Western Baltic spring-spawning herring (WBSSH, Clupea harengus L.) perform seasonal migrations between feeding grounds in the Skagerrak and Kattegat and their spawning sites in the Western Baltic Sea. The Sound, connecting the Kattegat to the Western Baltic Sea, is an important aggregation and transition zone for this herring stock during its spawning migration. We analysed data from the German autumn acoustic surveys of the years 1993–2009. These data revealed at least two different distribution patterns of herring in autumn: herring generally aggregated in the Sound, but in some years the majority of herring were detected further south, being outside of the Sound by the time of the survey. We tested whether observed annual differences in the herring migration can be explained by either stock characteristics (age and size) or hydrographical variables (salinity and oxygen concentration). Our results suggest that rather than being related to stock characteristics, the distribution pattern of herring was related to environmental conditions, i.e. to marine inflow events into the Baltic Sea. Barotropic inflow events in late summer and early autumn seem to prevent deoxygenation in the Sound and thereby favour the prolonged aggregation of herring in the Sound.
Atlantic herring (Clupea harengus L.) is an important commercially exploited fish species in the Baltic Sea and also an important prey for other fish, marine mammals, and birds (Sparholt, 1994; Börjesson et al., 2003; Guse et al., 2009). In the Baltic Sea, several herring stocks are distinguished, including the Western Baltic spring-spawning herring (WBSSH) and the Central Baltic herring (CBH).
The Baltic Sea is a semi-enclosed brackish water body connected by shallow straits to the Kattegat, Skagerrak and the subsequent North Sea (Figure 1) allowing for an outflow of low-saline water and a periodic inflow of high-saline water. The WBSSH stock is distributed in the Belt Sea (ICES Subdivision (SD) 22), the Sound (SD 23), the Arkona Sea (SD 24), and in the Skagerrak and Kattegat (ICES Division IIIa). Most individuals three years and older take part in summer feeding migrations to the Skagerrak and Kattegat and return for overwintering in autumn and spawning in spring, while juveniles stay in the Western Baltic Sea (SDs 22–24) year-round (Aro, 1989; Biester, 1989; van Deurs and Ramkær, 2007). During the summer feeding season, WBSSH mix with North Sea autumn-spawning herring in the Skagerrak and Kattegat (Bekkevold et al., 2011). Some CBH were observed together with WBSSH in the Arkona Sea (Podolska et al., 2006; Gröhsler et al., 2013). CBH, mainly distributed east of Bornholm island (SD 25–27, 29 and 32), and WBSSH can be distinguished based on differences in growth (Gröhsler et al., 2013).
The WBSSH mature at the age of 2–3 years, are iteroparous, and spawn benthic eggs that develop into pelagic larvae (Klinkhardt, 1996). Nielsen et al. (2001) observed that WBSSH aggregated in the Sound from August until March with maximum abundance in autumn, and that larger-sized herring left the Sound first. The Greifswald Bay (SD 24; Figure 1) was identified as a major spawning site with a high spawning activity of WBSSH (Biester, 1989; Oeberst et al., 2009). Spawning shoals in the Greifswald Bay consist mainly of individuals aged 2–6 years (Rajasilta et al., 2006).
Using data of the German autumn acoustic surveys carried out annually within the frame of the Baltic International Acoustic Survey (BIAS), differences in the autumn distribution pattern of WBSSH in the survey period 1993–2009 were observed. Although the Sound is considered to be an aggregation and overwintering area of WBSSH, in some years many herring appeared to have already moved out of the Sound and into the Arkona Sea by autumn. It is the objective of this study to describe and explain the observed annual differences in herring autumn distribution and migration timing.
Stock characteristics can influence herring migration behaviour. Individual size might affect migration distance due to energy constraints, through the choice of spawning grounds and overwintering areas (Slotte and Fiksen, 2000). Migration routes and timing are learned by social transmission from older experienced individuals (McQuinn, 1997; Corten, 2002; Kvamme et al., 2003). Migration patterns of Norwegian spring-spawning herring changed in relation to demographic properties of the stock (Dragesund et al., 1997). The distribution range of North Sea autumn-spawning herring could be related to changes in stock biomass, but a careful consideration of environmental variables has been suggested (Bailey et al., 1998).
Besides biological characteristics of the herring stock, environmental conditions may directly affect the migration and distribution patterns. The effects of environmental conditions on WBSSH migration timing have so far been inconclusive (Nielsen et al., 2001). It has been suggested that variations in temperature and salinity, indicating changes in water masses, could affect distribution patterns of herring in the North Sea (Maravelias and Reid, 1995; Röckmann et al., 2011). Migration timing of Pacific herring (Clupea pallasi) responded to changes in environmental conditions, i.e. temperature (Tojo et al., 2007). As herring can tolerate a wide variety of temperatures and salinities, the environment has been shown to affect herring distribution mostly indirectly, for example through the presence of zooplankton prey (Maravelias and Reid, 1995; Misund et al., 1998; Nøttestad et al., 2007; Broms et al., 2012). However, herring in the Central Baltic Sea were observed to avoid water layers with low oxygen saturation (Neuenfeldt, 2002).
In this study, we combined the analyses of acoustic surveys with environmental monitoring data to understand factors governing the autumn distribution of WBSSH in the Sound and in the Arkona Sea. We tested whether the distribution pattern was affected by stock characteristics of herring (ages and sizes of herring or by the estimated total number of herring in the area). We also considered environmental conditions related to the periodic marine autumn inflow events which influence the conditions in the Western Baltic Sea: the oxygen concentration in the Sound, which could directly affect herring migration behaviour, and the salinity in the deeper layers of the Arkona Sea as a direct reflection of marine autumn inflow events.
Acoustic survey and biological sampling
Data from the German autumn acoustic surveys for the period 1993–2009 (excluding 2001 when the survey did not cover the Sound) was used to summarize the autumn distribution of herring. The German autumn acoustic surveys are part of the Baltic International Acoustic Surveys (BIAS) and cover the ICES SDs 22–24 and the southern part of the Kattegat (Division IIIa) as the standard area (see Supplementary data), and have the main objective of estimating the stock size of clupeid fish in autumn (late September/early October) in the Baltic Sea. Only mature fish, three years and older (3+ herring), taking part in feeding migrations were considered. Because numbers of 3+ herring in the Belt Sea and the Kattegat were low, only data for the Sound and the Arkona Sea were included in the analyses. Due to the vertical diel migrations of pelagic fish, with high fish concentrations near the bottom during daytime, acoustic measurements and trawling were performed during night-time. During the night herring were observed mainly at 10–30 m depth in the Sound (see Supplementary data). The number of herring per rectangle was estimated from acoustic measurements supplemented by trawl samples. From biological samples, length distributions were recorded and otoliths of individual herring were collected for age estimation. Further details of the survey method are explained in the “Manual for the Baltic International Acoustic Survey (BIAS)” (ICES, 2011b).
Analysis of annual herring spatial distribution
Shifts in spatial stock distribution were identified by comparing the centre of gravity from one year to another (Óskarsson et al., 2009; Engelhard et al., 2011). The centre of gravity was calculated as the weighted mean latitude (Latcentre) and longitude (Longcentre):
Because movements in the Sound towards the Arkona Sea occur mostly along the north-south direction, Latcentre is a better indicator than Longcentre to distinguish years of differing spatial distribution patterns. Using Latcentre, the years were grouped into clusters, Glat, to identify years of differing migration strategies. The clustering algorithm PAM (partitioning around medoids), that minimizes within-group dissimilarities while maximizing between-group dissimilarities, was applied. The number of clusters was determined according to the highest average silhouette width; the silhouette width describes the dissimilarity of an object to its second best cluster, taking values from −1 to 1 (Struyf et al., 1997).
Analyses of stock characteristics
The characteristics of the stock in terms of the total number of individuals, their mean age and lengths at the time of the survey were analysed in relation to the herring distribution. To test whether the observed Latcentre was influenced by the mean age of herring (agemean) or the total number of herring (N), linear models were fitted in the general form:
Here, y represents the observed variable Latcentre, β the estimated parameters and x the respective variable describing a characteristic of the stock (agemean, N). The Durbin-Watson test indicated that there was no significant temporal autocorrelation in the residuals.
To test whether the length distribution could have affected the herring migration, the individual lengths of herring were compared between years of differing herring spatial distribution patterns, also accounting for differences between subdivisions (SD). Only 3+ herring of lengths between 21 and 31 cm were selected. The length distributions were aggregated per subdivision and year. The numbers of individuals per length class were standardized for each subdivision and year and rounded to the nearest integer. Instead of using Latcentre as a continuous variable in the model, years were clustered into groups Glat. A generalized linear model was fitted using a Poisson distribution with a log-link:
The length classes (L) were included together with its interaction with the factors subdivision (SD) and clustered years (Glat), which indicate their respective effects on the length distribution. A significant effect of variables was indicated at the 5% level.
To account for potential mixing of WBSSH and CBH in the Arkona Sea, the analyses of the lengths were performed on data with and without individuals classified as CBH. The numbers of potential CBH per age and length class were determined using a separation function (Gröhsler et al., 2013). The separation function was estimated to classify 97% correctly as WBSSH and CBH (Gröhsler et al., 2013). Since only data from 2005–2010 were used to estimate the separation function and growth may vary between years, the number of CBH in years prior to 2005 may be slightly underestimated.
Marine inflow dynamics through the Sound
The Sound represents a transition zone between the saline water of the Kattegat and the low-saline water of the Arkona Sea. The northern part of the Sound is separated from the Kattegat by a 25-m deep sill. A shallow sill (8 m depth, Drogden Sill) in its southern part restricts the marine water flow into the Arkona Sea (Figure 1). Inflows of saline water from the Kattegat through the Sound, over the Drogden Sill, are the major source of saltwater entering the Baltic Sea (Lintrup and Jakobsen, 1999). Inflowing denser saline water deposits near the bottom lead to a strong stratification of the water column (Matthäus and Schinke, 1999).
The marine inflows through the Sound are barotropic and occur periodically, driven by wind-induced sea level differences between the Kattegat and the Baltic Sea, whereas low-salinity surface water outflow from the Baltic Sea is not constrained by sill depth and is exchanged regularly. The inflowing high-salinity water enters the Arkona Sea, accumulates in a saltwater pool near the bottom below the permanent halocline (Feistel et al., 2003; Lass and Mohrholz, 2003; Lass et al., 2005). The residence time of the saltwater pool near the bottom in the Arkona Sea is estimated to be between 1 and 3 months (Lass et al., 2005).
The Sound is a eutrophic area with strong stratification where oxygen deficiency can occur in autumn (Göransson, 2002). For herring aggregating at high density in the area in autumn, oxygen concentrations may be critical. Barotropic inflow events can affect oxygen levels in the Sound. Therefore, the analyses of environmental variables in autumn were focused on (i) oxygen concentration in the Sound at 20–30 m depth, a depth characterized by inflowing oxygen-rich saline water and oxygen-consuming processes potentially affecting aggregating herring, and (ii) salinity in the Arkona Sea below the halocline, at 40–50 m depth, where barotopic saltwater water inflow can be detected.
Analyses of environmental conditions
To evaluate whether spatial differences in herring numbers can be related to local environmental conditions connected with marine autumn inflow events, we used hydrographical data available at the ICES Oceanographic database (ICES, 2011a). These data include salinity, temperature, and dissolved oxygen concentrations at different depths at geographical positions across the Baltic Sea. The weekly mean salinity, temperature, and oxygen concentration per water layer in 10 m intervals were calculated per subdivision.
A linear model of the general form in Equation 2 was used to test for an effect of local oxygen concentration in the layer of 20–30 m depth in August and September on the numbers of herring in the Sound. Correspondingly, the effect of salinity in the Arkona Sea in September at 40–50 m depth on the number of herring in the Arkona Sea was tested. The Durbin-Watson test indicated that there was no significant temporal autocorrelation in the residuals. The p-values were considered to be significant at the 5% level.
The centre latitude reflects the overall distribution of herring and depends on the presence of herring in multiple rectangles of the Arkona Sea. Its specific location may not be directly linked to migration timing. To distinguish years that differ in migration timing of herring leaving the Sound, years were clustered (Glat). The year clusters were compared with respect to mean oxygen concentration in the Sound at 20–30 m depth in August and September, and to mean salinity in the Arkona Sea at 40–50 m depth in September using Welch's t-test. Normal distribution within the clusters was tested and confirmed using the Shapiro-Wilk test at the 5% level of significance.
Observed spatial distribution patterns of herring in autumn
The highest numbers of herring three years and older (3+ herring) were observed in the Arkona Sea and in the Sound. Yet, considering the limited space available in the Sound (367 square nautical miles; standard survey area with >10 m depth) in comparison with the Arkona Sea (4622 square nautical miles; ICES, 2011b), the Sound can be considered an aggregation site for migrating WBSSH.
In the survey period 1993–2009, the centre of gravity of 3+ herring in the Sound and the Arkona Sea was variable (Figure 2). In some years high numbers of 3+ herring were found in the Arkona Sea, as indicated by lower values of the latitudinal centre of gravity (Latcentre). Depending on Latcentre, the years can be partitioned into either two or three strong clusters (average silhouette width of 0.68 or 0.75). In both cases, Latcentre of seven years clustered together were located north (cluster “north”), in other years Latcentre were shifted further south (Figure 2a). Herring numbers per rectangle revealed that in three years 1994, 1995 and 2007, the intermediate values of Latcentre were caused by relatively high numbers of herring in rectangles 39G3 and 39G4 in the Arkona Sea (see Supplementary data). In consequence, two clusters were used to contrast years of northerly distribution with the other years.
The years of “north” distribution describe the two periods 1997–1999 and 2003–2006 (Figure 2b). During these periods the relative numbers of herring in the Sound were high (45–55%) in comparison with the Arkona Sea. In other years the majority of herring (>70%) could be found in the Arkona Sea. The cluster of “north” distribution was not as easily identified from the absolute numbers of herring in one area (Figure 3). Local herring numbers were influenced by the total numbers, which in some years were high (e.g. 2006), and in others exceptionally low (e.g. 2007, 2009).
There was no significant relationship between mean age (df = 14, t = 0.41, p = 0.7) or the total number of herring (df = 14, t = 0.2, p = 0.6) and herring spatial distribution, i.e. Latcentre (Figure 4). The results of the generalized linear model (Equation 3) revealed no significant interaction between GLat and length (χ221 = 32.6, p = 0.051) or between GLat, length, and subdivision (χ221 = 12.9, p = 0.91). Only the interaction term between length and subdivision was significant (χ221 = 498, p < 0.01), indicating a significant difference in length structure between subdivisions. In the Arkona Sea, there were relatively more small (21–23 cm) and fewer large individuals (27–31 cm) than in the Sound (Figure 5). The model accounted for more than half of the total deviance in the data (R2 = 0.62).
In the Arkona Sea samples, the percentage herring classified as WBSSH varied between 63 and 97% per year (Figure 6). Including only herring classified WBSSH in the model also led to the conclusion that the length structure did not differ significantly between years clustered as “north” distribution and others but instead differed significantly between subdivisions.
Environmental variables in the Sound and in the Arkona Sea varied seasonally and spatially. Water temperature was generally highest in August and lowest in March. Similarly, oxygen concentrations varied seasonally with lowest concentrations in the summer due to low oxygen solubility and higher oxygen consumption at higher temperature. In August/September oxygen concentrations were lowest near the bottom (below 20 m) where exchange with the surface water is limited. In contrast with oxygen and temperature, there was no seasonality in salinity. Salinity increased with depth due to density differences and stratification of the water column. A horizontal salinity gradient exists with salinity being generally higher in the Kattegat and the northern Sound than in the Arkona Sea and further east in the Central Baltic Sea (SD 25).
Results from a linear model showed no significant effect of mean monthly oxygen concentration at 20–30 m depth in the Sound on local herring number in August (t = 1.2, p = 0.26) or September (t = 1.8, p = 0.1). However, the number of herring in the Sound was significantly related to the mean oxygen concentration at 20–30 m depth in Week 34 in August (t = 2.4, p = 0.03). The regression using a linear model did not indicate a significant effect of mean salinity at 40–50 m depth in September on the number of herring in the Arkona Sea (t = 1.7, p = 0.1).
The groups of years clustered according to the spatial distribution pattern of herring differed in the characteristic environmental conditions (Figure 7). Mean oxygen concentrations in Week 34 (August) at 20–30 m depth in the Sound were significantly higher in years when herring were northerly distributed (t = 2.8, p = 0.02, Figure 7b). In deeper layers of the Arkona Sea, at 40–50 m depth, mean monthly salinity in September was higher when herring were northerly distributed than in other years (t = 3.1, p = 0.01, Figure 7d).
Our results show that while WBSSH generally aggregate in the Sound, in some years relatively high numbers of herring were already found in the Arkona Sea in autumn. However, in two periods (1997–1999 and 2003–2006) the aggregation phase extended beyond the time of the surveys in September–October. We suggest that during a critical time in late summer/early autumn, oxygen deficiency in the Sound may trigger an early migration of herring into the Arkona Sea. In contrast, barotropic inflow of marine water from the Kattegat not only increased the salinity near the bottom in the Arkona Sea but also the oxygen concentration in the Sound, thereby favouring a prolonged aggregation of herring there. The total number of herring, the observed mean age, and the size of herring in the Sound and the Arkona Sea did not explain the observed differences in the distribution patterns. Instead our results suggest that environmental conditions were more likely to have triggered the early autumn migration from the Sound into the Arkona Sea.
Oxygen deficiency in the Sound may be most severe from August to early October. The low oxygen concentrations detected in summer and in early autumn may have been caused by elevated water temperatures (lower oxygen solubility, higher oxygen consumption), lack of westerly wind conditions and low wind force (inflow less likely, less mixing), and also by increased nutrient load (Mattsson and Stigebrandt, 1993; Meier et al., 2011a). For example, August–September 2002 was characterized by exceptionally warm weather, with low-wind conditions, increased stratification of the water column, and lack of barotropic inflow, contributing to very low oxygen concentrations (below 2 mg l−1) in the Sound (Feistel et al., 2003; HELCOM, 2003; Mohrholz et al., 2006). Barotropic autumn inflow events may counteract the oxygen depletion (Karlson et al., 2002). Major barotropic inflow events were observed in 1997, 2003 and 2004, preventing an oxygen deficit in the Sound (Feistel et al., 2006; Jonasson et al., 2012). Our results indicate that in these years herring aggregated in the Sound and fewer herring were found in the Arkona Sea.
Herring perform diel migrations. In the Sound, herring mainly aggregated at 15–30 m depth at night (Nielsen et al., 2001). We have shown that in years without marine inflow from the Kattegat, herring would experience low oxygen concentrations in this layer. During the day herring are observed to move to deeper layers where oxygen concentrations are lower (Orłowski, 2005). While Orłowski (2005) found that herring are able to tolerate oxygen deficiency for short periods of time, Neuenfeldt (2002) pointed out that herring avoid water layers of low oxygen saturation levels. Prolonged exposure, during both day and night, may lead to changes in herring migration behaviour such as the ones observed in this study. Low oxygen concentration was found to induce higher swimming speed herring, and shoals may change their schooling dynamics and increase in school volume to allow more oxygen per individual (Domenici et al., 2002; Herbert and Steffensen, 2006). Dommasnes et al. (1994) described local low oxygen levels in a fjord caused by overwintering Norwegian spring-spawning herring. At low oxygen levels in the Sound, oxygen concentrations may be depleted even further within the dense aggregations of herring schools due to respiration. In space-limited aggregation areas, such as the Sound, temporarily low oxygen concentrations could therefore contribute to earlier migration of WBSSH into the Arkona Sea.
In the much larger Arkona Sea, shoal volume is not space limited. Here, a lack of marine inflow into the Arkona Sea leads to weaker stratification and a deeper halocline, which allows for vertical mixing and reduction of hypoxia in layers above the halocline (Gerlach, 1994; Conley et al., 2002). Therefore, in years without barotropic inflow, oxygen availability in late summer and early autumn may be sufficient in the Arkona Sea, thus favouring early migration of herring there.
One might suspect that the higher numbers of herring in the Arkona Sea in some years were caused by more CBH migrating into the Arkona Sea. Yet the proportion of CBH, classified by a separation function (Gröhsler et al., 2013), did not indicate generally higher proportions of CBH in the autumn of these years. For example, in 2002, when herring distribution was shifted furthest south and the oxygen level in the Sound was very low, the proportion of CBH in the samples was at a minimum. Also, the number of herring in the Arkona Sea alone could not be related to salinity near the bottom there, i.e. marine inflow events.
Climatic changes could greatly affect spawning, recruitment and migration behaviour of WBSSH (MacKenzie et al., 2007; Cardinale et al., 2009). Since the 1980s saltwater inflow events decreased in frequency, which in combination with increased precipitation lead to decreasing salinity in the Baltic Sea (Ojaveer and Kalejs, 2005). Also, climate projections predict higher precipitation, lower westerly wind conditions, higher temperature, and lower frequency of barotropic inflow events in late summer and early autumn (Matthäus and Schinke, 1999; Meier et al., 2011b). Rising temperature will not only decrease oxygen solubility but also increases oxygen consumption, which could promote temporal hypoxia in the Sound (Conley et al., 2011; Bendtsen and Hansen, 2013). In the future, lack of marine inflow from the Kattegat counteracting oxygen depletion in the Sound may thereby trigger more frequently early migration of herring from the Sound into the Arkona Sea.
WBSSH recruitment has decreased consistently since 2000 (ICES, 2013). So far, it is not clear whether the annual differences in migration timing could have affected either the spawning activity and timing in the following spring, or the recruitment success. Evidence exists for phenotypic plasticity in spawning timing, as switching between winter and spring spawning in Greifswald Bay was observed (Bekkevold et al., 2007). Here, further research is required.
The German autumn acoustic surveys delivered the best available data on herring spatial distribution in autumn in the Western Baltic Sea over a long period; these data are also fishery-independent. The commercial fisheries in the Arkona Sea target spawning aggregations of herring with highest landings in March and April (Nielsen et al., 2001). It is generally difficult to use commercial landings as an indicator for annual changes in herring autumn migration patterns, because management measures (total allowable catches) and economic factors (e.g. fish and fuel prices) also affect temporal landing patterns (Haapasaari et al., 2012).
This study presented a case of habitat connectivity between the feeding and spawning grounds of WBSSH. The Sound, a migration corridor, is used as an aggregation or overwintering site by WBSSH. Environmental conditions were shown to affect migration timing. A better understanding of the migration behaviour of WBSSH in relation to environmental conditions may improve prediction of the stock response to climate change and ultimately help develop fisheries management strategies for sustainable exploitation. Considering the relatively low fishing mortality in the Sound (due to a trawling ban), and the activity of commercial fisheries in the Arkona Sea, migration timing may affect the exposure of herring to fisheries in the winter months. The results of this study can assist investigations into the effects of seasonal hypoxia on migratory fish species in areas characterized by eutrophication, upwelling, highly stratified water masses promoting oxygen depletion, and water bodies with long water residence times, such as enclosed seas, fjords, bays, and estuaries (Levin et al., 2009).
The following supplementary data is available at ICES Journal of Marine Science online. It includes a table with the number of herring per ICES rectangle as estimated from the survey for the period 1993–2009 and two additional figures: of a typical cruise track of the German autumn acoustic survey and acoustic measurements of herring in the Sound for 2006 as an illustration of herring vertical distribution in the Sound.
The research was partly financed and conducted as part of the Fehmarn Belt Science Provision Project.
We would like to thank Eberhard Götze for assistance in conducting the surveys and analysing the survey data, and Rainer Oeberst, Mikko Heino and two anonymous reviewers for comments on earlier versions of the manuscript.