This study is based on high frequency time-series data from 1994 to 2019 obtained from the Arendal Station 2 which is located in the northern Skagerrak, 1 nautical mile offshore at 75 m depth (58°23′N 8°49′E, Figures 1 and 2 ). The monitoring station is operated by the Institute of Marine Research in Norway. The Skagerrak is part of the north-west European Shelf and forms a transition area between the Baltic Sea and the North Sea proper. While the North Sea is a shallow shelf area, with an average depth of approximately 80 m, Skagerrak has as a mean depth of 200 m. The Skagerrak is directly connected to the Norwegian Sea via the Norwegian Trench, which extends southward from the Norwegian Sea and follows the Norwegian coastline into the center of Skagerrak where it reaches a maximum depth of 710 m. The predominant circulation in Skagerrak is anti-clockwise and is influenced by three main water masses of different origin. Warm, saline Atlantic water (salinity >35) enters the North Sea from the north and follows the western slope of the Norwegian Trench into Skagerrak, as the Norwegian Trench Atlantic Inflow (NTI; Furnes et al., 1986). Waters from the southern North Sea (Jutland Current, salinity < 34) carry relatively cold and fresh water along the west coast of Denmark and mixes with brackish water from the Baltic Sea, forming the Norwegian Coastal Current (NCC). The NCC flows along the Norwegian coast, transporting water out and northward from the Skagerrak and North Sea. Most of the water entering the North Sea shelf flows through Skagerrak before leaving, and many of the hydrographic events taking place in the North Sea will thus be reflected in this region. The coastal area of northern Skagerrak is strongly influenced by the brackish NCC and the outflow of freshwater from rivers, and the hydrographic conditions at the Arendal Station 2 are characterized by strong seasonal cycles in both salinity and temperature (Sætre et al., 2003; Albretsen et al., 2012). The site is influenced by relatively fresh coastal waters (25–32) in the upper 30 m and by more saline Skagerrak and North Sea water (32–35) deeper in the water-column. Surface water temperature ranges from 0°C to 20°C, with the annual minimum in March and maximum in August. The freshwater outflow combined with thermal heating results in the development of a pronounced pycnocline at 15-20 m from March to October. However, wind driven coastal upwelling is common along the coast, which may bring high salinity Atlantic water to the surface. A detailed description of the current system and hydrography in the North Sea-Skagerrak is given by Otto et al. (1990), Rohde (1996) and Sætre et al. (2003). A spring phytoplankton bloom, dominated by diatoms, usually starts between February and April and typically lasts for 2-4 weeks. Chlorophyll values are generally low during summer, followed by a smaller autumn bloom of dinoflagellates in August–September (Dahl and Johannessen, 1998; Dahl et al., 2005).

Plankton at the Arendal sampling site has generally been monitored once to twice a month since 1994. Zooplankton was sampled by vertical hauls from 50 m to the surface, using a standard WP2 net (specified in UNESCO, 1968) with mesh-size 180 µm. Each sample was split into two equal parts with a Motoda splitter (Motoda, 1959). One half of the sample was preserved in 4% borax buffered formaldehyde-seawater solution for species identification and enumeration, and the other half used for biomass measurements (not presented here). Calanus spp. were enumerated and identified to copepodite stage. In order to determine the proportions of C. finmarchicus and C. helgolandicus in the samples, 20 individuals of Calanus for each of the copepodite stages CV, CVI females and CVI males were identified to species according to Fleminger and Hulseman (1977). CV and female CVI of the two species were discriminated based on the morphology and teeth of the basipod of the fifth pair of swimming legs. Male CVI were identified on the basis of the relative lengths of the endopod and exopod of the fifth leg. The reliability of discriminating C. finmarchicus and C. helgolandicus as CV and CVI copepodites by morphology has been confirmed by molecular methods (Lindeque et al., 2006). The abundance of each species was obtained by multiplying the total number of Calanus spp. within each developmental stage, by the stage-specific proportions. Copepodite stages CI-IV were counted, but for these stages the two species cannot be separated morphologically (Frost, 1974; Fleminger and Hulsemann, 1977).

Sampling for Calanus was generally carried out once to twice a month (irregular intervals). To regularize the time series, data for the different developmental stages, sexes and species of Calanus spp., were averaged to monthly values for each year. When no sampling took place within a month, data for the month were imputed using the mean of the preceding and following month. For the different stage-resolved monthly time series, 12 to 18 missing values were imputed, which represents 4-6% of the 300 monthly values. If zero counts of individuals were found in the sample(s) of a month for each of the time series, these were not imputed, retaining the zero count observations. Subsequently, three more time series were created by aggregating (i) copepodite stages CI to IV for Calanus spp. (‘Calanus spp. I-IV. I-IV’), (ii) stages CV, CVI male and CVI female for Calanus finmarchicus (‘C.fin V-VI’) and (iii) stage CV, CVI male and CVI female for Calanus helgolandicus (‘C.hel V-VI’). These three aggregated time series are the ones for which long term trends and periodicities are explored in the present study. Throughout the 1994-2019 study period, for each of the time series, the percentage of zero counts was 1.33%, 1.67% and 2.67 for Calanus spp. I-IV, C.fin V-VI and C.hel V-VI, respectively.

Each of the three time-series was subjected to Seasonal and Trend decomposition using Loess (STL) (Cleveland et al., 1990). STL was selected over other methods as it allows the seasonal component to change over time and the smoothness of the trend-cycle to be user-defined. A preliminary examination of the data revealed that the seasonal component was not constant over time. A 5-year window for the trend smoother was chosen, while the seasonal window was set to 12 months, thus allowing for variable amplitude and/or timing of the seasonal cycle. The selected 5-year trend smoother captured relatively well the long-term changes observed in the abundance of each aggregated time-series. Wider smoother-windows (e.g., 10 years) were tested, but led to over-smoothing, thus missing patterns with evident high/low abundance values, whereas lower values (e.g., 1 year) led to a very wiggly smoother from which long-term trends could not be easily discerned.

A Generalised Additive Model (GAM) (Hastie and Tibshirani, 1990; Wood, 2017) was used to evaluate potential changes in the seasonal patterns and abundance of the two species throughout the study period, for each of the three time series. The Tweedie distribution was chosen for its ability to handle zero values in the dataset through the parameter p which defines the model distribution. When 1<p<2, the model is a Poisson mixture of Gamma distributions. Let Y(t) be the observed abundance of the copepods in the studied time series. We assume that:

where ϕ is the dispersion parameter and the mean µ_i is linked to the linear predictor through a known link function g,

where X_i is a vector of covariates and β is a vector of unknown regression parameters. GAM fitting was done by setting the family argument to ‘tw’ for automatic estimation of the parameter p and generalized cross validation (GCV) was used to determine the optimal model fit. A tensor product smoother using ‘Month’ and ‘Year’ was applied to the abundance of each time series. All analysis was carried out in R 4.1.1 (R Core Team, 2021). For the STL decomposition and the GAM fitting the ‘fable’ (0.3.1, O'Hara-Wild et al., 2021) and ‘mgcv’ (Wood, 2017) packages were used, respectively. GAM plots were created using the ‘ggplot2’ (Wickham, 2016) and ‘mgcViz’ (Fasiolo et al., 2020) packages.

The seasonal variation in abundance of Calanus spp. was pronounced, and differences in the seasonal dynamics between Calanus finmarchicus and C. helgolandicus were evident. In Figures 3 and 4 , data from all years have been pooled and thus show the general seasonal patterns over the period studied. Due to pooling of data for all years, between-year differences in timing must be assumed to result in a wider distribution than would be the case for any single year.

Young Calanus copepodites of stages CI-III generally appeared in late February ( Figure 3A ). These stages displayed a marked pulse that peaked from mid-March to mid-April (medians of ~ 3350-8000 ind. m^-2, for the upper 50 m) and was strongly reduced by early May. This pulse of young copepodites belonging to the new generation of the year (G1) was well timed with the spring phytoplankton bloom, which occurred before the temperature in upper waters showed any notable increase. Still, the peak of these young copepodites showed a slight delay of about two weeks compared to the phytoplankton peak ( Figure 3A ).

Calanus finmarchicus of stage CV and adults (pooled) showed increasing numbers already from the beginning of January, a peak during mid-April to mid-May (~ 2500-3200 ind. m^-2), and thereafter decreasing numbers ( Figure 3B ). Still, these older stages of C. finmarchicus remained present in moderate concentrations until early autumn. The overall distribution of these stages indicates that the older part of the C. finmarchicus population reached its peak about one month after the phytoplankton bloom had collapsed, and the population started to decrease about 4 months before the temperature had reached its maximum.

Calanus helgolandicus of stage CV and adults (pooled) were present throughout the year, though with their main occurrence during July-November with a long-lasting peak from about August to October (medians of ~ 1100-2400 ind. m^-2, Figure 3C ). The seasonal population distribution of older stages of C. helgolandicus mirrored the seasonal temperature cycle, with the highest abundances occurring when temperature was peaking. The highest abundance was at least one month delayed relative to the autumn phytoplankton bloom. Moderate increases of C. helgolandicus, however, were also noted in April-May when water-temperature was far lower (5-8°C).

Adult females of Calanus finmarchicus were practically absent from the station in winter until early January ( Figure 4 ). Following a further modest increase in late January, the number of females rose markedly until reaching the annual maximum in late March (median of ~ 600 ind. m^-2). Some variability in the spring abundances of adult females was evident, and in early May the concentrations were almost at the same level as in late March. Whether the spring distribution for abundance of C. finmarchicus adult females represents one single but somewhat irregular peak, or a composite of more peaks with differing timing is not clear ( Figure 4 ). From mid-May the numbers decreased, and from July and for the rest of the year the levels were low (< 80 ind. m^-2), except for a small increase in August. The seasonal development of the generations is carefully reviewed in the “Discussion” section.

Adult males of C. finmarchicus were present in low numbers at the station as early as January. Even if the medians were unchanged during January and February (8 ind. m^-2), some years had higher abundances of males towards the end of February. ( Figure 4 ). Thereafter the abundances generally remained at zero ind. m^-2 or a very low level until March. The annual maximum of adult males occurred in the first half of May (median of 45 ind. m^-2) ( Figure 4 ). Male abundances then decreased, and remained at lower levels until mid-August, indicating a small new peak in July-early August. From October and for the rest of the year, C. finmarchicus adult males were practically absent.

C. finmarchicus copepodites of stage CV occurred in two annual peaks, the first peak very pronounced while the second peak far more subtle ( Figure 4 ). CV were present in low numbers in winter (< 90 ind. m^-2 during early September – mid April), their numbers thereafter increased rapidly to the annual maximum in early May (median of ~ 2700 ind. m^-2). Abundances then fell abruptly to lower levels, thereafter increasing very slowly to reach a minor second peak in early July (median of ~1200 ind. m^-2).

C. helgolandicus CV and adults displayed a rather synchronous seasonal pattern at the station. The species was present throughout the year and observed in low and variable numbers from January to July, with somewhat elevated occurrences of CV during April - early May being the most noticeable feature during the first half of the year ( Figures 3 and 4 ). A simultaneous and strong increase in abundance of all three stages occurred from July to late August. C. helgolandicus CV reached its annual maximum (~1350 ind. m^-2) in late August and showed elevated abundances until December. The peaks for adults occurred during August-September, followed by decreasing numbers, and low levels in November and December.

Calanus spp. stages CI-IV were present in low numbers during late December - late February (medians < 80 ind. m^-2 of CI-IV pooled) ( Figure 4 ). A strong increase in the abundance of CI occurred in early March, and the annual maximum for this stage was reached in the last half of March. Copepodite stages CII-III showed slightly delayed patterns compared to that of CI, with abundances peaking in the first half of April. Copepodite stages CI-III were present in variable though lower numbers also later in the season, but without showing clear and consistent secondary peaks. The seasonal pattern for stage CIV was similar to the patterns for CI-CIII, but with a little delay. Maximum abundances of stage CIV occurred in early April – early May. CIV also displayed considerable levels later in the year, with a rather wide secondary peak around July.

The decomposition of the three aggregated Calanus time series (i.e., Calanus spp. stages CI-CIV, C. finmarchicus CV-VI and C. helgolandicus CV-VI) into trend and seasonal components is shown in Figure 5 . Bars at the left side of each panel represent the relative scales of the components. The variation in size of the grey bars stems from the different scales of the respective plots; if the y-axis of each plot were set to the same range, the grey bars would have the same size. The short bar in the seasonal plots indicates that for all three time series, the seasonal cycle is explaining more of the observed variation in the data than the trend component (longer bars in the trend plots). The residuals (lowermost panels) show that there is substantial variation in the time series that is not explained by either the seasonal nor the long-term trend component (similar size to the grey bar of the ‘Data’ plot).

For Calanus spp. stages CI-IV no unidirectional trend was found. A period with high abundances was observed between 2003 and 2012, with a slight decrease in the middle of this period ( Figure 5A , second panel). A few years with relatively high abundances were also observed at the beginning of the time series, whereas in the last decade the trend was generally decreasing.

Nor did we observe any unidirectional long-term trend for the abundance of C. finmarchicus CV-VI ( Figure 5B , second panel). However, strong interannual variability was observed, with alternating periods of low and high densities. A period with high abundances was observed around 2009-2011, while much lower values were observed around the years 2001 and 2015. The annual abundances as represented by the trend for C. finmarchicus varied by a factor of 5 between high (~ 2010) and low (~2001) periods.

C. helgolandicus CV-VI ( Figure 5C , second panel) had low abundance at the beginning of the time period. The population showed a marked increase from 2000 to 2003 and remained at a high level until 2014. The very last years, the levels decreased.

The seasonal trend-decomposition revealed that the seasonal component in all three time-series varied over time ( Figure 5 ). Calanus spp. CI-IV ( Figure 5A , third panel) showed a regular seasonal pattern, however with a period of higher amplitudes during 2002-2008. Since 2008 the amplitude of the peaks has declined. In C. finmarchicus CV-VI ( Figure 5B , third panel) the spring peak dominated throughout the time series, however with varying amplitude. The variation in the amplitude mirrored the interannual trend in abundance, suggesting that the trend in C. finmarchicus is mainly related to variations in the spring abundances. In C. helgolandicus (CV-VI) one single annual peak was observed during the first years of the time-series ( Figure 5C , third panel). However, since 2002 a bimodal seasonal pattern emerged, with a small spring peak, and a stronger peak later in the year with increasing amplitude.

Figure 6 shows the modelled seasonal dynamics over the years carried out with the GAM. This makes it possible to examine if there have been interannual changes in phenology and abundance during the time-period. The model suggests that there were two annual peaks in Calanus spp. CI-IV (deviance explained = 56.4%, Figures 6A, B ). The seasonal timing of the spring peak has been consistent over the years, regularly occurring in mid-March. In contrast, the model suggests that the timing of the secondary peak has advanced by one month during 1994-2018, moving from July to June ( Figure 6A ).

One strong annual peak in C. finmarchicus CV-VI was predicted by the fitted model (deviance explained = 61.5%), with large interannual variability in abundances ( Figures 6C, D ). The fitted model suggests a gradual change in the timing of the annual peak in C. finmarchicus CV-VI, which has advanced by about one month during the time series.

A strong annual peak in C. helgolandicus CV-VI in autumn (August-September) was predicted by the model (deviance explained = 63.4%), with increasing abundances over the years ( Figures 6E, F ). Concurrent with the increasing abundances, the model indicates an earlier timing of the autumn peak over time, suggesting an advancement from mid-September to mid-August. Moreover, the model suggests that the seasonality of C. helgolandicus has changed from a unimodal to a bimodal pattern. In the early years of the time series, one single annual peak of C. helgolandicus was found, whereas after around 2002 an additional small peak appeared, occurring in mid-April. This is also evident from the time-series decomposition model shown in Figure 5C .

The Arendal time series revealed strong interannual variability in the abundances of C. finmarchicus and C. helgolandicus, however with clear interspecific differences in their temporal patterns. This suggests that the population sizes of the two species are affected by different mechanisms, as has previously been shown by Montero et al., (2021). While a generally increasing trend, apart from the very latest years with decreasing levels, was observed for C. helgolandicus, no unidirectional long-term trend was detected in C. finmarchichus. In fact, the abundance of C. finmarchicus has, based on expectations that this species is decreasing in the North Sea and Skagerrak area and retreating northwards (Beaugrand et al., 2002; Chust et al., 2014), remained surprisingly unchanged over time. However, intermittent shorter periods with high and low abundances were observed.

The lack of a consistent long-term trend in C. finmarchicus at this monitoring site is in contrast with previous studies based on CPR data, showing a long-term decrease in occurrences of this species in the North Sea during the last decades. Since the 1960s the biomass of C. finmarchicus has declined by 70% in the North Sea (Fromentin and Planque, 1996; Reid et al., 2003) which has been related to rising temperatures. The most dramatic reduction in C. finmarchicus occurred in the late 1980’s, often referred to as the North Sea regime shift (Beaugrand, 2004; Alvarez-Fernandez et al., 2012). The Arendal time series was initiated after the major decline of Calanus in the North Sea, and the abundances at this site prior to the 1980’s are not known. A reduction in the C. finmarchicus population at this site may thus have taken place prior to the initiation of the time-series presented here, which may explain why no consistent long-term decline was observed. If so, we might speculate that C. finmarchicus now is at a lower level at this site compared to the decades before the present time series.

A period with high abundances of C. finmarchicus was observed around 2009-2011. Comparing our data with the CPR data for the relevant standard area (B1) reveals similar patterns for C. finmarchicus in years after 1994 (Montero et al., 2021) with peak abundances in 2010. Similarly, a period with a higher ratio of C. finmarchicus to C. helgolandicus was reported from the North Atlantic CPR data during 2009-2011, which was related to years of low winter North Atlantic Oscillation index (NAO; Edwards et al., 2014). Indeed, the period of high abundances of C. finmarchicus reported here coincides with a period of strong negative NAO index in winters 2009/2010 and 2010/2011, reflected in below average SST in the NE Atlantic (González-Pola et al., 2020).

C. finmarchicus populations in the North Sea are not self-sustained and are highly dependent on the inflow from the Norwegian Sea (Heath et al., 1999; Gao et al., 2021). Variability in C. finmarchicus abundances in the North Sea have previously been linked to the NAO (Fromentin and Planque, 1996) which again is correlated with the inflow of water to the North Sea and the Skagerrak (Winther and Johannessen, 2006; Hjøllo et al., 2009). In the late 1980’s, this relationship broke down, which was explained by a decline in the overwintering stock in the southern Norwegian Sea and the Faroe Shetland Channel, and a northward shift of the species as a result of ocean warming (Heath et al., 1999; Reid et al., 2003). In the southern Norwegian Sea, a reduction in the abundance of C. finmarchicus and C. hyperboreus have been observed after 2003, and the abundances have remained at lower levels since then (Kristiansen et al., 2019; Skagseth et al., 2022). These changes have been attributed to reduced inflow of subarctic and arctic waters into the Norwegian Sea from west, which in turn has been linked to changes in the North Atlantic subpolar gyre (Kristiansen et al., 2019; Skagseth et al., 2022). However, the underlying mechanisms for the variations in C. finmarchicus population size in the North Sea and Skagerrak are still poorly understood and both large scale physical processes as well as local drivers need to be considered (Papworth et al., 2016). Recent studies suggest that the decoupling of NAO in the late 1980’s mainly occurred in western areas, while in the northeastern North Sea the fluctuations in C. finmarchicus remain related to NAO (Montero et al., 2021). In agreement with this, Gao et al. (2021) concluded that the inflow of C. finmarchicus biomass into the North Sea through the Norwegian Trench, accounted for 41% of the biomass in the North Sea and thus is a strong driver for the interannual variability of this species. Furthermore, Maar et al. (2013) suggested that the annual abundance of C. finmarchicus in the North Sea is sensitive to the degree of overwintering within the North Sea, because it allows individuals of this species to utilize the spring bloom more efficiently and independently of the timing and amount of oceanic inflow. It has also been suggested that the inflow of C. finmarchicus through the Norwegian Trench can facilitate favorable nursing conditions to the spring spawning North Sea cod (Huserbråten et al., 2018). The Norwegian Trench is thus of significance both as a gateway for the transport of Calanus to the area (Gao et al., 2021), as well as a habitat for local overwintering (Heath et al., 2004).

An increasing trend in the abundance of C. helgolandicus at the Arendal station was found until about 2013, being especially pronounced during 2000-2003. This is in accordance with numerous studies based on CPR data, reporting that this species has become more abundant and widely distributed in the North Sea over the past 50 years (Fromentin and Planque, 1996; Beaugrand, et al., 2002; Reid et al., 2003; Chust et al., 2014). Similar observations have been made at fixed point monitoring stations around the North Sea: At Helgoland (southern North Sea), C. helgolandicus has been increasing since the 1980’s, while in waters off Stonehaven (Scotland) an increase was not observed until 1999 (Bonnet et al., 2005). In contrast, Maud et al. (2015) found no temporal change in the abundance of C. helgolandicus at the Plymouth L4 site (English Channel) during 1988-2012. This suggests spatial differences in the dynamics of C. helgolandicus within the North Sea, implying that the species is more sensitive to environmental changes nearer the edge of its geographical range (Helaouët et al., 2013; Maar et al., 2013). The expansion of C. helgolandicus in the North Sea has been attributed to increased oceanic temperatures due to climate change (Reygondeau and Beaugrand, 2011) and temperature has been considered as the main factor affecting growth and population dynamics in C. helgolandicus (Møller et al., 2012; Montero et al., 2021). Sea temperatures in Skagerrak have increased by 1 C° since 1990 (Albretsen, 2012). We suspect that higher water temperatures have contributed to the increase in C. helgolandicus at this site.

The present study indicates a shift in the phenology of Calanus spp. during the last 25 years at the northern Skagerrak study site. Since 1994 the annual autumn peak of Calanus spp. CI-IV seems to have advanced by about one month. In contrast, we observed no changes in the seasonal timing of the spring peak in Calanus spp. CI-IV. It should be kept in mind that the copepodite stages CI-IV were not identified to species, and the autumn peak was probably a mixture of both species.

Moreover, the models suggest an earlier timing of the annual peaks for both C. finmarchicus CV-VI and C. helgolandicus CV-VI. This is in line with studies reporting a shift towards earlier seasonal timing in a number of zooplankton taxa in the North Sea during the last decades (Edwards and Richardson, 2004; Mackas et al., 2012; Reygondeau et al., 2015). These changes have been related to rising temperatures, suggesting that plankton tend to follow an ‘earlier when warmer’ strategy (Mackas et al., 2012). Also, in the more northern Norwegian Sea, a small change in the timing of C. finmarchicus has been observed, where the spring peak was suggested to occur four days earlier (Dupont et al., 2017). The phenology of C. finmarchicus is strongly correlated with temperature (Mackas et al., 2012), and in the central North Sea the seasonal peak of this species has advanced by 10 days during 1958-2002 (Edwards and Richardson, 2004; Bonnet et al., 2005). Similarly, at Helgoland (southern North Sea), C. helgolandicus has shifted to an earlier timing by at least 1 month during the years 1975-2006 (Mackas et al., 2012). An earlier seasonal timing in Calanus species has been related to shorter development time with increasing temperatures (Cook et al., 2007) or to variations in the duration of the spring phytoplankton bloom (Aßmus et al., 2009).

Variations in timing and phenology of Calanus spp. may also be caused by advection of cohorts from other areas (Espinasse et al., 2018). The Arendal sampling site is located in a highly advective system, with a large degree of temporal variation (Rodhe, 1996). Consequently, advective supply of Calanus spp. to the site is expected to play an essential role in contributing to the observed temporal variations. However, despite the high temporal variability, spatial homogeneity has been documented for the area (Dahl and Johannessen, 1998). The observations made at the Arendal sampling site can thus be considered as representative for a larger area of the Skagerrak.

An important result of his study is the appearance of a small spring peak in C. helgolandicus at the Arendal station. The fitted models suggest a change in the seasonal pattern of C. helgolandicus, switching from a unimodal to a bimodal pattern around year 2002, when a small spring peak emerged. A similar change in the seasonality of C. helgolandicus has been described from other regions of the North Sea. Beare and McKenzie (1999) observed a change from one to two annual peaks in C. helgolandicus in the northern North Sea, where a small spring peak occurred during 1980-1999. In the western North Sea (Stonehaven time series), a small spring peak in C. helgolandicus developed during the late 1980’s (Bonnet et al., 2005; Wilson et al., 2015). Strand et al. (2020) reported spring occurrences of C. helgolandicus along a section in the northeastern North Sea based on CPR data from 2008-2016. Our study not only supports these findings, but also reveals that this spring peak appeared around the year 2002. We suggest that C. helgolandicus in the northeastern North Sea is approaching a seasonality reported as typical in more southerly areas (Bonnet et al., 2005).

The seasonal cycle of C. helgolandicus is mainly driven by the temperature regime, and different cycles have been observed for different areas (Bonnet et al., 2005). In warmer regions (southern North Sea and the Mediterranean), the peak in annual abundance occurs earlier in the year compared to the northern North Sea which is considered to be related to optimal temperature for fecundity and hatching (Bonnet et al., 2005; Wilson et al., 2016). Surprisingly, the small spring peak for C. helgolandicus at the Arendal station occurs in April-May when temperatures are below 8°C. Experimental studies suggest that C. helgolandicus is not able to develop from eggs to adults at these temperatures (Bonnet et al., 2009; Møller et al., 2012). The presence of C. helgolandicus at these low temperatures may indicate an adaption to lower temperatures as were suggested by Møller et al. (2012). However, the occurrence of C. helgolandicus in spring may also be explained by advective transport of spring populations from more southern and warmer regions in the North Sea. Analysis of possible relationships between the population dynamics and environmental conditions is outside the scope of this study. Whether the observed shifts in phenology is a response to changes in local conditions, or to large scale variations in advective processes will be addressed in future studies.

In order to reveal possible mechanisms behind phenological changes in the Calanus sibling species, further studies should consider detailed demographic data and multiple phenological indices. Moreover, extended sampling to full ocean depth, including the deeper areas of the Norwegian Trench, would provide a more complete picture of the Calanus populations in this area.

The observed changes in annual abundance and phenology of Calanus spp. at the Arendal station mainly took place around the years 2000 – 2005. During the same period a regime shift was identified in the North Sea (late 1990’s - early 2000’s; Alvarez-Fernandez et al., 2012; Beaugrand et al., 2014). Furthermore, several studies have suggested that major shifts took place in the coastal waters of the Norwegian Skagerrak in the early 2000’s, including a shift in stoichiometry (Frigstad et al., 2013), a change in the macroalgae community (Moy and Christie, 2012), and recruitment failure of coastal gadoid species (Johannessen et al., 2012). Even though this station is located at the shelf off the Norwegian coast and may thus not be representative for the larger North Sea, it might represent zooplankton changes that have taken place in Skagerrak and along the south coast of Norway. The present time-series may thus be more relevant indices, potentially explaining recruitment variability and failure in coastal fish populations, among others the decline in coastal cod populations (ICES, 2021).