Migratory connectivity of two Baltic Sea salmon populations : retrospective analysis using stable isotopes of scales

Jyrki Torniainen1*, Pekka J. Vuorinen2, Roger I. Jones1, Marja Keinänen2, Stefan Palm3, Kristiina A. M. Vuori4, and Mikko Kiljunen1 Department of Biological and Environmental Science, University of Jyväskylä, PO Box 35, FI-40014, Jyväskylä, Finland Finnish Game and Fisheries Research Institute, PO Box 2, FI-00791, Helsinki, Finland Institute of Freshwater Research, Department of Freshwater Research, Swedish University of Agricultural Sciences, Stångholmsvägen 2, S-17893, Drottningholm, Sweden Department of Biology, University of Turku, FI-20014, Turku, Finland *Corresponding Author: tel: +358 408 053 847; fax: +358 146 172 39; e-mail: jyrki.t.torniainen@student.jyu.fi


Introduction
Migratory connectivity refers to the extent to which individuals of a migratory population behave in unison (Webster et al., 2002;Hobson and Norris, 2008).Strong migratory connectivity exists when the individuals of a population bred in one area all migrate to the same non-breeding area.In contrast, wide dispersal of individuals in the non-breeding regions indicates weak migratory connectivity.The strength of migratory connectivity has significant consequences for the ecology, evolution and conservation of migratory animals (Esler, 2000;Martin et al., 2007).
Populations of anadromous Atlantic salmon (Salmo salar L.) are river-specific with respect to spawning, but their feeding migration is thought to vary on a year-to-year basis (Ikonen, 2006;Thorstad et al., 2011).However, based on tag-recovery studies, populationspecific annual migratory connectivity of salmon in the Baltic Sea has been considered rather fixed (Aro, 1989;Ikonen, 2006).Salmon from most of the northern rivers mainly utilize the southern main basin of the Baltic Sea as their principal feeding area (Salminen et al., 1994;Ikonen, 2006;Jutila, 2008).In contrast, many salmon originating from rivers flowing into the Gulf of Finland remain there to feed and show reduced migration activity, although appreciable numbers do migrate farther to the Baltic Proper (Ikonen, 2006).Differences in migratory behaviour are mainly due to the evolutionary histories of the populations (Kallio-Nyberg and Ikonen, 1992).
Atlantic salmon is an economically and ecologically valuable fish species (Kulmala et al., 2012).Many salmon populations have faced adverse environmental impacts during the last century (Mills, 2003), mainly affecting the river phase of the salmon life cycle.Degradation of spawning rivers has led to fewer wild young salmon entering the sea, and consequently many salmon populations have been compensated through stocking with hatchery-reared salmon.However, some studies have indicated that stocked and wild individuals do not necessarily display similar migratory behaviour (Salminen et al., 1994;Kallio-Nyberg et al., 1999, 2006).Better understanding of the migratory connectivity of salmon of different origin could allow commercial fishing in sea areas to be better targeted in order to promote conservation of endangered wild stocks.
Tag-recovery has been widely used to study fish movements, but is an inefficient tool to study migration patterns in sufficient detail to conserve stocks.This method underestimates the movements of fish due to limited recapture effort (Lucas and Baras, 2000).Tagged salmon are usually captured only once so conclusions about migration movements are based on just two observations.Intrinsic markers such as stable isotopes (SIs) have been widely used to study avian (e.g.Hobson and Wassenaar, 1996;Wassenaar and Hobson, 2000;Hobson et al., 2004) and insect (Hobson et al., 2012) migrations, although applications to aquatic organisms are increasing (Best and Schell, 1996;Me ´nard et al., 2007;Dempson et al., 2009;MacKenzie et al., 2011a;Zbinden et al., 2011;Quinn et al., 2012).Crucially, the stability of the isotopic value in many tissues and structures allows ecological topics to be studied using archived sample material when available (Dawson and Siegwolf, 2007).Teleost fish scales offer valuable opportunities to study the ecology of fish via stable isotope analysis (SIA) (Hutchinson and Trueman, 2006;MacKenzie et al., 2011aMacKenzie et al., , 2011b)).Moreover, incrementally growing scales offer the possibility of obtaining ecological information such as changes in diet and location of fish over time (Perga and Gerdeaux, 2003;Pruell et al., 2003).
We conducted SIA of carbon (C) and nitrogen (N) from the last fast-growth region of scales (the outermost set of widely spaced circuli representing the last phase of marine growth) from Atlantic salmon caught ascending the River Simojoki in northern Finland and the River Kymijoki in southeastern Finland after final feeding in an unknown area, and compared the values with those from scales of salmon caught during the previous winter from three defined Baltic Sea areas during the years 1989-2011 (reference salmon).Our main aim was to identify the most likely final feeding areas of salmon before ascending their natal river and any annual variation in the use of these areas, and hence to assess the migratory connectivity within these salmon populations, using a novel method based on salmon scale isotopic composition.A second aim was to test differences in migratory connectivity between wild and stocked salmon from the R. Simojoki.We also made a methodological comparison between SIA-based assignment and Carlin-tagged and recaptured salmon.

Study site
The Baltic Sea is the largest and best-studied brackish water basin in the world (Lass and Mattha ¨us, 2008;Ojaveer et al., 2010).The progressive north to south increase in marine influence (Kullenberg, 1981) leads to divergence of stable isotope ratios (C and N in this study) between different sea areas (e.g.Rolff and Elmgren, 2000;Kiljunen et al., 2008), allowing inferences regarding salmon feeding through the sea.The Baltic Sea has been divided into several subdivisions (SubDivs) for fish catch statistics according to the International Council for the Exploration of the Sea (ICES; Figure 1).For this study the relevant SubDivs are the Baltic Proper [SubDivs 25, 26, and 28 (excluding the Gulf of Riga)], the Bothnian Sea (SubDiv 30) and the Gulf of Finland (SubDiv 32), hereafter referred to respectively as BPr, BS and GoF.Due to the SubDiv-classification of recaptured salmon (reference salmon) caught from the sea, these SubDivs define the areal segregation and accuracy of possible Baltic Sea salmon feeding areas in this study.
The R. Simojoki flows into the northern end of the Gulf of Bothnia (Figure 1) and is one of the few northern Finnish rivers that still support a wild migratory salmon stock; many other large rivers in the area have been dammed and regulated to the extent that their salmon stocks have been lost.The R. Kymijoki flows into the eastern part of GoF and is almost exclusively reliant on salmon introductions because the original stock was destroyed by damming in the early 1950s.Since 1995 there has been evidence of natural reproduction with annual smolt production of ca.Baltic salmon migratory connectivity 20 000 compared with the annual stocking of ca.600 000 smolts (ICES, 2012a).

Acquiring salmon scales
Scales were from salmon caught between 1989 and 2011.One set of scales included 675 female salmon caught for the M74 monitoring programme of the Finnish Game and Fisheries Research Institute (FGFRI) when ascending the R. Simojoki and the R. Kymijoki (Table 1).The second set of scales included 1195 reference salmon (sex not known) individuals (Table 1) in their feeding phase caught from five different ICES SubDivs: 25, 26, 28, 30 and 32 (Figure 1).These scales of the second set were from the scale archives of the FGFRI, the University of Turku and the Swedish Institute of Freshwater Research.Since marked differences in scale isotope composition were not found between SubDivs 25, 26 and 28, these were combined and represent BPr, while SubDiv 30 represents BS and SubDiv 32 GoF (Table 1).Scales of salmon from the feeding phase provide a geographical reference against which data from salmon ascending the R. Simojoki can be compared for each year.To ensure correct sea area for reference SI values, only scales from feeding phase fish collected from 1 September to 31 March in each winter were used in this study.One-sea-winter salmon and multispawners may retain traces of riverine SI values in their scales and hence were excluded from the study, and only two-sea-winter salmon or older were accepted as reference material.Origin of salmon (wild or hatchery-reared) was determined from the scale nucleus (Hiilivirta et al., 1998).

Salmon Carlin-tag recovery data
Carlin-tag recovery data for salmon were provided by FGFRI.The timescale was divided into two parts: the years 1989-1998 are referred to as the 1990s and the years 2004-2011 as the 2000s.The data used represent recoveries (years 1989-2009) of wild salmon tagged as smolts with Carlin-tags after being trapped during their downward run or of hatchery-reared salmon on release at the R. Simojoki mouth (n 1990s ¼ 341, n 2000s ¼ 46) and of salmon stocked at the R. Kymijoki mouth (n 2000s ¼ 51), which were caught from the same sea areas of the Baltic Sea as the SIA reference salmon.The recoveries used are from the same months as the reference salmon scale data.

Sample preparation and SIA
Scale growth is mainly radial beginning from the scale nucleus where narrow circuli are laid down in wintertime, and wider circuli correspond to summertime growth (Hutchinson and Trueman, 2006).The chemical composition (in this study the C and N isotopic composition) of the outermost growth region (formed by the outermost widely spaced circuli) yields information about the diet of an individual during the final marine feeding phase (summer).However, formation of a new annual collagen -bioapatite layer below the previous layer prevents the use of earlier, inner scale growth rings (Hutchinson and Trueman, 2006).As a pre-treatment for SIA, scales were soaked in deionized water, cleaned with fine non-lint paper tissue to remove tissue other than scale material (e.g.mucus, pigment, guanine, adhered paper).The last year growth region was determined using a microfilm reader and was cut using a scalpel against a glass edge to sample a minimum of 0.2 mg of scale tissue.Although salmon do not exhibit somatic growth during the spawning migration, some minor extra growth on the scale edge was occasionally observed.This additional growth was removed.To prevent bias in SI values from extraneous carbonates, scale material was acidified for 2 min in 1.2 N HCl, rinsed five Reference salmon are from three different feeding areas (BS ¼ Bothnian Sea, BPr ¼ Baltic Proper, GoF ¼ Gulf of Finland).Ascending salmon (wild, reared or unidentified) of the rivers Simojoki and Kymijoki were caught from the river estuary on route to their spawning sites.
times in deionized water and dried overnight at 608C (Perga and Gerdeaux, 2003).SIAwas performed at the University of Jyva ¨skyla ¨, using a FlashEA 1112 Elementar Analyzer connected to a Thermo Finnigan DELTA plus Advantage mass spectrometer (Thermo Electron Corporation, Waltham, MA, USA).Pike (Esox lucius L.) white muscle tissue was used as an internal working standard.Results are expressed using the standard d notation (d 13 C, d 15 N) as parts per thousand (‰) differences from the international standard.The reference materials used were International Atomic Energy Agency (IAEA) standards of known relation to the international standards of Vienna Pee Dee Belemnite (for carbon) and atmospheric N 2 (for nitrogen).Precision for each run was better than 0.35‰ for C and 0.20‰ for N based on the standard deviation of replicates of the internal working standards.Sample analysis also yielded percentage of carbon (%C) and nitrogen content (%N) of samples from which C:N ratios (by weight) were derived.These additional data from SIA were included in statistical analyses as they improved the precision of assignment to sea feeding areas.
Predatory fish in particular show a positive relationship between fish size and trophic level (Romanuk et al., 2011).Therefore larger salmon tend to exhibit higher d 15 N values (Satterfield and Finney, 2002;MacKenzie et al., 2011aMacKenzie et al., , 2011b;;Trueman et al., 2012), which could lead to incorrect assignment to sea areas exhibiting higher d 15 N values.This is especially the case when river-ascending salmon tend to be on average larger than salmon in the feeding phase.In contrast, d 13 C values in fish tissues do not show this bias (Vander Zanden and Rasmussen, 1999).To adjust for size effects, d 15 N values were therefore standardized to salmon lengths for each annual salmon dataset when linear regression (Y ¼ a + bX) between length (X) and d 15 N (Y) value was statistically significant (years 1989-1992, 1997, 2004-2005 and 2009).

Statistical analyses
Due to a lack of reference scales from the years 1999-2003, these years were excluded from the discriminant analysis (DA).Multivariate analysis of variance (MANOVA) was used to test SI differences between sea area SubDivs and years.DA was applied to estimate the most likely final sea feeding area for each salmon caught ascending the R. Simojoki and the R. Kymijoki.For each year the ascending salmon SI values were compared (via DA) to previous wintertime reference salmon scale SI values to ensure correct assignment to sea feeding area.Scales were also missing from BS for the years 1996, 1998, 2004, 2006, 2008 and 2011, from BPr for 1998 and from GoF for 2005 and 2006.For those years, previous and/ or following year scale reference material was used.As with the tag-recovery data, the SIA dataset was divided into two sets: decades 1990s and 2000s.A stepwise method was used for DA with variables d 13 C, d 15 N, %C, %N and C:N (Table 2).The criterion for including a variable was Wilks' l (F ≥ 3.84).The stepwise approach minimizes the multicollinearity occurrence and between variable correlations; tolerances of the variables entered at each step were checked (tolerance .0.4) for confirmation.Annual average probabilities were calculated for each year from ascending salmon individuals as an indicator of migratory connectivity (Webster et al., 2002).Differences in size between ascending and sea-phase salmon were evaluated with the Kruskal-Wallis non-parametric test.Kruskal-Wallis and Bonferroni-corrected Mann-Whitney non-parametric tests were conducted to test differences in salmon feeding area probabilities within and between the 1990s and 2000s to assess possible differences in migratory connectivity within the record.Possible differences between wild and hatchery-reared salmon with respect to the most likely last feeding area distribution were tested using the Pearson X 2 test.Possible change in d 13 C and d 15 N through the timespan was tested via linear regression.All statistical analyses were performed using PASW Statistics 18 for Windows (SPSS Inc., Chicago, IL, USA).

Isotope values of salmon scales
Differences in isotope values between salmon scales from the three sea areas were greater for d 15 N than for d 13 C (Figure 2).Values were generally highest for GoF and lowest for BS, with intermediate values from BPr. Fluctuations in both d 13 C and d 15 N values were apparent over the 22-year period, particularly for BS.Overall, d 13 C values became more depleted towards the end of the timescale (linear regression, r 2 ¼ 0.22, p , 0.001, n ¼ 1880) and the same was true (but less markedly) for d 15 N values (linear regression, r 2 ¼ 0.023, p , 0.001, n ¼ 1880).Standard deviations of both d 13 C and d 15 N values of the reference salmon generally decreased (although not statistically significantly) towards the end of the 2000s, leading to a better areal discrimination.Both d 13 C and d 15 N values of salmon from the R. Simojoki mostly resembled those of reference fish from BPr, and the isotope values of the R. Kymijoki salmon were most similar to those from GoF.

Definition of the final feeding area
The mean length (+ SD) of ascending salmon (R. Simojoki: 87.4 + 9.3 cm, n ¼ 605; R. Kymijoki: 90.1 + 9.4 cm, n ¼ 70) was greater than that of reference salmon from each of the three sea areas: BS (74.3 + 9.4 cm, n ¼ 217), BPr (78.7 + 11.1 cm, n ¼ 707) and GoF (72.6 + 10.2 cm, n ¼ 271) (Kruskall-Wallis: H R. Simojoki ¼ 461.166, p , 0.001; H R. Kymijoki ¼ 153.424, p , 0.001).Without interannual standardization of d 15 N against the salmon lengths these size differences would have affected use of d 15 N values in the Baltic salmon migratory connectivity discriminant analysis.The average DA cross-validation (leaveone-out method) accuracy for salmon assignment to final feeding area was 85.4% and range 73.9 -96.9 % (Table 2).Annual mean probability estimates of the most likely feeding areas for all salmon ascending the R. Simojoki indicated that BPr was the most likely feeding area (mean + SD for ascending fish probability 0.59 + 0.32).The probability remained highest for BPr throughout most of the record; only in 1997 did GoF have a higher probability (Figure 3a).The probabilities that an ascending salmon had been last feeding in BS (0.21 + 0.26) or in GoF (0.20 + 0.27) were similar (Bonferroni-corrected Mann-Whitney, p .0.05) and clearly smaller than for BPr.
BS ranked higher than GoF during the 1990s (with the exception of 1997), although the difference between assignment probabilities was not statistically significant (Bonferroni-corrected Mann-Whitney, p .0.05).In the 2000s the situation reversed with reduced ascending probabilities from BS (with the exception of 2009), although this was not statistically significant either (Bonferroni-corrected Mann-Whitney, p .0.05) (Figure 3a).On average salmon had a statistically significantly higher probability of ascending from BS in the 1990s than in the 2000s (Bonferroni-corrected Mann-Whitney, p ¼ 0.033), while there was no significant differences of ascending from BPr or GoF between the 1990s and the 2000s (Bonferroni-corrected Mann-Whitney, .0.05).
Most of the R. Kymijoki salmon from the two years of study were assigned to the GoF feeding area (Figure 3d; 0.71 + 0.42) but with an appreciable probability of feeding in BPr (0.29 + 0.41); BS was not a likely feeding area.Tag-recovery data (Figure 4a and b) and SIA-based assignments (Figure 4c and d) appeared similar for both river populations.

Wild and hatchery-reared salmon
The probabilities of the most likely last feeding sea areas for R. Simojoki stock did not show statistically significant differences between wild or hatchery-reared salmon in any year [X 2 , p .0.05 (Figure 3b and c)].The stocked salmon seem to show some cyclicity, observed as a 4 -6 year cycle in BS assignments (Figure 3c).

Discussion
The likely final sea feeding areas of salmon before ascending their natal river and the degree of migratory connectivity of wild and hatchery-reared salmon were identified from salmon scale isotopic composition.According to DA, produced from SI data, most R. Simojoki salmon had likely been feeding in BPr.The remainder of the population was quite uniformly distributed to other likely feeding areas, BS and GoF.Most R. Kymijoki salmon had likely been feeding in GoF, with an appreciable probability of feeding in BPr; BS was unlikely to have been used as a feeding area.The two salmon populations both showed weak migratory connectivity to a single region irrespective of wild or stocked origin.DA assignments were consistent with the mark-recaptured salmon data, and moreover can provide information based on individuals.
Values of both salmon scale isotopes exhibited marked variation, both within a year and in mean values over the 22-year timespan.This variation is caused by abiotic and biotic factors (Satterfield and Finney, 2002), such as annual variation in primary producer isotopic values transferred through the foodweb (DeNiro and Epstein, 1978;DeNiro and Epstein, 1981;Vander Zanden and Rasmussen, 1999;McCutchan et al., 2003) and factors related to seawater temperature changes (MacKenzie et al., 2011a).The decrease in d 13 C of atmospheric CO 2 through global usage of fossil fuel and deforestation (the Suess effect; Verburg, 2007) over recent decades has also affected the CO 2 -d 13 C value in aquatic environments (Schloesser et al., 2009).While these sources of variation provide the basis for isotopic segregation between possible salmon feeding areas in the Baltic Sea, the observed trend in scale isotopic composition of ascending salmon over time is not necessarily due to changes in migratory behaviour, but could be a function of the abovementioned factors.
During the 22-year study period, the Baltic Sea pelagic foodweb has undergone a well-documented regime shift (e.g.Mo ¨llmann et al., 2009;Heikinheimo, 2011), whereby the system has changed from a cod (Gadus morhua)-dominated foodweb to a more sprat (Sprattus sprattus)-dominated foodweb in the 1990s (Mo ¨llmann et al., 2009).Consumers are fundamentally influenced by the isotopic composition of organisms at the bottom of the foodweb (e.g.Vander Zanden and Rasmussen, 1999;Satterfield and Finney, 2002).During this period there has been a shift in stock size of the two main prey species of salmon, from herring (Clupea harengus membras), which was previously more abundant, to sprat, which have dominated more recently (Mo ¨llmann et al., 2004, Mikkonen et al., 2011).Of these two prey species, sprat appear to hold a lower trophic position (Kiljunen et al. 2008).Therefore, we suggest that the changes in stock biomass over time of the two most important prey species for salmon is the main driver of the declining trend in d 15 N and d 13 C.However, from the perspective of migratory behaviour of salmon, such temporal change in SI values is not critical to this study, since reference scales were sampled during the feeding period equivalent to the last growth region of scales to obtain the most reliable assignment of final sea feeding area.
When SI-based assignment probabilities were compared with tag-recovery data, results for final feeding areas obtained from both methods were remarkably similar.Our SI-based assignment results are broadly consistent with previous reports of salmon feeding distribution in the Baltic Sea (Christensen and Larsson,  1979;Salminen et al., 1994;Ikonen, 2006), although the SI-based analyses indicate that GoF may play a more important role as a feeding area than previously thought.This difference could be explained by the fact that some salmon, which have fed in BPr, migrate via GoF on their way to the Gulf of Bothnia natal rivers (Aro, 1989;Ikonen, 2006), or have fed in nearby areas and therefore reflect isotope values close to GoF.It is also likely that salmon move between areas within the feeding season and will reflect mixed SI values from those areas via dietary integration.
There are several advantages to the use of SI-based methods in migration studies compared with conventional tagging methods.
(i) Conventional methods can be strongly susceptible to the bias that samples are obtained from a commercial fishery from the sea and therefore do not represent the whole feeding salmon population, but are heavily dependent on fishing effort.The use of SI-based methods will not entirely overcome the problem, but because SI values of ascending salmon represent seasonal averages of the feeding area (Hutchinson and Trueman, 2006) Tag-recaptures are ineffective for studying the consequence of migration patterns (e.g.final feeding area) on individual characteristics which are invisible in tagged fish but appear later in the life cycle; for example M74 mortality of yolk-sac fry, which affects Baltic Sea salmon (Ikonen, 2006;Keina ¨nen et al., 2012), is related to the diet of salmon and therefore also to the feeding areas, and SIA provides better information on individual dietary history than does tagging.(iv) Analyses of surviving and ascending salmon better represent the migratory connectivity of the actual breeding population than a tagged subsample of the population caught from the sea.Indeed, the cost-effective SI method can give individual-based data on feeding areas, unlike recapture data, which only reveal where a salmon was when it was caught.Webster et al. (2002) showed that strong migratory connectivity can be deduced when all individuals of a certain population spend their non-breeding season in the same area and weak connectivity when a population is distributed to several areas.Hence, our data suggest that both salmon populations show rather weak migratory connectivity.In the years 1996 and 2006, R. Simojoki salmon seemed to show stronger migratory connectivity by most likely feeding in BPr, which suggests some flexibility in year-to-year migratory behaviour and an ability of the population to respond to emerging resource and environmental changes (Webster et al., 2002).Salmon may move widely in search of better foraging areas.Evidence of such movement in the Baltic Sea comes from an observation that the largest salmon catches in the GoF were in 1996 and 1997, when prey species biomasses there were exceptionally large (ICES, 2012b).
We found no statistical differences in migratory connectivity between wild and hatchery-reared salmon in the selection of most likely last feeding area.Palm et al. (2008) reported evidence of similar behaviour, whereby adult salmon shoals in the sea are a mixture of individuals of different origin and relation, which would be consistent with our findings.However, these results contradict other reports (Salminen et al., 1994;Kallio-Nyberg et al., 1999, 2006) in which stocked immature salmon from northern rivers showed more restricted migration than wild salmon.This more restricted migration has been attributed to the decreased migration activity (Jutila et al., 2003) and the larger size of stocked smolts and therefore a better ability to feed on prey fish available in BS (Kallio-Nyberg et al., 2011).Methodological issues can also provide possible explanations for the contradictory results.Conventional tag-recovery studies deal with salmon caught from the sea that might not have survived to ascend the rivers, whereas our SI study only includes individuals that did survive to ascend their natal river.The suggestion of 4-6 year cyclicity in BS assignments of reared salmon cannot be verified due to the lack of data from the middle of the timespan.
Wild salmon populations are endangered in the Baltic Sea.Although the situation is improving through the developing fishing policy, conservation of the remaining wild populations is still crucial.Unfortunately, the similar migratory connectivity of wild and reared R. Simojoki salmon shown in our study offers no real possibility for accurately guiding allocation of commercial fishing towards the reared salmon while at the same time ensuring adequate conservation of wild salmon.However, as most of the R. Simojoki salmon population evidently feed in the BPr year after year, possible conservation actions should be focused on that area when considering the fishing quotas for salmon in the Baltic Sea.Because of this, and to promote sound conservation actions, further investigations of migratory connectivity of other salmon river stocks are required to develop an overall view of the migratory behaviour of Baltic Sea salmon.

Figure 1 .
Figure 1.Map of the Baltic Sea showing the ICES subdivisions in the sea-areas used in the study (indicated in grey): Baltic Proper (SubDivs 25, 26 and 28), Bothnian Sea (SubDiv 30), and Gulf of Finland (SubDiv 32).The locations of the River Simojoki and the River Kymijoki are also shown.

Figure 2 .
Figure 2. Annual salmon Salmo salar scale mean SI values for (a) d 13 C values, and (b) d 15 N values.Whiskers represent standard deviations.

Figure 3 .
Figure 3. Discriminant analysis (DA) results of annual mean probabilities of ascending salmon Salmo salar from three different Baltic Sea areas.(a) All, (b) wild origin, (c) stocked origin River Simojoki salmon, and (d) stocked origin River Kymijoki salmon.Whiskers represent variance measures of likely last feeding area probabilities of ascended salmon.

Table 1 .
Numbers of Atlantic salmon Salmo salar from different years analysed for the study.

Table 2 .
List of variables used in discriminant analysis (DA) and cross-validated (leave-one-out method) precision accuracy (%).