Atlantic climatic factors control decadal dynamics of a Baltic Sea copepod Temora longicornis

ECOGRAPHY 26: 672–678, 2003

Atlantic climatic factors control decadal dynamics of a Baltic Sea copepod Temora longicornis Jari Ha¨nninen, Ilppo Vuorinen and Georgs Kornilovs

Ha¨nninen, J., Vuorinen, I. and Kornilovs, G. 2003. Atlantic climatic factors control decadal dynamics of a Baltic Sea copepod Temora longicornis. – Ecography 26: 672–678. We discovered, using transfer functions, that climatic changes in the Atlantic control the abundance of Temora longicornis, a dominant pelagic copepod of the Baltic Sea. The seawater salinity was increasing and copepod numbers were high from 1960s up to 1970s. Then the freshwater runoff started to increase, which resulted in decreasing salinities and abundance of the copepod. At the end of 1990s, runoffs remained at a high level, and the decrease of surface salinities and Temora leveled off. Due to time lags between variables studied, we also make predictions of changes expectable in early 2000s. The total freshwater runoff to the Baltic Sea followed the North Atlantic Oscillation with an immediate lag. Salinity followed the runoff non-linearly with a lag of 4–9 months. Temora longicornis followed the salinity with a lag of 1 – 3 months. Predicted abundance of T. longicornis will remain low implicating poor feeding conditions for planktivores. Our study points out the importance of physical factors in control of pelagic environments compared to ecological interactions, such as top-down and bottom-up. J. Ha¨nninen ( [email protected]) and I. Vuorinen, Archipelago Research Inst., Uni6. of Turku, FIN-20 014 Turku, Finland. – G. Kornilo6s, Lat6ian Fisheries Research Inst., Dauga6gri6as Str. 8, LA-1007, Riga, Lat6ia.

The climate of the North Atlantic was early anticipated to control the Baltic Sea hydrography and biota through the effect of westerly winds over the North Sea (Ka¨ndler 1949). Using transfer function (TF) models, Ha¨nninen et al. (2000) presented a chain of events between changes in the North Atlantic Oscillation (NAO) and subsequent changes in the Baltic Sea runoff and salinity. Changes in salinity have been suggested to explain the present poor condition and starvation of the Baltic herring, as well as the decline in the Baltic cod stock. Both of them have their origin in the 1980s and have caused problems for fisheries and fisheries’ management (Lumberg and Ojaveer 1991, Dickson and Brander 1993, Rudstam et al. 1994, Flinkman et al. 1998). Several studies on pelagic plankton have also shown a decline of large neritic copepods during the 1980s (Lumberg and Ojaveer 1991, Wasmund et al. 1996, Ojaveer et al. 1998, Vuorinen et al. 1998). These

separate findings suggested including biological time-series in a TF exercise. The same reasoning that was used for the physical processes by Ha¨nninen et al. (2000), is in this paper hypothetically extended to a calanoid copepod, Temora longicornis (Mu¨ller). It is one of the dominant pelagic copepods in the Baltic Sea (e.g., Wasmund et al. 1996), and an important food item in the diet of pelagic predators, the Baltic herring Clupea harengus membras L., and sprat Clupea sprattus (L.) (e.g., Flinkman et al. 1998). Ha¨nninen et al. (2000) also presented a possibility to predict above-mentioned interactions, due to the existence of a time lag of several months between a climatic-driven change in the Atlantic and the hydrographic response in the Baltic Sea. In this paper we also used TFs in order to present predictions on expected changes in the runoff, salinity, and the abundance of T. longicornis.

Accepted 28 April 2003 Copyright © ECOGRAPHY 2003 ISSN 0906-7590

672

ECOGRAPHY 26:5 (2003)

Materials and methods Data The NAO data used were a seasonal index of the difference in normalized sea level pressures between Ponta Delgada, Azores and Stykkisholmur/Reykjavik, Iceland in 1961 –2000 (data taken from the NCAR Climate Analysis Section B http://goldhill.cgd.ucar.edu/  jhurrell/nao.html\). The runoff data were monthly values (m3 s − 1) of total freshwater discharges from catchment area to the Baltic Sea from 1961 to 1998, provided by the Swedish Meteorological and Hydrological Institute (SMHI), and transformed to quarterly values for this study. To observe long-term changes in hydrography and plankton we used data of the Latvian Fisheries Research Institute (LatFRI), collected in the period 1961 –1997 from the area of International Council for the Exploration of the Sea (ICES) sub-division 28 in the Gotland basin. These data were sampled during quarterly seasonal surveys (February, May, August, October) in a net of monitoring stations in the Central Baltic Sea Proper (Fig. 1). In addition to mesozooplankton, the data include observations of salinity (PSU) measured from surface to bottom with fixed 5 –10 m intervals. Mesozooplankton was sampled using a Juday-net (Anon. 1968), and hydrography using a Nansen-type water sampler (1-liter capacity). Salinity was measured either by the Knudsen-method or with

an inductivity salinometer. The Baltic Marine Environment Protection Commission (HELCOM) provided some additional zooplankton data from the Gotland Basin. Data from the surface layer ( B 25 m and above the thermocline) were averaged over the sub-area and aggregated to seasonal quarters. Logarithmic transformation ln(x) was used for zooplankton data to achieve homogeneity of variances.

Statistical analysis In the statistical analysis we applied transfer function models (TF), also called dynamic regressions (Box and Jenkins 1976, Pankratz 1991), using the Scientific Computing Associates (SCA) Statistical System Software, release VI.2a (32-bit) (Liu and Hudak 1992). Transfer functions are able to connect one series not only with its own past values, but also with past and present values of other, related time-series. This is done by merging the basic concepts of the general regression model with that of traditional ARIMA models. In regression analysis, the response of a dependent variable is related to values of potential explanatory variables. The deficiency of regression analysis is evident when the error terms of the model are serially correlated, which results in an ineffectual or incorrect model (e.g., Box and Newbold 1971). In order to account for the autocorrelated structure of time-series data, autoregressive-integrated-moving average (ARIMA) models were introduced (e.g., Box and Jenkins 1976). ARIMA time-series analyses comprise an iterative procedure for modeling and encompass three main phases; identification of a time-series, parameter estimation, and diagnostic checking. Once an appropriate model is determined, it can be used for forecasting, control, or simply in order to better understand the structure of the time-series. The univariate ARIMA models are useful for the analysis of a single time-series. In such a case modeling is limited to the information contained in the series’ own past. In many cases, however, it may be possible to relate the response of one series not only to its own past values, but also to the past and present values of other, related time-series. TF models were introduced as a class of models particularly useful for these kinds of applications. The general form of the TF model is Yt = C+

Fig. 1. LatFRI and HELCOM monitoring stations in the central Baltic Sea. ICES sub-division 28 is indicated with square. ECOGRAPHY 26:5 (2003)

v(B) X t + Nt , d(B)

(1)

where Yt = response (output) variable, Xt = explanatory (input) variable(s), and C = constant term, which indicates a possible trend in the series when existing. The parameter v(B)=TF between Yt and Xt, either in a linear form, when d(B)= 1, or as a rational form, when d(B)"1. The value B represents the delay of the 673

response in the process. The v values (v= v0, v1, v2, …) are referred to as the TF weights for the input series Xt (Box and Jenkins 1976). TF weights provide a measure of how the input series affects the output series. Thus the weight v0 is a measure of how the current response is affected by the current value of the input series, while v1 is a measure of how current response is affected by the value of the input series one period ago, etc. The sum of all weights is called the steady-state gain and represents the total change in the mean level of the response variable. Nt =disturbance term, which is not assumed to be ‘‘white noise’’ and is modeled as an ARIMA process. ARIMA modeling gives approximations that are more reasonable for Nt, which results in more efficient estimates of the TF weights. At the beginning of modeling, the disturbance term gains a first-order autoregressive, AR(1), approximation, that is Nt =

1 at, 1-fB

when there is no seasonality present, and, N=

1 at, (1-f1B) (1-F2Bs)

with seasonality (seasonal period s). Here at represents a sequence of random errors that are independently and identically distributed with normal distribution. B represents the backshift (or lag) operator, which refers to previous values of data series (can be extended over longer lags). The parameter f indicates the autoregressive (AR) operator in non-seasonal series and F in seasonal series. However, the models are usually more complex than AR(1) order, e.g., autoregressive operator (non-seasonal or seasonal or both) is more than one in order [e.g., AR(2) or AR(3)], or there is a still more complex multiplicative form of operator than the simple seasonal series. Moreover, the disturbance term can have a moving average (MA) operator, represented by u and U in non-seasonal and seasonal series, respectively. MA operator(s) are always placed in the denominator of the formula. Another case is the series where the disturbance term is mixed with AR and MA operators. For more detailed discussion of these properties see, e.g., Box and Jenkins (1976). There are two principal assumptions of the TF models. First, the input series can affect the response variable, but not conversely, i.e., the relationship between Xt and Yt is unidirectional. Second, the input series is assumed independent of the disturbance. There are many similarities between TF and general regression models, but the models differ in two important respects: the assumption regarding the error (disturbance) term and the complexity of the parameter representation. 674

The TF test statistics of significance of the parameters is

t=

(estimate) −(hypothesized value) (estimated STD of estimate)

The t value is the value associated with a one-sample t-test with a test of ‘‘parameter=0’’. This statistic is then compared with a critical value of the t distribution with n −p degrees of freedom (n =number of observations, p =number of parameters estimated). As a rule of thumb the parameter is significant when t ] 1.96, when the number of observations is high ( ] 120). This is the case almost without exception in the time-series analysis. Only significant parameters are included into the TF models. For more comprehensive presentation of the TF models see, e.g., Box and Jenkins (1976), and Liu and Hudak (1992). Here, only the most convenient model was selected for presentation. Three following nested criteria were used for selection of the models presented: 1) The smallest residual standard error obtained. 2) Parsimony, i.e. the simplest obtained model (model with the fewest number of parameters). 3) The highest proportional decrease of error term when TF model residual standard error was compared with ones of univariate ARIMA model of the same response variable (the decrease of error term was seen due to inclusion of convenient exploratory variables into the model).

Missing values and outliers Before modeling, occasional missing values are replaced with appropriate values identified by the program included in the SCA software. These values are averages of two or more adjacent observations, depending on the series stationary and periodicity. During the modeling, four types of outliers (additive and innovational outliers, level shifts and temporary changes) are detected and adjusted in the fitted models by the software. Depending on their nature, outliers may have a substantial impact on an analysis, and better modeling and estimation are obtained by detecting and adjusting them. In fitted model outlier detection, SCA tests the series residuals against a prespecified critical value, which is dependent on the underlying model and the sample size. Therefore, only broad guidelines can be provided for a general choice of the critical value. In practice, the value 3.0 provides reasonable sensitivity to outliers. Lower sensitivity is provided by using larger critical values, and vice versa. The value 3.0 is recommended when the number of observations is between 120 and 250 (Liu and Hudak 1992), and it was used in our analyses. ECOGRAPHY 26:5 (2003)

Table 1. Identified transfer function models, initial estimates of the parameters with standard errors, t-values and p-values. Coefficients of determination for the models are calculated with r2 =1−[(n−1)/(n−p)][(SSresid.)/(SStotal)], where n =number of observations and p = number of estimated parameters. Time-series are NAO = seasonal NAO index in 1961–2000, RUNOFF= total Baltic Sea freshwater runoff in 1961–1998, SALINITY =seawater salinities at 0–25 m in 1961–1997, and TEMLON =T. longicornis abundance at 0–25 m in 1961–1991. All the time-series are quarterly means. For more detailed description see text. r2 =0.71, n =152 (1−B4)RUNOFFt =v0(1−B4)NAOt+(1−U4B4)/(1−f1B)at Estimate 3.29 0.81 0.38 SE 0.95 0.05 0.08 t-value 3.46 16.50 5.46 p-value 0.001 B0.001 B0.001 B. Salinity 0–25 m vs tot runoff r2 =0.90, n =148 (1−B4)SALINITYt=(v2+v3/1−dB)(1−B4)RUNOFFt+(1−U4B4)/(1−f1B)at Estimate −0.62 −0.86 −0.92 0.74 0.59 SE 0.28 0.29 0.03 0.07 0.08 t-value −2.19 −3.04 −34.81 10.95 7.72 p-value 0.029 0.002 B0.001 B0.001 B0.001 C. T. longicornis 0–25 m vs salinity 0–25 m r2 =0.48, n =123 (1−B4)TEMLONt =vt(1−B4)SALINITYt+(1−U4B4)at

A. tot Baltic runoffs vs seasonal NAO

Estimate SE t-value p-value

1.73 0.43 4.04 B0.001

Results

surface salinity with a lag of 1 –3 months (Table 1C, Fig. 2C). None of the series indicated trends during observed period. Predictions demonstrated that runoffs have remained at a high level while the decrease of surface salinities has leveled off during the latter part of 1990s. This indicates that the abundance of T. longicornis should also remain exiguous for some time to come (Fig. 2).

The TF models were seen to fit well with the observed series (Table 1). In the original time-series a high seasonal NAO was generally related with increased runoff in the Baltic Sea catchment area, which, subsequently, showed inverse coupling with Baltic Sea salinity. The effect was extended to the abundance of T. longicornis, which generally reflected the changes in salinity. This resulted in the observed high abundance of the copepod from 1960s until the early 1970s, with a decline thereafter. All the substantial modeled parameters showed statistical significance, and coefficient of determination values (r2) varied between 0.48 and 0.90, which are considered satisfactory in statistical time-series analysis. Total freshwater runoff to the Baltic Sea followed changes in the NAO with an immediate lag (shorter than resolution of data) and was directly connected with the NAO (positive t-value, i.e., high NAO index also denotes increased runoff) (Table 1A, Fig. 2A). Seawater surface salinity showed a non-linear and inverse response to freshwater runoff as the model attained rational form and showed negative t-values for observed lags (Table 1B, Fig. 2B). The model revealed that during increased runoff, salinities began to significantly decrease (or vice versa) already during a lag of 4– 6 months (as the resolution of data was quarters) but the major response was not attained before a lag period of 7 –9 months (higher t-values for latter lags). Immediately thereafter no response was detected. Furthermore, T. longicornis abundance directly followed changes in ECOGRAPHY 26:5 (2003)

0.84 0.05 15.41 B0.001

Discussion Our results provide the first mathematical models, and their statistical testing, of a remote control of marine pelagic biota by climatic and subsequent hydrological factors in the Baltic Sea area, here exemplified by the NAO and the freshwater runoff, respectively. The existence of a climatic control of the marine pelagic ecosystems has been duly demonstrated in the North Sea (e.g. Aebischer et al. 1990, Fromentin and Planque 1996) and in the Pacific (Roemmich and McGowan 1995). Recently, the first statistical models for response of Calanus finmarchius populations to climate variability in northwest Atlantic have been presented, as well (e.g. Conversi et al. 2001). In the Baltic Sea area, several early oceanographers have suggested that salinity changes control ecological interrelationships e.g., species distribution and pelagic food chain (Ka¨ ndler 1949, Hela 1951, Purasjoki 1953, Segerstra˚ le 1969, Lumberg and Ojaveer 1991). The decisive role of salinity in controlling Baltic Sea biodiversity is most likely due to 675

Fig. 2. Time series studied. On the left side of each panel are presented model fit scatterplots (observed values in X-axis, estimated values in Y-axis). On the right side, modeled (black spots) and observed changes (open circles) in the time series, and predictions (gray spots with 95% confidence intervals) based on the models. Letter next to each panel refers to corresponding model in Table 1. A) Total freshwater runoff (m3 s − 1) from the Baltic Sea catchment area. B) Baltic Sea surface salinity (PSU; 0 – 25 m) in ICES sub-division 28. C) Temora longicornis abundance in above thermocline surface water (Ind. m − 3, 0 – 25 m) in ICES sub-division 28 (note that values at the right are backtransformed from ln(x) -transformation). Smooth lines are drawn with distance-weighted least squares method.

676

ECOGRAPHY 26:5 (2003)

its causing osmotic stress to fauna and flora. In the case of copepods there is even a field study demonstrating lethality of low salinity for Calanus spp. in a convincing combination with laboratory studies (Zajaczkowski and Legez; yn´ ska 2001). However, all such observations have been scattered, anecdotal and lack statistical corroboration. While the NAO was recently shown by Fromentin and Planque (1996) to be associated with the North Sea mesozooplankton, it is not surprising that, in our case, a copepod species in the Baltic Sea pelagic ecosystem was behaving in a parallel way. The existence of this chain of events was already hypothesized by Ha¨ nninen et al. (2000), although no biological data were included in their modeling. The controlling mechanism is conceptually described as a series of events having a general climatic cause with various consequences that are modeled in this paper, in hydrological, hydrographic and biological terms. During the years of high NAO in the 1980s, there was a substantial increase in the rainfall over an area extending from northern Germany and the British Isles over Scandinavia (Førland et al. 1996). An increase in runoff and decrease in salinity mirrored this in the Baltic Sea hydrology and hydrography. Similarly, during the years with low NAO index values, from the early 1950s to the late 1970s salinities generally remained at high level in the Baltic Sea. This study corroborated that these changes were reflected also in pelagic biota, here exemplified by one copepod species. Our example, although carried out with one copepod species only, implies profound changes in the entire pelagic food chain, as the copepod used, T. longicornis is a central prey of the planktivores of the Baltic Proper (Flinkman et al. 1998, Flinkman 1999). Other species, which could have been included, are Centropages hamatus and Pseudocalanus elongatus. Indeed, these species also showed the same initial increase in abundance, and then the decrease during the low salinity years in late 1970s and the early 1980s (Wasmund et al. 1996, Ojaveer et al. 1998, Vuorinen et al. 1998). We have shown a direct and predictable cascade of effects from Atlantic climatic factors down to single plankton species in the pelagic ecosystem of the Baltic Sea. The result of a change in salinity, and its cascading effect in e.g., mesozooplanktonic copepods is fundamental, and has been shown to effect directly, or indirectly the whole pelagic ecosystem (for a general discussion see Flinkman 1999). The pelagic ecosystem in the Baltic includes the Baltic herring and sprat as planktivores, and the cod and salmonids as primary carnivores. Baltic Sea cod, herring and salmonids have drastically decreased in abundance in late 1970s and early 1980s, and since then their abundances have remained low, while a reverse trend has been recorded for sprat (Anon. 2001). Increasing abundances of the latter species are hypothesized to result from relief of predation pressure by Baltic cod (Flinkman 1999). Indeed, our results could theoretically have been affected ECOGRAPHY 26:5 (2003)

by increasing selective predation by sprat, but even in that case, the ultimate cause behind changes in biota is changing salinity. The effect of salinity to Baltic cod reproduction is a direct one, i.e. sinking and subsequent loss of eggs due to low buoyancy in low saline water (e.g., Flinkman 1999). Therefore, decreasing predation by cod on sprat is also due to the salinity change, albeit indirectly. Flinkman (1999) has discussed growth changes in the Baltic herring in detail, and it appears again that they may be both direct and indirect. Rajasilta et al. (1999) presented the idea that clupeids in general, and especially Baltic herring, may directly adapt to low salinity by decreasing its growth. Furthermore, a change in the diet (fewer copepods in stomachs) was documented in 1990s by Flinkman et al. (1998) with a resulting decrease in the growth. Even for zooplankton e.g. T. longicornis, and other copepods the effect may be both direct and/or indirect; through a combination of osmotic stress and food limitation. Phytoplankton quantity and quality has an effect on copepod growth (e.g., Koski et al. 1998), thus the documented increasing eutrophication of the Baltic Sea (e.g., Elmgren 2001) may have contributed to changes in pelagic copepods. Pelagic food chains have certainly been altered due to this, but at present, and especially with the present small study, we have no way of judging the relative importance of eutrophication, or more generally, of concurrently operating top-down and/or bottom-up controls of the pelagic ecosystem. Furthermore, the predictive character of the transfer function models renders them of profound importance for the understanding, future management and conservation of the coastal seas, e.g., the Baltic Sea. Our predictions were in this case relatively short-term only, extending just a few months or years ahead of time. This is due to shortness of the time lag, which was a period of only a few weeks from a change in atmospheric pressure to an alteration in the precipitation and runoff of the Baltic Sea watershed. A longer time period of some months is required for the effect to reach salinity of the Baltic Proper. One way to increase the predictive span would be to look at factors that are controlling changes in the NAO itself. Work on this aspect has barely just begun (a recent review by Visbeck et al. 2001). There is a hypothesis that sea surface temperatures could be used for this, but with little evidence so far. However, a possibility exists (Ulbrich and Christoph 1999, Visbeck et al. 2001), that anthropogenic climate change might influence variability modes of the NAO, with a hypothetically high mode being favored by an intensified greenhouse effect. Thus if we accept, for the sake of argument, the possibility that the present high mode of NAO is anthropogenically affected, we may also then deduce that the effects of global change started to be seen in Baltic Sea hydrography and biota already in 1970s with the increase in runoff and resulted in the decrease of salinity, and some its consequences described here. 677

Aside from the large-scale Atlantic climate effect, which might be at least partly anthropogenic, further research is needed on other factors that are more traditionally considered as manmade (pollution, eutrophication, overfishing, etc., for a review see Jansson 1997). Furthermore, studies should be done on factors that are traditionally expected to control the function of pelagic ecosystem, such as top-down or bottom-up controls (for a discussion concerning the Baltic Sea, see Flinkman 1999). A more thorough analysis would be required comprising all the dominant plankton and planktivore species before any detailed conclusions about the relative importance of driving forces can be made. Nevertheless, our study points out that even in pelagic environments, and especially in relatively shallow coastal sea areas the abiotic factors may be remote, but still extremely influential. Acknowledgements – This study is a contribution to the BASYS project (Baltic Sea System Study) in the MAST III Programme of the European Commission DG XII Contract MAS3-CT96-0058, and also to the AqValue project (Aquatic Biodiversity Research Programme/Finnish Biodiversity Research Programme FIBRE) of the Academy of Finland and Maj and Tor Nessling Foundation. We would also like to thank Christian Mo¨ llmann for some of the references, Juha Flinkman for some of the data, and Kevin O’Brien for checking the English of the manuscript.

References Aebischer, N. J., Coulson, J. C. and Colebrook, J. M. 1990. Parallel long-term trends across four marine trophic levels and weather. – Nature 347: 753 –755. Anon. 1968. Zooplankton sampling. Monographs on oceanographic methodology, 2. – UNESCO Press, NY, USA. Anon. 2001. Environment of the Baltic Sea area, 1994 – 1998. – Balt. Sea Environ. Proc. 82A: 1 –24. Box, G. E. P. and Newbold, P. 1971. Some comments on a paper of Coen, Gomme, and Kendall. – J. Roy. Stat. Soc. A Sta. 134: 229 –240. Box, G. E. P. and Jenkins, G. M. 1976. Time series analysis: forecasting and control, (rev. ed.). – Holden-Day, USA. Conversi, A., Piontkovski, S. and Hameed, S. 2001. Seasonal and interannual dynamics of Calanus finmarchicus in the Gulf of Maine (northeastern US shelf) with reference to the North Atlantic Oscillation. – Deep-Sea Res. 48 (Pt. II): 519 – 530. Dickson, R. R. and Brander, K. M. 1993. Effects of a changing windfield on cod stocks of the North Atlantic. – Fish. Oceanogr. 2: 124 –153. Elmgren, R. 2001. Understanding human impact on the Baltic ecosystem, changing views in recent decades. – Ambio 30: 222 – 231. Flinkman, J. 1999. Interactions between plankton and planktivores of the northern Baltic Sea: selective predation and predation avoidance. – Walter and Andre´ e Nottbeck Found. Sci. Rep. 18: 5–24, Ph.D. thesis. Flinkman, J. et al. 1998. Changes in northern Baltic zooplankton and herring nutrition from 1980s to 1990s: top-down

678

and bottom-up processes at work. – Mar. Ecol. Prog. Ser. 165: 127 – 136. Førland, E. J. et al. 1996. Changes in ‘‘normal’’ precipitation in the North Atlantic region. – DMNI Rep. No. 7/96, Klima, Oslo, Norway. Fromentin, J.-M. and Planque, B. 1996. Calanus and environment in the eastern North Atlantic. II. Influence of the North Atlantic Oscillation on C. finmarchicus and C. helgolandicus. – Mar. Ecol. Prog. Ser. 134: 111 – 118. Ha¨ nninen, J., Vuorinen, I. and Hjelt, P. 2000. Climatic factors in the Atlantic control the oceanographic and ecological changes in the Baltic Sea. – Limnol. Oceanogr. 45: 703 – 710. Hela, I. 1951. On the occurrence of the jellyfish Aurelia aurita L. on the south coast of Finland. – Arch. Zool. Soc. ‘‘Vanamo’’ 6: 71 – 78. Jansson, B.-O. 1997. The Baltic Sea: current and future status and impact of agriculture. – Ambio 26: 424 – 431. Ka¨ ndler, R. 1949. Die ha¨ ufigkeit pelagisher Fisheier in der Ostsee als masstab fu¨ r die Zu- und Abnahme der Fischbesta¨ nde. – Kieler Meeresforsch. VI: 73 – 98. Koski, M., Klein Breteler, W. and Schogt, N. 1998. Effect of food quality on rate of growth and development of the pelagic copepod Pseudocalanus elongatus (Copepoda, Calanoida). – Mar. Ecol. Prog. Ser. 170: 169 – 187. Liu, L. M. and Hudak, G. B. 1992. Forecasting and time series analysis using the SCA Statistical System. Vol. 1. – Scientific Computing Associates, IL, USA. Lumberg, A. and Ojaveer, E. 1991. On the environment and zooplankton dynamics in the Gulf of Finland in 1961 – 1990. – Proc. Estonian Acad. Sci. Ecol. 1: 131 – 140. Ojaveer, E., Lumberg, A. and Ojaveer, H. 1998. Highlights of zooplankton dynamics in Estonian waters (Baltic Sea). – ICES J. Mar. Sci. 55: 748 – 756. Pankratz, A. 1991. Forecasting with dynamic regression models. – Wiley. Purasjoki, K. 1953. Beobachtungen u¨ ber die Einwirkung gesteigerten Salzgehalts auf das Auftreten einiger marinen Zooplanktonarten ausserhalb Helsinki. – Arch. Zool. Soc. ‘‘Vanamo’’ 8: 101 – 104. Rajasilta, M. et al. 1999. Intensive monitoring of spawning populations of the Baltic herring (Clupea harengus membras L.). – Final report of the EU study project ref. No 96-068, 1997 – 1999, Univ. of Turku, Finland. Roemmich, D. and Mcgowan, J. 1995. Climatic warming and the decline of zooplankton in the California Current. – Science 267: 1324 – 1327. Rudstam, L. G., Aneer, G. and Hilde´ n, M. 1994. Top down control in the pelagic Baltic ecosystem. – Dana 10: 105 – 129. Segerstra˚ le, S. G. 1969. Biological fluctuations in the Baltic Sea. – Prog. Oceanogr. 5: 169 – 184. Ulbrich, U. and Christoph, M. 1999. A shift of the NAO and increasing storm track activity over Europe due to the anthropogenic greenhouse gas forcing. – Clim. Dynam. 15: 551 – 559. Visbeck, M. H. et al. 2001. The North Atlantic Oscillation: past, present and future. – Proc. Nat. Acad. Sci. USA 98: 12876 – 12877. Vuorinen, I. et al. 1998. Proportion of copepod biomass declines together with decreasing salinities in the Baltic Sea. – ICES J. Mar. Sci. 55: 767 – 774. Wasmund, N. et al. 1996. Pelagic biology. – In: Anon. (ed.), Third periodic assessment of the state of the marine environment of the Baltic Sea, 1989 – 1993. Balt. Sea Environ. Proc. 64B: 89 – 100. Zajaczkowski, M. J. and Legez; yn´ ska, J. 2001. Estimation of zooplankton mortality caused by an Arctic glacier outflow. – Oceanologia 43: 341 – 351.

ECOGRAPHY 26:5 (2003)