Discussion
The present study identifies regime shifts in a long-term metric of biological productivity in relation to regime shifts in potential causal drivers, using a Bayesian online change point detection (BOCPD) algorithm. Juvenile Atlantic cod in Norwegian coastal Skagerrak experienced two stepwise reductions in mean abundance, beginning in 1975 and 1999, ultimately attaining a level ~30% of what it was in the first half of the 20th Century. Regime shifts in indices for the NAO and zooplankton abundance since 1960 were directionally the same, the opposite, or independent of directional shifts in cod catch rates. The degree of temporal overlap between regimes shifts in cod abundance and water temperature depended on season. Single regime shifts of increased mean temperature (by 1 to 20C) during the winter and spring months either preceded the 1999 cod regime shift by more than a decade or occurred several years thereafter. However, summer-autumn temperature regime shifts (1 to 20C increase) were either concomitant with, or occurred slightly in advance of, the cod regime shift (decrease) in 1999. The earlier cod regime shift (1975) was not associated with a regime shift in water temperature for any month of the year.
The relative importance of hypothesized drivers can be ascertained by the temporal proximity of their regime shifts with regime shifts in cod. It is clear that shifts in some factors, such as the NAO index and abundance of C. finmarchicus , do not always have obvious biological consequences for Atlantic cod. Against the background of temporal changes in fishing mortality and spawning stock biomass of North Sea cod, the earliest regime shifts in NAO (1961; a decline from 0.2 to -2.0) and C. finmarchicus (1982; a ~75% decline) were not linked with regime shifts in cod (Fig. 3a). The same was true of the 1-20 C regime-shift increase in water temperature during the winter-spring months in 1988.
One interpretation of this lack of influence is that drivers of cod productivity are less likely to manifest biological change when (i) they act singly, (ii) human-induced mortality is relatively low, and (iii) cod population size is relatively high. When the NAO shifted in 1961, fishing mortality on North Sea cod was less thanFlim and SSB was 1.36Blim ; when zooplankton abundance shifted downwards in 1982, fishing mortality was increasing but population biomass remained high (1.53 Blim ). Regarding the winter-spring increase during the regime shift in water temperature that began in 1988, it is notable that, despite these increases, temperatures remained well within the range thought to be optimal for cod eggs and larvae (Nissling 2004; Righton et al. 2010). Indeed, these January-June regime shifts might have had a beneficial influence on cod productivity, mitigating to some extent any steadily increasing effects of persistently high fishing mortality and declining SSB.
Our analyses reveal two occasions when regime shifts in potential drivers of cod productivity preceded regime shifts in cod catch rate by a time period sufficiently brief (< 5 yr) that they could plausibly have influenced the subsequent abundance of cod aged 0+ to 2+ years (Fig. 6). Even though a decline in NAO in 1961 had no discernable effect on cod, the increase in 1972 might well have, insofar as the first cod regime shift followed two years later. This supposition is supported by studies (e.g., Stige et al. 2006) that have concluded that an increased NAO index has a negative influence on cod productivity in the northeast Atlantic. One possible reason for why the NAO apparently affected cod (beginning in 1974) is that the magnitude of the NAO regime shift (−2.0 to 1.5) was the greatest of the three shifts that occurred between 1864 and 2018. It is also notable, however, that the 1974 cod regime shift occurred during a period of steadily increasing and unsustainably high fishing mortality (1.5Flim ), potentially affecting the ability of cod to resist environmental changes caused by the NAO, changes to which an unfished population might have been resilient. This underscores the challenge in disentangling the effects of fishing and climate-related indices on biological productivity.
The cod regime shift that began in 1999 was preceded by a ‘perfect storm’ of multiple concomitant changes in the environment. Summer-autumn temperatures jumped 1-20 C; the NAO index declined from 1.5 to 0.5; C. finmarchicus had plunged to its lowest level in the time series; fishing mortality was at its highest level since 1963 (2.0 Flim ); and spawning stock biomass was at its lowest level in the time series (0.8 Blim ) en route to a minimum of 0.41 Blim in 2006. The effects of the NAO, C. finmarchicus , and temperature on cod productivity were undoubtedly accentuated by the directionality of their regime shifts in 1999. As the NAO index declines, so does primary and secondary productivity in Skagerrak (Tiselius et al. 2016), and increased water temperatures are associated with increasingly unfavourable conditions for C. finmarchicus (Fromentin et al. 1998).
Among the putative drivers of cod productivity, the 1996 regime shift in NAO may have been the most benign, given that (i) a reduction in the index did not have its anticipated positive effect on cod and that (ii) the index had returned to levels characteristic of the 1868 to 1960 period. The very considerable reduction in C. finmarchicus(beginning in 1997; Fig. 3) was likely a much more prominent factor, given the exceedingly low levels to which this key prey species of juvenile cod had declined.
There is, however, reason to believe that the summer-autumn increase in water temperature was of considerably greater importance than either the NAO or zooplankton abundance. Successive regime shifts from 1994 to 1999 during July through October raised temperatures to their highest recorded levels in coastal Skagerrak since 1925, when the time series began. In some years, mean August temperatures exceeded 200C, approaching the critical thermal maximum for Atlantic cod (Righton et al. 2010; Norin et al. 2019).
We hypothesize that cod did not respond positively to the presumed increased in food supply in 2008 because of the physiological stress associated with increased summer-autumn water temperatures. Based on tagging studies at sea of almost 400 cod from 8 northeast Atlantic cod stocks, Righton et al. (2010) found that although the total thermal niche of adult cod ranged between −1.5 and 19.00 C, the temperature range was considerably narrower during the spawning period when larval and juvenile cod are developing (1 to 80 C). Nissling (2004) reported that survival of larval cod in the laboratory declined considerably when water temperatures exceeded 100 C.
One fundamentally important element to consider when evaluating the consequences of climate-related and environmental regime shifts on population productivity is the size of the population relative to a metric of long-term sustainability, such as carrying capacity or population size in an unfished state. This is because small populations are more vulnerable to environmental stochasticity than comparatively large populations (Lande 1993). This link between population size and susceptibility to environmental change has been repeatedly considered when assessing the recovery capacity of depleted cod populations (Hutchings and Myers 1994; Hutchings and Kuparinen 2017). But it has also been made with respect to potential drivers of cod regime shifts. Based on an analysis of cod populations on the European Shelf south of 62oN, including North Sea cod, Brander (2005) concluded that environmental variability, as represented by the NAO index, only affects cod when the spawning stock biomass is low. Brander’s (2005) argument is both theoretically compelling and empirically supported by the present study.
There are several attributes to the methodology we have applied here. Firstly, the same algorithm is used to identify regime shifts in a metric of biological productivity and putative causal drivers of that metric. Secondly, our approach greatly reduces the subjectivity inherent in deciding the magnitude of data change which constitutes a regime shift (the ‘effect’ size) and when it is that a regime shift occurs; we did not presume the existence of a regime shift in any given year for any given variable. A third improvement is that the BOCPD algorithm accounts for changes in the variance in the data, not simply the mean.
One limitation in our interpretation of the relative importance of fishing and the environment on regime shifts in cod productivity is our use of estimates of fishing mortality and spawning stock biomass for North Sea cod as metrics of fishing pressure and population size for Skagerrak cod. But if we were to account for fishing mortality in our analyses, we needed to avail ourselves of the best available data in this regard, and these data were available for North Sea cod. There are empirically defensible reasons for our application of North Sea cod estimates of F and SSB to Skagerrak cod. Firstly, North Sea cod genotypes exist along the Norwegian Skagerrak coast (Knutsen et al. 2018). Secondly, Skagerrak has long been considered part of the North Sea cod stock unit (ICES 2019). Thirdly, limited estimates of fishing mortality available for Skagerrak cod confirm that fishing mortality can be exceedingly high. Kleiven et al. (2016) reported that recreational and commercial fisheries for cod in Skagerrak fjords resulted in a mortality rate of 55.6% for the years 2005 to 2013, equivalent toF =0.81. For comparison, the average F for North Sea cod over the same time period was 0.61 (ICES 2019), suggesting that the fishing mortality experienced by North Sea cod may be comparable to, and possibly less than, that experienced by Norwegian Skagerrak coastal cod in some years.
The concept of regime shifts permeates the marine ecological and fisheries literature. Definitions vary considerably. The ecological literature tends to interpret regime shifts as community-level changes between alternative stable states with the implication that such shifts are difficult to reverse (Conversi et al. 2015; Ling et al. 2015). In contrast, regime-shift analyses of meteorological factors tend not to focus on alternative stable states, being much more accepting of regime-shift ‘reversibility’ (e.g., Dippner et al. 2014; Jaagus et al. 2017). The fisheries literature is perhaps intermediate with respect to the question of regime-shift reversibility. Some work draws attention to long-term, slow-to-reverse discontinuities in ecosystem properties (Möllmann et al. 2009), whereas neither reversibility nor regime-shift time period have been integral to a lack of temporal stationarity in fish-stock productivity (e.g., Vert-pre et al. 2013; Perälä and Kuparinen 2015).
Our analyses emphasize the utility in examining multiple regime shifts when trying to understand the causal mechanisms responsible for regime shifts in metrics of biological productivity. Doing so allows one to formulate hypotheses and to draw conclusions concerning the conditional probabilities that an environmentally related regime shift will affect biological productivity. One hypothesis that emerges here is that the strength of the effect of an environmental or climate-related regime shift is accentuated when it coincides with other regime shifts. A second hypothesis, underscoring the findings of previous work (Brander 2005; Hutchings and Kuparinen 2017), is that climate-related regime shifts are more likely to affect populations when they are relatively small. The present study affirms the dominant role that fishing has on the probability that populations will respond to regime shifts in environmental variables, underscoring the fundamental necessity of accounting for fishing mortality in any analysis of regime shifts in commercially exploited marine fishes (de Young et al. 2004; Möllmann et al. 2009).
For our case study of Norwegian Skagerrak cod, our work suggests that steadily increasing fishing mortality from commercial and recreational fisheries has increasingly sensitized the cod to regime shifts in NAO, zooplankton abundance, and water temperature. Fishing mortality remains unsustainably high in the region (Kleiven et al. 2016). This, coupled with small population size and increased summer and autumn water temperatures that broach the thermal limit for the species, are likely major factors limiting the recovery capacity for cod in southern coastal Norway.