Comparison of two methods for distinguishing wild from hatchery-reared salmon (Salmo salar Linnaeus, 1758) in the Baltic Sea

Transcription

1 ICES Journal of Marine Science, 55: Article No. jm Comparison of two methods for distinguishing wild from hatchery-reared salmon (Salmo salar Linnaeus, 1758) in the Baltic Sea P. Hiilivirta, E. Ikonen, and J. Lappalainen Hiilivirta, P., Ikonen, E., and Lappalainen, J Comparison of two methods for distinguishing wild from hatchery-reared salmon (Salmo salar Linnaeus, 1758) in the Baltic Sea. ICES Journal of Marine Science, 55: A discriminant function based on scale characters was used for distinguishing wild and hatchery-reared Baltic salmon. The function was based on the following three variables: the width of the first freshwater annual zone; the maximum number of circuli per year in the freshwater zone; and the mean distance between circuli in the freshwater zone. Using this function, it was possible to correctly classify 92.63% of fish of known origin. Use of the so-called visual method resulted in a 100% correct classification of the same sample. When the two methods were compared using a scale sample of unknown origin, the results showed that the two methods classified 219 and 257 fish the same. The suitability of the methods for stock identification and the discrepancies between them are discussed International Council for the Exploration of the Sea Key words: Salmo salar, stock identification, discriminant function, scale pattern. Received 10 June 1992; accepted 23 April P. Hiilivirta: Deceased. E. Ikonen: Finnish Game and Fisheries Research Institute, PO Box 6, FIN Helsinki, Finland. J. Lappalainen: Department of Limnology and Environmental Protection, PO Box 27, FIN University of Helsinki, Finland. Correspondence to E. Ikonen: Fax: ; erkki.ikonen@rktl.fi Introduction Various techniques have been used to identify Atlantic salmon (Salmo salar) stocks. Discriminant analysis using such characters as the river age, the numbers of circuli in the second annual river zone and the first annual sea zone, and the anterior radii of these zones, has been used to discriminate between salmon originating from North America and Europe (Lear and Sandeman, 1980). Reddin and Burfitt (1983) used similar characters but separated the counts of circuli in the winter and summer growth portions of the first sea year. European salmon stocks have also been discriminated using scale shape measurements (de Pontual and Prouzet, 1986a,b). In Norway, salmon stock identification based on scale characters, deformation of fins, and pigment analysis is important in assessing the amount of escapees from fish farms (Anon., 1991a; Lund and Hansen, 1991). In the Baltic Sea area the main problem has been to distinguish wild from hatchery-reared salmon. This subject is very important because the wild salmon stocks are decreasing due to overfishing. Antere and Ikonen (1983) describe a method to visually distinguish the stocks on the basis of the structure of the freshwater zone of the scale. Sych and Tuszynska (1983) introduced the use of discriminant analysis for distinguishing stocks of Baltic salmon. Borzecka et al. (1990) used discriminant analysis based on the following characters: the number of circuli in the freshwater annual zones and the width of these zones; the number of circuli in the first annual zone in the sea zone; and the width of this zone, and the freshwater age of the fish. The aim of this study is to compare the visual method (Antere and Ikonen, 1983) with the discriminant function method. This is done in two phases. In the first phase, the visual and the discriminant function methods are tested with scales of known origin, in the second phase, both methods are tested using a routine catch sample from the Bothnian Bay. Materials and methods Scales and scale reading Most of the scales used in this study were obtained from commercial fishermen. Instructions had been given to /98/ $30.00/ International Council for the Exploration of the Sea

2 982 P. Hiilivirta et al. 12 E R. Simojoki R. Iijoki 65 North Montaja Laitakari 31 R. Oulujoki N km Figure 1. ICES Subdivisions and the localities. the fishermen where the scale samples should be taken, but it is possible that some scales were not taken from the right location. The possible error was minimized by excluding scales taken from the lateral line and regenerated or abnormal scales. Impressions of the scales were made on polycarbonate strips using a mechanical press. For discriminant analysis, scales were examined with a computer-aided scalereading apparatus (henceforth called CSR, for details see Ikonen et al., 1993). The measurements and circuli counts were made from the focus along the longest axis of the scale (this was not always possible due to flaws in the impressions). All the measurements were made with the precision of mm depending on the magnification used. For visual classification, the criteria used are those presented by Antere and Ikonen (1983). Calibration data The material on which the discriminant function is based consists of 214 hatchery-reared and 183 wild fish. The hatchery-reared fish originated from the stocks of the rivers Oulujoki (n=100) and Iijoki (n=114) (Fig. 1). All of these fish were tagged, i.e. their origin was known with certainty. The origin of wild fish was known with reasonable certainty: either the fish had been tagged as smolts, or the scales had been collected during the smolt run. All these fish belonged to the river Simojoki stock. Only characters in the freshwater zone of the scales were chosen as the variables to be used in the discriminant analysis, because scales of smolts were also included in the calibration material. A correlation analysis was applied to find uncorrelated variables to be used

3 Distinguishing wild from hatchery-reared salmon 983 Table 1. The Pearson correlation coefficitions between scale variables (n=397). fw1 fw2 fwc1 fwc2 fwcmax fwcmin fw fwc fwc fwcmax fwcmin cmean Variables: fw1, the width of the first freshwater annual zone in mm; fw2, the width of the second freshwater annual zone; fwc1, the number of circuli in the first freshwater annual zone; fwc2, the number of circuli in the second freshwater annual zone; fwcmax, the maximum number of circuli per year in the freshwater zone; fwcmin, the minimum number of circuli per year in the freshwater zone; cmean, the mean distance between circuli in the freshwater zone (the total number of circuli in the freshwater zone/fwcmax). in a stepwise discriminant analysis (SAS, 1988). Those variables which had the greatest discriminating power in the stepwise discriminant analysis were chosen to be used in a leave-one-out discriminant analysis (Lachenbruch, 1975; SAS, 1988). Comparison of the visual method and discriminant analysis using scales of known origin (Test I) In order to compare the visual method and the discriminant function method, a blind test was made where scales of fish of known origin were analysed by the same scale reader using both methods. The reader did not know the origin of the fish. The scales used were from different fish than those used to develop the discriminant function. For the test, 138 scale impressions were randomly chosen by a person not participating in the scale reading. The impressions were numbered using a table of random numbers. Of these, 43 had to be discarded owing to regenerated scales, poor quality of the impressions, or similar reasons. Thus the final test material consisted of 95 impressions, of which 51 were of hatchery and 44 of wild origin. The scales of the hatchery-reared fish were collected in connection with tag returns. Of these, 47 belonged to the river Iijoki stock (stocking age 3 years) and four to the river Oulujoki stock (stocking age 2 years). The fish were tagged in the years and in The scales of wild fish (n=44) used in this test originated from the Merikoski rapids in the river Oulujoki running into the Bothnian Bay. The scales had been collected by T. H. Järvi in Considering the situation in post-war Finland, it is quite reasonable to assume that these fish were of wild origin. The origin of the fish was determined visually using a Bell & Howell SR-VIII microfilm reading device (magnification ca ). After this, the scale reader examined the scale impressions with the CSR recording the scale data. Comparison of the visual method and discriminant analysis using a routine catch sample (Test II) The discriminant function was compared with the visual method using part of a routine catch sample from the Bothnian Bay near the mouth of the river Simojoki (isle of Montaja, Fig. 1). The material consisted of scale samples from 257 fishes collected in 1990 during the following periods: 28 May to 6 June, 23 June to 30 June and 5 July to 28 July. These periods were chosen to ensure the inclusion of wild fish in the material. As hatchery-reared fish have a tendency to run later in the season and because a relatively large part of the material consists of fish running early, the material does not give a true picture of the proportions of wild and hatcheryreared salmon. Adipose-fin-clipped fish were excluded from the analysis for two reasons. Firstly, the freshwater zones of scales of fish stocked when 1-year-old tend to be almost identical to those of scales of wild fish and, secondly, the origin of adipose-fin-clipped fish is known anyhow. The material was analysed as in Test I, except that the visual examination of the scales was conducted from the video monitor while recording the scale data with the CSR. Results Discrimination analysis of calibration data Only three scale variables showed correlation coefficients less than 0.5 ( r >0.5): the mean distance between circuli in the freshwater zone (=cmean), the maximum number of circuli per year in the freshwater zone (=fwcmax) and the width of the first freshwater annual

4 984 P. Hiilivirta et al. Table 2. Results of the stepwise discriminant analysis. Step Variable Entered Removed Number In Partial R2 F statistic Prob> F 1 cmean fwcmax fw Table 3. The results of the leave-one-out discriminant analysis using variables fw1, fwxmax, and cmean. Table 4. Results of the discriminant analysis of fish of known origin (Test I). Actual group Groups into which the fish were classified Hatchery-reared Actual group Groups into which the fish were classified Hatchery-reared Hatchery-reared n % n % n % Hatchery-reared n % n % n % zone (=fw1) (Table 1). These variables were used in the stepwise discriminant analysis to find the best discriminating variables between hatchery-reared and wild stocks. The fwcmax had the greatest discriminating power (Table 2), but since none of the three variables were removed (Wilks lambda, p<0.15) from the stepwise discriminant analysis, they were all used in a leave-one-out discriminant analysis. The leave-one-out discriminant analysis revealed that with the variables fw1, fwcmax, and cmean, 85.05% of the hatchery-reared salmon were classified correctly as were 90.71% of the wild (Table 3). The error rate of the calibration data was 12.34%. The error rate in this study is defined as the proportion of misclassified samples over both groups (Lachenbruch, 1975). The misclassification rate is defined as the percentage of misclassified salmon within a group. Thus the linear discriminant functions are: = *fwcmax *fw *cmean Hatchery-reared= *fwcmax *fw *cmean To determine the origin of a fish, one places the measured values of the variables in the functions. The fish belongs to the group the function of which gives the higher value. Table 5. Comparison of the discriminant analysis and visual methods using a routine catch sample. Classification by the visual method Classification by discriminant analysis Hatcheryreared Hatchery-reared n % n % n % Comparison of methods (Tests I and II) In Test I, using the visual method, all of the fish of known origin were classified correctly. The discriminant analysis correctly classified 92.63% of the fish of known origin (Table 4). Three hatchery-reared fish were incorrectly classified wild, and four wild fish were incorrectly classified hatchery-reared. On the basis of this test it seems that the discriminant function may slightly underestimate the proportion of wild fish.

5 Distinguishing wild from hatchery-reared salmon 985 In Test II, visual classification and discriminant analysis gave different origins for 38 fish of the routine catch sample (Table 5): 31 fish classified hatchery-reared by the visual method were classified wild by the CSR, and seven fish classified wild by the visual method were classified hatchery-reared by the CSR. The percentage of different classifications is 14.79% (38 differently classified fish out of 257) when the whole material is concerned. If, on the basis of Test I, it was considered that the visual method gives the right result, the differences were and 14.89% for hatchery-reared and wild fish, respectively. In this case the discriminant function method tends to overestimate the proportion of wild fish in comparison with the visual method. Discussion Antere and Ikonen (1983) show that it is possible to visually distinguish wild and hatchery-reared salmon on the basis of the structure of the freshwater zone of the scales. The use of this method requires no special equipment, but it does include a subjective element even if a skilful scale reader is employed. On the other hand, discriminant analysis provides its user with functions which can be used to discriminate between wild and hatchery-reared salmon even by a person with limited experience in scale reading. Using the scale data on which the discriminant function derived in this study is based, the error rate is 12.34%, which is of the same order of magnitude as the error rates obtained in other studies (Lear and Sandeman, 1980; Reddin and Burfitt, 1983; Borzecka et al., 1990; Ikonen et al., 1993). Besides, the misclassification rates of wild and hatchery-reared fish were almost equal, being 9.29 and 14.95%, respectively. This is more important than the reduction of the error rate, especially since the ratio of wild salmon to hatcheryreared salmon in the Baltic Sea is approximately 1:10 (Anon., 1991b). An example will illustrate the situation. A sample containing 100 wild and 1000 hatchery-reared salmon is analysed using the new discriminant function. If the fish are misclassified according to the above mentioned misclassification rates of 9.29 and 14.95%, 241 fishes would be classified wild and 859 hatcheryreared. If, on the other hand, the misclassification rates were 7.23 and 21.38%, as in Ikonen et al. (1993), 307 fishes would be classified wild and 793 hatchery-reared. Thus, the relative abundances are estimated more accurately with the discriminant function obtained in this study. The results using scales of known origin indicate that the differences between the visual and discriminant function method are quite small, the misclassification rates being 0 and 7.37%, respectively. Compared with the error rate (12.34%) of the basic material these figures seem rather low. The reasons for the better discrimination of the material used in Test I using scales of known origin are not clear, but one reason may be the greater homogeneity of the material, which contained only fish of northern stocks. In this connection, it must be noted that the sea age of the fish from the Merikoski Rapids in Test I was mostly 3 or 4 years. In contrast to this, the sea age of the hatchery-reared fish in Test I was mostly 1 or 2 years. In spite of the fact that the appearances of the Merikoski scales were very typically wild, the difference in the sea ages may have given an advantage to the scale reader. However, this advantage has apparently been of little importance, since the discriminant analysis misclassified only four (=9.09%) of the 44 Merikoski fishes. The results of Test II using a routine catch sample are more difficult to interpret because the origin of the fish is unknown. It seems, however, that no great discrepancies between the two methods exist, since the overall difference between them was no more than 14.79%. One reason for the greater discrepancy in this case may be the fact that the stocking age of hatchery-reared fish is sometimes very difficult to determine, due to the irregular structure of the freshwater zone of the scale. It is virtually certain that the proportion of hatchery-reared 3-year-old fish was overestimated: only about 1 2% of the salmon stocked in Finland consists of 3-year-old smolts (Anon., 1986; Anon., 1988; Anon., 1989; Anon., 1991c). In the catch sample, 8.1% of the fish visually classified hatchery-reared were considered to be 3 years old when stocked. It is conceivable that some of these fish have been classified erroneously wild by the discriminant analysis, because of the lower counts of circuli and smaller widths of annual zones in the freshwater zone, resulting from the overestimation of the stocking age. Some error may also have been due to the fact that it was not always possible to measure the annual zones and count the circuli along the longest axis of the scale. Also this may have resulted in underestimating the widths of the freshwater annual zones and in overestimating the proportion of wild fish. The possible use of non-preferred scales may also have had some influence on the results, but its significance is very difficult to assess. Summary On the basis of the present study it seems that the visual method is quite suitable for the discrimination of wild and hatchery-reared salmon in the northern parts of the Baltic Sea. However, in some instances it may have drawbacks that can be avoided by the use of discriminant analysis. The present results indicate that it is possible to derive a discriminant function which distinguishes wild and hatchery-reared salmon in the

6 986 P. Hiilivirta et al. above mentioned area almost as accurately as the visual method. It seems reasonable to believe that the accuracy of the discriminant function method may be increased to the level where its applicability may even exceed that of the visual method, at least in situations where scale readers of limited experience have to be employed. Acknowledgements The authors thank Dr S. E. Campana, Dr D. G. Reddin, Mr Pekka Vakkari and Mr Mikael Hildén for comments on the manuscript. Mr Pekka Vakkari also gave us valuable assistance with statistical methods. Mr Jonathan Hutchins kindly checked our English. References Anon Statistics on fishing and fish farming in Finland in Suomen Kalatalous, 52: Anon Statistics on fishing and fish farming in Finland in Suomen Kalatalous, 54: Anon Statistics on fishing and fish farming in Finland in Suomen Kalatalous, 55: Anon. 1991a. Report of the Workshop on Identification of Fish Farm Escapees and Salmon. ICES CM 1991/M: 2, 33 pp. Anon. 1991b. Report of the Baltic Salmon and Trout Assessment Working Group. ICES CM 1991/Assess: 13, 99 pp. Anon. 1991c. Statistics on fishing and fish farming in Finland in Suomen Kalatalous, 58: Antere, I., and Ikonen, E A method for distinguishing wild salmon from those originating from fish farms on the basis of scale structure. ICES CM 1983/M: 26, 7 pp. Borzecka, I., Ikonen, E., and Torvi, I Attempts of using the discriminant function based on scale structure of Baltic salmon to distinguish between the wild and hatchery-reared smolts. ICES CM 1990/M: 21, 7 pp. Ikonen, E., Hiilivirta, P., and Lappalainen, J (eds). Baltic Salmon Scale Reading. Report of the Baltic Salmon Scale Reading Workshop. ICES Cooperative Research Report, 192: Lachenbruch, P. A Discriminant analysis. Hafner Press, New York. 129 pp. Lear, W., and Sandeman, E Use of scale characteristics and discriminant functions for identifying continental origin of Atlantic salmon. In ICES/ICNAF Joint Investigations on North Atlantic salmon. Rapports et Procès-Verbaux des Réunion du Conseil International pour l Exploration de la Mer, 176: Lund, R. A., and Hansen, L. P Identification of wild and reared Atlantic salmon, Salmo salar L., using scale characters. Aquaculture and Fisheries Management, 22 (4): Pontual, H. de, and Prouzet, P. 1986a. Atlantic salmon stocks discrimination by scale shape analysis. ICES CM 1986/M: 11, 14 pp. Pontual, H. de, and Prouzet, P. 1986b. Scale shape measurements to discriminate between Atlantic salmon stocks. ICES CM 1986/M: 12, 18 pp. Reddin, D., and Burfitt, R An Update: The use of scale characters and multivariate analysis to discriminate between Atlantic salmon (Salmo salar L.) of North American and European Origin Caught at West Greenland. ICES CM 1983/M: 11, 14 pp. SAS SAS/STAT User s Guide, Release 6.03 Edition. Cary, NC: SAS Institute Inc pp. Sych, R., and Tuszynska, L Attempts of using the scale characteristics for separation of some Baltic Salmon and Sea Trout stocks. ICES, CM 1983/M: 29, Appendix 6, 37 pp.