Establishing traceability for cod (Gadus morhua): determining location of spawning and harvest

Deliverables

Using the microchemistry of otolith cores to place cod back to their origin gave the following results. There were no significant differences between areas in otolith core concentrations of Mg or Pb. The otoliths of fish from the Baltic had significantly more Mn and Ba, and lower Sr that those of fish from other areas. The three sampling seasons were grouped together for discriminant analysis. Classification was based on 8 elements: Na, Mg, Mn, Cu, Rb, Sr, Ba and Pb. Pb made no significant contribution to the discrimination. The most important elements in the classification were Rb, Mn, Sr, Ba. However, better classification was achieved using 7 elements, excluding Pb. The percent correct classifications ranged from 71% for Baltic fish to 38% for Celtic Sea fish. The cross validated success of the Artificial Neural Network (ANN) was mixed, ranging from 0 to 82.4%. With the exception of CTB (Celtic Sea) and CTF (Norwegian waters), the classification success was above the expected random probability. In the case of CTF, there were relatively few samples with which to train, cross validate or test the ANN. CTC (Icelandic waters), CTD (Irish Sea), CTE (North Sea) and CTG (Icelandic fish farm) all had a similar classification success of between 45 and 60%. The clearest classification success was with CTA (the Baltic) despite a relatively small number of samples analysed. In the case of CTB (Celtic Sea) the highest proportion of misclassified samples was to CTD (Irish Sea). CTB (Celtic Sea), CTC (Icelandic waters) and CTD (Irish Sea) all had been 15 and 22% of the fish misclassified as coming from CTE (North Sea). No fish were misclassified as coming from CTF (Norwegian waters), however, the training data set was small for this area. The linear regression of the ANN model was relatively high for the Baltic (CTA) and the Scottish fish farm (CTH), reflecting the high correct percentage of classification. The regression coefficient for the Icelandic waters (CTC) and, surprisingly, the Norwegian waters (CTF) also had reasonably high regression coefficients. All other areas had a relatively low linear regression coefficient that reflected the relatively low level of classification success. Testing: Autumn (feeding aggregation) core samples. The cross validated success of the ANN was mixed, ranging from 0 to 60.0%. Where test data existed, with the exception of CTB (Celtic Sea) the classification success was above the expected random probability. The highest classification success was with the two fish farms CTG (Icelandic fish farm) and CTH (Scottish fish farm). In the case of CTB (Celtic Sea) the highest proportion of misclassified samples was to CTD (Irish Sea) with a reasonably large proportion also in to CTC (Icelandic waters). The majority of the CTE (North Sea) were misclassified as coming from the Irish Sea (CTD). No fish were misclassified as coming from CTF (Norwegian waters) or CTB (Celtic Sea). The highest linear regression of the ANN model were for the two fish farms (CTG and CTH) reflecting the reasonably high correct percentage of classification. All other areas had a relatively low linear regression coefficient that reflected the relatively low level of classification success.

The project was designed to determine if CODTRACE techniques were able to provide legally indisputable evidence for the harvest location of unknown cod. This was tested through transplants of eggs, and also through double blind testing of cod samples. With the transplantation experiments, eggs from Irish Sea cod were successfully transplanted to Iceland and reared in common conditions with Icelandic cod. Siblings from the same batch of Irish Sea eggs were reared in Irish Sea water in the Isle of Man concurrently with the fish in Iceland. All three groups of fish were sampled at five months of age; Irish Sea fish reared in Irish Sea water, Irish Sea fish reared in Icelandic water, and Icelandic fish reared in Icelandic water. Two of the CODTRACE techniques, the bacterial and the parasite assemblages, produced no useful results because the fish were too small and because of the water treatment used in a hatchery-rearing environment. Otolith microchemistry results indicated that many elements (Sr, Na, Pb, Ba, and Cu) were influenced by environmental conditions, but some elements (Mg and Mn) reflected the stock origin regardless of local conditions. Microsatellite characterisation revealed significant divergence between the Icelandic fish and the Irish Sea fish, whether transplanted or not. The otolith morphometry characterisations revealed that most of the transplanted fish were classified as Icelandic rather than Irish Sea fish; otolith shape was more strongly influenced by environmental factors (rearing conditions) than previously assumed. In summary, only genetic characteristics and some selected isotopes in the otoliths can be considered as invariant markers of the cod source population considered here. Body and otolith morphology, and most otolith microchemistry were affected by local conditions, thus reflecting harvest location, therefore these latter analyses are more likely to determine where a cod was harvested independently from the location of its natal population (important for separating where a migrating cod was actually caught).

Two Peer-Reviewed publications have resulted, and 3 other publications have already been submitted. In addition, 2 PhD theses and one MSc thesis were completed as the result of this project. Publications (submitted for review): Higgins, RM, B.S. Danilowicz, JA. Balbuena, AK. Daníelsdóttir, AJ. Geffen, WG. Meijer, J Modin, FE. Montero, C Pampoulie, D Perdiguero, A Schreiber, MÖ. Stefánsson, & B Wilson. Multivariate Fingerprints Reveal the Harvest Location of Atlantic Cod (Gadus morhua). Marine Ecology Progress Series. Pampoulie C., Jörundsdóttir Þ. D., Steinarsson A., Pétursdóttir G., Stefánsson M. Ö & Daníelsdóttir A. K., submitted. Genetic and growth rate comparisons of farmed and wild Icelandic populations of Atlantic Cod (Gadus morhua L.). Aquaculture. Pampoulie, C., Stefánsson, MO, Jörundsdóttir, TD, Danilowicz BS & Daníelsdóttir, AK. Recolonisation history and large scale dispersal in the open sea: Evidence from the North Atlantic cod, Gadus morhua L. Biological Journal of the Linnean Society. PhD thesis: Higgins, Ruth. Optimising classification analyses for the traceability of individual fish within the north-eastern Atlantic Ocean. Department of Zoology, University College Dublin, Ireland. 2002-2006. Perdiguero-Alonso, Diana. Parasite communities of cod in European waters: use as indicators of harvest location. University of Valencia, Spain. 2003-2007. MSc Thesis: Nkambo, Mujibi 2006 -Assessing welfare and condition in farmed fish (Atlantic cod, Gadus morhua) using otoliths and body morphology- MSc, Department of Biology, University of Bergen

Composition and diversity within the bacterial assemblage Phylogenetic analysis revealed that there was substantial diversity among sequences and when compared with database sequences, they were most similar to isolates and clones retrieved from marine samples. Sequences corresponding to the ?-proteobacteria dominated the clonal libraries for each sampling site and had high similarity (97-100%) to 16S rDNA sequences of Photobacterium sp., Psychrobacter sp., Acinetobacter sp. and Alteromonas sp. In contrast, ?-proteobacteria dominate free-living water marine bacteroplankton communities. Indeed, only a single member of the ?-proteobacterial group, with a 100% 16S rRNA gene identity to Loktanella salsilacus, was identified in our study and it comprised 4% of the Icelandic Sea clone library only. ?-Proteobacteria dominate libraries of bacteria isolated from marine systems. Clones clustering with 16S rDNA sequences from the CFB group showed the lowest level of similarity to published 16S rDNA sequence data, with two clones showing only a 92% similarity to the nearest relative (Bacteroides sp. ASF519) and it is possible that these clonal sequences (Accession Numbers DQ263705 and DQ263706) might represent novel genera. Discriminant analyses clustered T-RFLPs by sample location, suggesting that the biogeography of bacterial communities differs significantly with water mass location. In this study, members of ?-proteobacteria consistently dominated cod epidermal communities throughout the North-East Atlantic, in correspondence with previous studies of the bacterial flora of fish indigenous to this area. However, within the ?-proteobacterial group, there were site-specific differences in the numbers of cod colonised by either Photobacterium sp. or Psychrobacter sp.. These results are in agreement with previous studies based upon culturable bacteria which showed that samples from spatially distinct sites were frequently dominated by different species. This study confirms the ubiquity of the CFB previously reported for marine samples. If we assume no PCR (or other) bias, the dominant phylotypes detected in this study using T-RFLP are found to be existing in significant abundance, and must be assumed highly active. The oligotrophic nature of sea water and scarcity of nutrients mean that bacterial association with attached surfaces (such as marine aggregates and organisms) represents a good growth strategy and explains the composition of communities within fish epidermal mucous. Organisms related to the ?-proteobacteria (similar to those identified in this study) have been found to make up a significant proportion of particle-attached communities and to dominate these communities at depths greater than 50 m, similar to the depths at which the dermersal cod exist. Bacteria related to Psychrobacter sp., Acinetobacter sp. and Photobacterium sp. have all been previously isolated from fish; Photobacterium sp. (and closely related Vibrio sp.) have received much attention due to their luminous symbioses with marine organisms and chemotactic response to fish mucous. In this study, relatively high levels of CFB bacteria on all fish sampled were found. This correlates well with their affinity for nutrient-rich environments and the phenotypic property of the group for the degradation of HMW compounds and biopolymers. Since mucous consists primarily of hydrated long chain polysaccharides, it is very likely that CFB bacteria are well adapted to metabolise this abundant source of dissolved organic matter (DOM). In a review of the group, Kirchman states that the levels of DOM in oceans is too low to support the high abundance of CFB bacteria that are found in a free-living state and suggests that it is the release of these bacteria from associated communities which contributes to these unusually high numbers. The CFB bacteria were found to be a dominant group in the mucous of almost all cod sampled in this study. It is therefore likely that these fish-associated communities serve as a significant reservoir for the group in aquatic systems. The work in this study confirmed that a relative few phylogenetic clusters dominated the assemblages in the epidermal mucous of Atlantic Cod and that these comprised both resident and transient organisms. The data obtained in this study suggest that there may be a stable periodicity to bacterial community structure but the degree to which temporal and spatial scales affect this composition cannot be elucidated without analysing additional data over an extended sampling period. Ultimately, the classification success varied from season to season, but this data when analysed across seasons and sites simultaneously (see Chapter 3G- classification by combined analysis) classification was at 78.3% overall. This is a very high success rate for classification, and indicates that this method is worth continued development as a tool for the traceability of cod.

Using the microchemistry of otolith edges to place cod back to their origin gave the following results: 1. The best Artificial Neural Network (ANN) for these data utilise the isotopes Na23, Cu63, Rb85, Sr86 and Ba137 and gave an Mean Stardard Error of 0.14 for the training and cross validation. 2. Of the statistics used, and amongst the wild populations, the ANN had the least problem classifying the Baltic fish. Icelandic fish could be 62% classified correctly. All others were <40% correctly classified. 3. In wild populations, the Baltic, Icelandic, Irish Sea and North Sea classifications were above a random probability. 4. The ANN classified 60% and above of individuals into the two fish farms (Icelandic and Scottish).

For most of the location investigated, we did not observe any temporal genetic differentiation (first and second spawning seasons). Moreover, in most of the sampling areas we did not observe differences between harvest and spawning locations. Therefore, the samples could be combined for the analysis, which renders the assignment analysis stronger. The total percentage of individual assignment to their population of origin greatly varied from 43% (North Sea) to 83% (Baltic Sea). Even if the obtained percentage of assignment is very high in this study for a marine species, it will nonetheless be difficult to assign properly individuals coming from Irish, Celtic and North as well as individuals from Norway and Iceland. This is mainly due to the lack of genetic differentiation of population living in those areas. Celtic, Irish and North Seas are geographically relatively closed to each other and no obvious barriers to migration are present. The absence of genetic differentiation between those areas could merely be due to migration events between the populations living in those areas, thus rendering correct individuals assignment difficult. On the contrary, because of the large geographical distance involved between Norway and Iceland, it is unlikely that migration, which reduces the level of genetic differentiation between two populations, is responsible for the lack of genetic divergence between these two populations. On the contrary, recent common shared could explain the lack of genetic dissimilarity between Icelandic and Norwegian populations. Concerning the assignment of individuals to their origins, the observed genetic pattern was stable over the investigated seasons e.g. for the spawning and harvested locations: for the assignment and the origin of fish, the first spawning season was slightly different from the other seasons. In fact, during the first season, 14.61% of the individuals were not correctly assigned (17.14% unexplained, 30% because of potential migration and 52.86% because of potential common ancestor). For the second spawning and the harvest location, around 25% of the individuals were not correctly assigned among which 35% were due to migration, 58% to common ancestry and 7% were unexplained. This slight difference can in fact be due to sampling effect. Indeed during the first season, the number of individuals we caught per sampling location was not constant and could have induce a bias in the different percentage of individuals assigned. For example we only caught 29 fishes in North Sea for the first spawning location in this area and 60 during the second spawning period. Moreover, most of the fish caught in this area during the second spawning season could be considered as migrant, thus increasing the percentage of wrongly assigned individuals. Our results clearly demonstrated that individual assignment to population of origin may be, in some cases, particularly accurate, while in other cases individual origin can not be assessed. Microsatellite loci are useful genetic markers for traceability of individuals and assignment to population of origin, but genetic differentiation should be pronounced between the considered units. Here, at the individual base, the assignment was reasonable for Icelandic samples (farm and wild), the Baltic Sea, The Irish-North-Celtic Seas and the Scottish farm. The sample size of the Norwegian samples was probably too small to give an accurate estimate of the position of this population in this genetic pattern. Finally, 76.25% of the individual analysed could be assigned to the correct group. However, some individuals were not correctly assigned with STRUCTURE while the basic genetic analysis, especially the PCA-gene analysis clearly separated the investigated groups, except the Irish and Celtic Seas. Therefore, the traceability of individuals may not be clear, mainly because of common shared ancestor between the population considered, but the combination of the group (PCA) and individual (STRUCTURE) assignment should allow correct determination of fish origins.

10 Conference presentations and 4 Poster Presentations were provided on aspects of this project. Oral presentations: Geffen AJ, Modin J, Pampoulie CJ and Danielsdottir A. 2004. Comparison of otolith and genetic signatures in Northeast Atlantic cod (Gadus morhua) populations Third International Symposium on Fish Otolith Research and Application. July 11-16 2004, Townsville, Queensland, Australia. Geffen AJ. "Acquisition, analysis and interpretation of multimodal signals extracted from calcified structures: Chemical Information". Fifth EFARO Workshop "How can otolith research contribute at improving fisheries sciences?" December 2004, Brest, France Montero, F.E., Ferrer, E., Perdiguero-Alonso, D., Raga, J.A. & Balbuena, J.A. (2003). Providing traceability for cod captures: the role of parasites. Spring Meeting of the British Society of Parasitologists. Manchester, UK. Montero, F.E., Ferrer, E., Perdiguero-Alonso, D., Raga, J.A. & Balbuena, J.A. (2003). Establishing harvest location for atlantic cod: playing Sherlock Holmes with parasites. XXI Symposium Scandinavian Society of Parasitologist,Bergen, Norway. Montero, F.E., Perdiguero-Alonso, D., Raga, J.A. & Balbuena, J.A. (2004). Parasites as biological tags: a new conceptual approach to traceability of fish captures. IX European Multicolloquium of Parasitology. Valencia, Spain. Pampoulie, C., Þ. D. Jörunsdóttir, M. Ö. Stefánsson, B. S. Danilowicz and A. K. Daníelsdóttir Genetic variability of the Northeast Atlantic cod: past, present and future. ICES 2004. Perdiguero-Alonso, D., Aznar, F.J., Kostadinova, A., Montero F.E & Balbuena, J.A. (2004). Parasite infracommunities of Baltic and North Sea cod: a comparative approach. Spring Meeting of the British Society of Parasitologists. Chester, UK. Perdiguero-Alonso, D., Montero, F.E., Raga, J.A. & Balbuena, J.A. (2005). Accidental infections in Gadus morhua L.: observations on monogenean host-specificity. 5th International Symposium on Monogenea. Guangzhou, China. Perdiguero-Alonso, D., Montero, F.E., Raga, J.A. & Balbuena, J.A. (2006). Morphology and host specificity of Diclidophora merlangi (Monogenea). XI International Conference of Parasitology. Glasgow, UK. Perdiguero-Alonso, D., Montero, F.E., Raga, J.A. & Balbuena, J.A. (2006). Cuticular structure in three Ascarophis species. XI International Conference of Parasitology. Glasgow, UK. Poster presentations: Geffen, AJ. CODTRACE: Establishing traceability for cod (Gadus morhua): determining location of spawning and harvest. Sjømat for alle Seafood Conference, September 2003, Bergen Norway Kruber C, Geffen AJ, Tumyr O and Kosler J. Analysis of fish otoliths by laser ablation ICP mass spectrometry as a tool for tracing migration of cod. Nordic Conference on Plasma Spectrochemistry, June, Loen, Norway. 2006. Perdiguero-Alonso, D., Montero, F.E., Raga, J.A. & Balbuena, J.A. (2004). Geographical variability of parasite infracommunities of Atlantic cod as revealed by an Unsupervised Neural Network. IX European Multicolloquium of Parasitology. Valencia, Spain. Perdiguero-Alonso, D., Montero, F.E., Raga, J.A. & Balbuena, J.A. (2005). The species of Diclidophora in cod Gadus morhua L.. 5th International Symposium on Monogenea. Guangzhou, China.

1. Data on otolith shape has been extracted as Fourier coefficients and thereafter compiled into a database of descriptors that have been delivered for further analysis of combined data. 2. Otolith shape does not discriminate well between a large set of sample origins. In this study mean classification success ranged from 58.8% to 61.9% for 7 sample sites that were replicated three times. 3. Low variation in shape descriptors between sample origins does not necessarily imply that the classification success will increase. Otoliths from the Baltic were more easily distinguished from other otoliths compared to otoliths from the North Sea despite that both sample sites had a low variation in shape. 4. The number of samples of otolith shape from different origins needs to be balanced in order to secure a proper statistical analysis. Furthermore the subtle differences between otoliths suggest that sample size should exceed 1000 specimen in order to be representative. 5. Otolith shape appears to vary between seasons. Reference samples of otolith shape therefore need to be taken in the same time period as the samples that need to be discriminated. 6. Otoliths are readily available in large numbers from practically all fishery labs around the NE Atlantic. The otolith shapes are fairly easy to determine by standard image systems. Therefore investigations that need to discriminate between stocks and detect the harvest origin of individual cod can be made swiftly and with a relatively low budget compared to other discriminating methods. 7. Fishery managers and legislators are often more interested in fish that may derive from only two areas. Thus, investigations of fish origin are usually initiated when a fisherman are accused of fishing in closed areas but claim that he has been fishing in an adjacent area. This study suggests that otolith shape can predict the fish origin between two areas with a high accuracy.

A database consisting of n = 36.904 morphometric mesurements for n = 1.318 individuals is utilized to characterize the external morphology of the Atlantic cod, and its differentiation during ontogenetic growth and within the geographical range of the species in European waters. The size-related allometry is investigated for every one out of 22 variables from a truss network. A linear model confirms that both life age and geographical origin influence the body shape of cod significantly, the former being more important than the latter. Principal component extraction in factor analysis groups the geographical test groups according to the spatial location of the collection sites, but only when considering one principal component out of five that explains 2.5% of the morphometric variance. Body size variance alone, chiefly due to growth, accounts for 95.5% of the morphological variation. Although the geographical component of shape poylmorphism is rather low, as many as 96.4% of the female cod, and 94.98% of the males, could be assigned as individuals to originate from one out of eight sea basins. Cod of five years of age, or older, are assigned with 100% precision to their origins, whereas two or three years old cod produce somewhat lower assignment rates. Cod from certain geographical populations are assigned with greater precision than are others. Morphometric data that had been standardized by means of allometric growth functions for every size variable alone-yielded less satisfactory assignment rates than did size-adjusted data that were in addition analysed separately for every single year class. The dual control of ontogenetic shape transformation by (i) estimating the fishes' life age from otolith diagnosis and (ii) size-adjustment of raw data is proposed in order to assign individuals to geographical populations. If life ages are known, year classes permit geographical assignment even without size-standards. Within the truss system of measurements taken in every cod specimen, most body size variables are intercorrelated, but orbital diameter and occiput length retain a certain independence of the remaining measurements.

As expected, cod from farms could not be allocated to harvest location based on studies of parasite assemblages. Due to artificial feeding these cod are seldom infected with parasites transmitted through the food web. However, parasite evidence can still be used to discriminate cod from farms or wild locations, since few cod from farms were infected with parasites, whereas only two wild cod (from the Baltic) were devoid of parasites. Although 61 parasite species were found, a number between 10 and 15 species seemed reasonable in terms of cost-benefit to allocate cod to five or six harvest locations since the addition of more species to the models did not improve allocation results substantially. The suite of predictor parasite species varied between datasets, but eight species were selected in all of them, and one additional species was chosen in three of the four datasets. These nine species are helminths (two trematodes, six nematodes and one acanthocephalan), which are commonly found in either the digestive tract or the visceral cavity of cod in European waters. None of the species with prevalence equal or less than 10% was useful as predictor of harvest location. In practical terms, this means that focussing on collecting relatively common species is enough to develop methods to establish harvest location of cod in European waters. The analyses suggested better classification ability of both predictive models (BBP network and LDA) when the season samples were analysed separately, since their performance increased correct classification rates of cod to harvest location by about 5%. This result suggests that a merged data set is probably too noisy to provide satisfactory classification results. In the second progress report of CODTRACE, large variation in parasite abundances between seasons was reported, but no consistent patterns accounting for such variation could be found. Further implementation of the method in a fishery management context would require steady sampling in designated areas and constant update of the predictive models. All predictive models could assign cod from the Baltic Sea, Iceland and Norway to their respective harvest location efficiently (over 95% for the Baltic and Norwegian cod and about 90% for the Icelandic cod), but failed to deliver similar correct allocation rates in cod from the Celtic, Irish and North Seas. Unfortunately the datasets, except Autumn 2002, were unbalanced (some localities included more cod than others), being cod was from the Celtic and North Seas (Spring 2002 dataset) and Irish Sea (Spring 2003 dataset) underrepresented in the learning processes (Table 1). Since both the BBP network and LDA's ability to correctly classify cases depends on the number of examples for each category used in the training process, additional data from the Celtic, Irish and North Sea might probably improve the correct classification rates in the test sets. Unbalanced samples might also account for the apparently erratic behaviour of the correctly classified cod in the test set of cod from the Celtic Sea, ranging from as bad as 57% in the most unbalanced data set (Spring 2002) to 92% in the balanced Autumn 2002 dataset (Table 2). Note also the better overall performance of the BBP network with the balanced Autumn 2002 dataset that probably illustrate the full potential of the technique if sample sizes of each category are similar in model training. Although the accuracy of the BBP network was similar to LDA, it is possible that the accuracy of the neural network could improve with larger sample sizes. Neural networks have traditionally been used to decipher and classify very subtle patterns in large datasets. The advantage being that the more "experienced" they become, the better they are at detecting patterns. This ability would be an asset in regular sampling programmes, greater resolution potentially arising out of their application to increasingly large amounts of data. However, room for prediction improvement is probably constrained by similarities in parasite assemblages of cod, particularly between the Irish, Celtic and North Sea. The fact that about 10% of the fish analysed was misallocated by two statistical methods based on quite different assumptions is a strong suggestion of an underlying unpredictability in the composition of parasite assemblages of cod that cannot solved by the sophistication of the statistical method applied. However, this report shows that even if this possibility is confirmed, analyses of parasite assemblages are useful to predict harvest location of Baltic, Icelandic or Norwegian cod and might still be crucial used in combination with other biomarker techniques developed under the CODTRACE project.

Three hundred seventy-two variables were extracted from each of the 1100 individual cod collected for optimising the techniques from wild populations and from two cod farms. An additional 220 cod were collected in a double-blind manner to independently test the accuracy of the techniques. Three changes were made in optimising the CODTRACE techniques. Allozymes could not be uniformly extracted across individuals producing too space of a data set; ultimately this technique was not useful for traceability of cod in this study. The data from otolith core and edge microchemistry was combined given the broad similarity of the data produced; this proved more effective towards traceability than using each of the data sets in isolation. As the specific DNA locus (Syp I) proved only useful for isolating the Icelandic populations from others, and this information was largely redundant to the microsatellite date, it was not necessary to use this locus data. The limits of optimisation (highest percent correct classification) for each technique were determined, and across seasons and locations the highest correct placement was as follows: body morphometry (84.7%), parasite assemblage (80.4%), bacterial assemblage (78.3%), microsatellite loci (70.1%), otolith morphometry (68.0%), and otolith microchemistry (66.9%). This successfully accomplished the objective of optimising each technique; the absolute classification limit for each technique was established, and no individual technique was able to confer 100% classification (perfect traceability) upon the collected cod. Another objective was to optimise the method for combining analytical and statistical results into a more powerful classification tool. For the combined analyses, the size of large resultant data set needed to be reduced to allow for statistical confidence in the results. The data generated by each technique were reduced to probabilities representing the likelihood of each cod belonging to each source population, in two steps. First, each cod was classified as being of wild or farmed origin using logistic regression of individuals known to belong to wild or farmed populations. Second, if a fish was classified as being of wild origin, it was assigned to one of the five wild populations using multinomial logistic regression. This step was conducted separated for cod classified as being of farmed origin, using logistic regression for assignment. Using the same multivariate model for all seasons and years sampled, 100% of cod were correctly placed to their population of origin.

Searching for OpenAIRE data...

Deliverables

Share this page

Download