it has been reported as potentially invasivein places where it is exotic (Balart et al., 2009).
2. Materials and methods
2.1. Fish sampling
The 315 juveniles and adults of Sparus aurata used in the presentstudy were collected on amonthly basis between January 2008 andJanuary 2009, in the Gulf of Lions (France, NW Mediterranean Sea,Fig. 1). In the area, S. aurata spawns at sea in the winter (Audouin,1962) but colonizes coastal lagoons, mainly from March to October(Lasserre and Labourg, 1974). This lagoon use is facultative for theadults but seems to be the rule for the juveniles, at least during
L. Mercier et al. / Estuarine, Coastal and Shelf Science 92 (2011) 534e545 535Fig. 1. Map of the sampling area for Sparus aurata. All the adults were caught within the shcollected within the Mauguio lagoon.ing area of the sh auction market of Ste (dotted line) while age class 0 juveniles were
l andtheir rst year of life. The legal minimum catch size is 20 cm TL inthe area (Crespi, 2002). Therefore, while all the S. aurata individualsabove 20 cm were collected all year round from the sh auctionmarket in Ste (Fig. 1), smaller sh were caught only between Apriland December 2008 in the Mauguio lagoon, one of their mainnursery sites in the area (Quignard et al., 1983). Fish captureinvolved demersal trawls and shing lines at sea, and capchades(traditional Mediterranean shing nets) and beach seines in thelagoons. All the sh collected were measured (total length TL, inmm), weighted (W in g) and sexed on the day of capture, asjuvenile, when no gonad was detectable from visual observation,and as male, female, intersexual or undetermined other-wise. They were then stored in the freezer until otolith extraction.
2.2. Otolith reading
For each sh, both sagittal otoliths were removed, but only theleft sagitta was used for age estimation. As annual growth markswere poorly visible on the whole otolith, all sagittae wereembedded in Epoxy resin (Araldite, 2020) and transverse thinsections (500 mm) including the otolith core were made usinga precision low speed saw (Buehler, Isomet 1000). Sections weremounted on microscope slides using epoxy resin and polished toexpose the nucleus. Following current protocols (Panli et al.,2002), all sections were etched with EDTA 5% during 2 mn 30 sand subsequently stained with Toluidine Blue during 2 mn in orderto increase the contrast between translucent and opaque growthbands and improve otolith reading. Each otolith section wasexamined under a stereomicroscope (Olympus SZX12) at 10e30magnications with reected light and photographed andmeasured with a Computer-Assisted Age and Growth Estimation(CAAGE) system, consisting in a stereomicroscope (OlympusSZX12) linked to a camera (C5 Jenoptik) connected to a PC imageanalysis system (ProgRes CapturePro 2.5, Jenoptik). For age esti-mation, annual growth bands were identied and counted twice bytwo different readers. If there was any discrepancy, a third obser-vation was carried out to reach a nal decision. When the threereadings differed, the otolith was rejected from the analyses.
For each sh, individual agewas calculated by adding the annualfraction between the capture date and the theoretical birth date tothe number of complete annual growth bands identied in theotolith, following Gordoa and Moli (1997). As S. aurata spawns inNovembereFebruary in the Gulf of Lions (Lasserre, 1974), January1st was used as the theoretical date of birth for all sh. In order toprecisely assess the timing for annual mark (opaque growth band)formation in all age classes, monthly percentages of otoliths withopaque and translucent margins were calculated based on allotolith readings. To ascertain this measure, average marginalincrement widths were also assessed monthly for all age classesabove 1 year. For this, the marginal increment was dened as thedistance from the last opaque zone to the otolith edge and it wasmeasured to the nearest mm on the dorsal radius of each otolithsection (Fig. 2). Measurements of the distances (in mm) from thecore to each opaque zone on the dorsal radius also allowed esti-mation of growth for each age class.
2.3. Growth modelling
All statistical analyses were carried out using R software (RDevelopment Core Team, 2009). The growth of Sparus aurata wasassessed based on the age-at-length data derived from otolithobservation and shmeasurement. All the individual data gatheredwere used to generate 3 datasets: the whole population dataset(N 315), the group 0 juveniles dataset (including the 173
L. Mercier et al. / Estuarine, Coasta536juveniles of the year collected in the lagoon, with TL 21 cm) andthe commercial landings dataset (including the 142 remainingindividuals, between 1 and 6 years of age, mainly collected from theauction market, and with TL between 15 and 50 cm). Six differentgrowth models were tted to these 3 datasets (Table 1). Amongthem, the traditional Von Bertalanffy growth function (VBGF) is themost studied and commonly applied model among all the growthmodels. However, the Gompertz model (GM) has been shown tobetter describe absolute growth for many aquatic species(Katsanevakis, 2006). These two models both assume asymptoticgrowth. However, many sh and other aquatic organisms seem notto grow asymptotically (e.g. Rafail, 1971), and non-asymptoticmodels such as the Power function (PF) and the Tanaka model (TM)(Tanaka, 1982) have been proposed in these cases. The Persistencemodel (PM) and the Extended Power model (EPM) are non-asymptotic quasi-sigmoid models recently developed for the studyof species area-relationships (Tjorve, 2009). They both tendtowards the power function (PF) when age becomes large, whilethe TM is a quasi-sigmoid function with non-asymptotic limits(Tanaka, 1982). Their inclusion within the set of models testedtherefore allowed encompassing most of the likely shapes for shgrowth.
All models were tted using the Gauss-Newton algorithm fornon-linear regression, with residual sum of squares (RSS) as a crite-rion. Under the assumptions of residual normality and homosce-dasticity (previously veried with Lilliefors test), RSS was directlylinked to the likelihood (Burnham and Anderson, 2002). Forcomparison with past studies, the growth parameters associated tothe VBGF (LN, K and t0) were calculated for all datasets and overallgrowth performance V0 was estimated in each case following theformula (Munro and Pauly, 1983; Pauly and Munro, 1984):
F0 logK 2logLN:The comparison of the models was performed within an infor-
mation theory approach based on the likelihood theory (Burnhamand Anderson, 2002). For each dataset, R2 was calculated toassess the percentage of variance in sh length explained by age.However, because R2 is generally not consistent for model selectionwithin the non-linear regression context, Akaike InformationCriterions (AIC) were computed, following:
where RSS is the residual sum of squares, n the number of obser-vations and k the number of estimable parameters.
Differences between the AIC for each model and the minimumAIC obtained over the k models tested (DAIC) were calculated tocompare between models, with only DAIC variations superior to 2being considered signicant following Burnham and Anderson(2002). The AIC is a very efcient tool for selection betweencompeting models (Katsanevakis, 20...