Global fish production is under threat from overexploitation and climate change (Jackson et al., 2001; Drinkwater et al., 2003; Brander, 2007, 2010). The strongest impact from climate variability occurs in northern areas like the Barents and Baltic Sea (ACIA, 2015), as well as in enclosed areas, e.g., the Black and Baltic Seas (BACC Author Team, 2008; Philippart et al., 2011). The Baltic Sea is an intra-continental dilution basin with a total area of 415,000 km² (including Kattegat, Figure 1) that has a highly productive but low-diversity ecosystem with cod, sprat, and herring as top predators. It is also characterized by a closed basin circulation (Voss et al., 2005) and by strong horizontal and vertical gradients in salinity and in ecosystem variables. The distribution of salinity is responsible for the low biodiversity of the Baltic Sea with marine species at the entrance to the North Sea decreasing eastwards and ultimately replaced by freshwater species in the inner reaches. A local minimum in species numbers occurs in the 5–7 salinity range (Remane, 1934).

Inter-annual and inter-decadal variability of the Baltic Sea is influenced by the climate variability of the Northern Hemisphere and by major Baltic inflows of water with high salinity and oxygen from the North Atlantic passing through the North Sea into the deeper parts of the Central Baltic Sea (Matthäus and Frank, 1992). The Baltic Sea ecosystem has been shown to be strongly influenced by the interplay of non-periodic shifts from freshwater to marine conditions and back during the early Holocene. These shifts resulted in changes in the relative composition of marine and freshwater species as well as cold and warm water taxa, in the occasional intrusion of non-indigenous species, and changes between oxygen-rich and -poor conditions in the deep basins (BACC Author Team, 2008).

In this paper we show that the decrease in the mean weight of 3-year-old central Baltic Sea herring is caused by climate variability via a complex mediator chain described in the following.

A statistical downscaling between the time series (Figure 4; Table 1) shows a significant correlation between the AMO and precipitation over the catchment area without any time lag. AMO and river runoff show a time lag of between −2 and 0 years and the rain and river runoff of −1 year. The three identified relationships between AMO and river runoff (lag = −2, −1, 0) have all the same correlation coefficient r and the same skill β (Table 1). Cumulative river runoff and salinity in the Gotland Sea are highly correlated with a lag of between −4 and −3 years, the time taken by the rain in the catchment to reach the central Gotland Sea. Here again both identified relationships have the same correlation coefficient and the same skill. Salinity and P. acuspes are weakly correlated with a lag of −1 year, as are the zooplankton and the herring weight, which show a time lag of −3 to 0 years. This lag is meaningful regarding the weight of 3-year-old herring. The four identified relationships (lag = −3, −2, −1, 0) have slightly different correlation coefficients (r = 0.66, 0.64, 0.59, 0.63) and different skills (β = 0.41, 0.39, 0.33, 0.37). The combination with the highest values (lag = −3) is used. The red filtered curves in Figure 2 show a strong similarity in their inter-decadal variability, which is comparable to the identified long-term trends across different trophic levels in the North Sea (Aebischer et al., 1990).

The results of the downscaling experiment are supported by a complex wavelet analysis, which suggests that the AMO forces the precipitation over the Baltic Sea catchment directly without phase difference (Figure 5A), whereas river runoff is hardly linked to the AMO (Figure 5B). This might be related to the result that the cumulative river runoff responds to the precipitation with an additional time lag of ~3 years (phase difference 270°) in lower periods (~4 years; Figure 5C). A small gap exists in the coherence between the cumulative river runoff and the precipitation from the end of the 1970s to the end of the 1980s. This coincides with the period of negative North Atlantic Oscillation (NAO) index (Reid et al., 2001; Dippner et al., 2014). A correlation of the river runoff with the salinity in the upper 30 m in the Gotland Sea only exists since the 1980s on periods of 8 to 15 years with a time lag of 4 years (Figure 5D). Long-term influences of the salinity on the mesozooplankton biomass can only be seen from the 1990s; however, although this result is statistically significant (Figure 5E), it cannot be interpreted because it is in the area of the cone of influence and therefore, the amplitude of coherence is influenced by boundary effects due to zero padding (Torrence and Compo, 1998). The mesozooplankton biomass is directly correlated to the weight of herring on periods of < 4 years (Figure 5F). Here a shift in the response period can be identified from 4 years to 1 year and back to 4 years. The short response period of 1 year between mesozooplankton biomass and herring weight coincides with the period of positive NAO index (Reid et al., 2001; Dippner et al., 2014).

The sea surface temperature (SST) of the North Atlantic has a distinct signal of multi-decadal variability (Schlesinger and Ramankutty, 1994). The AMO and its index are defined as the 10-years running mean of the detrended Atlantic SST anomaly of the Northern Hemisphere (Kerr, 2000). A positive AMO anomaly or a positive SST anomaly in the North Atlantic favors broad cyclonic pressure anomalies (Knight et al., 2006) and is related to a lower mean sea level pressure and higher rainfall in North West Europe during summer (Sutton and Hodson, 2005). Monthly mean precipitation over the catchment area of the Baltic Sea for the period 1950–2000 is significantly correlated to the AMO index for the whole year (Figure 3) indicating a heterogeneous and patchy distribution. Our findings also demonstrate that the effects of increasing rainfall originating in the 1970s have sustained for some decades (Figure 4B). The change in the mid 1970s might be explained by a regime shift of climatic origin, as the trend in the AMO has reversed (Kerr, 2000; Dima and Lohmann, 2007). After the regime shift in 1988/1989 (Reid et al., 2001; Dippner et al., 2014), two further processes might have caused additional warming of the North Atlantic, which also might contribute to increased rainfall in North West Europe. The observed weakening of the sub-polar gyre allowed subtropical warm water to extend farther to the north (Hátún et al., 2005). This period was also characterized by a warming caused by an increased heat transport from the tropics to the extra-tropics (Hoerling et al., 2001). The consequence was a significant increase in precipitation, which caused an increase in total river runoff (Knight et al., 2006), decreasing salinity, resulting in a shift in habitat for marine species, and lower biomass of P. acuspes in the Baltic Sea (Vuorinen et al., 2015). Previous studies have reported that a reduction in the salinity (Casini et al., 2010) of the Gotland Sea (Figure 1) is forced by the NAO and impacts marine copepod abundance and total freshwater runoff (Hänninen et al., 2000). Our investigations do not support these findings. An direct association between herring conditions and salinity has been documented by various authors (Rönkkönen et al., 2004; Casini et al., 2010). However, we could not identify a direct link between the NAO winter index and decreasing salinity or the abundance of P. acuspes. Climate signals have been identified in different mesozooplankton species, but never in Pseudocalanus sp. (Dippner et al., 2000, 2012). However, a significant correlation has been identified between the abundance of P. minutes and the herring recruitment in the Gulf of Finland (Dippner et al., 2001). It seems that our findings support the conclusion of Rajasilta et al. (2018) who argued that salinity acts as background signal that affects fish indirectly through changes in the prey community and fish diet.

The trend of reducing salinity might continue into the future. Different climate change emissions scenarios (IPCC SRES A2 or B2) with regional climate models for the Baltic Sea (Meier et al., 2006; BACC Author Team, 2008) project a decline in salinity of between 8 and 50%. This is likely to lead to a further reduction in habitat availability for marine species (Vuorinen et al., 2015) and even lower abundance and biomass of the marine copepod P. acuspes, the favorite food for central Baltic Sea herring. A continuation of this development might end in a similar situation to that occurring in the early Holocene: a freshwater Baltic Sea.

Two apparently contrasting hypotheses of herring growth regulation, so-called unspecified “hydro-climatic factors” and inter-specific density dependence, have been recently presented (Casini et al., 2010). Spawning stock biomass of sprat, the main food competitor for herring, were increasing up to 1997 (BACC Author Team, 2008), which might explain the reduction in herring weight. However, after 1997 the spawning stock biomass of sprat declined with no positive impact on the weight of herring. Observations in the Gulf of Finland (Figure 1) have shown that a density dependent mechanism is unlikely to have been the cause of the changes in herring weight at age (Rönkkönen et al., 2004); in this region zooplankton community structure changed supporting the theory of bottom-up control. Therefore, we conclude that the weight of herring was controlled by climatic and physical factors, i.e., an indirect mediator chain from the AMO via precipitation, river runoff, and a decrease in salinity (Dippner, 2006). This led to a reduction in habitat available to marine copepods with a lower spring biomass (Möllmann et al., 2009), which affected the weight of central Baltic Sea herring. Due to different time lags a direct coherence between the AMO and herring weight cannot be identified (Figure 6), except for a weak signal around 1984, meaning that any trophic interaction between fish and plankton was likely to be indirect and climate driven (Aebischer et al., 1990).

Fluctuations in harvestable fish stocks have major economic consequences for human society. The observed reduction in the mean weight of central Baltic Sea herring from 50 to 70 g in the late 1970s to 25–30 g today has resulted in a catch deficit in the weight of fish over this period. This deficit is equivalent to one million tons of herring (by weight), includes a decline in the quality of fish, and represents a strong reduction in ecosystems goods and services. Using a net price of 0.1 € kg⁻¹ according to the European Parliament Report “Industrial Fisheries in the Baltic Sea,” the loss in catch is equivalent to ~100 million € (Lassen, 2011).

Herring in the Central Baltic Sea is managed by means of a special management plan that is in accordance to the maximum sustainable yield concept (ICES Advice, 2014). With decreasing fishing mortality the stock has shown signs of increasing stock biomass since the mid 2000s, although this was not accompanied by an increase in mean weight. The difference between the actual and the potential condition are not only a direct loss to the economy (catch) and ecology (prey) but also a loss in product quality (reduced fat content). Even though fisheries management is one of the few levers human have for “managing” ecosystems like the Baltic Sea, our results show that these attempts may have limited impact when overruled by extrinsic factors such as climate change.

JD performed the statistical downscaling and BF the wavelet analysis. JD, BF, and CH contributed to the writing of the main text.