Philosophical Transactions of the Royal Society B: Biological Sciences
You have accessResearch article

Synchronous marine pelagic regime shifts in the Northern Hemisphere

G. Beaugrand

G. Beaugrand

Centre National de la Recherche Scientifique, Laboratoire d'Océanologie et de Géosciences’ UMR LOG CNRS 8187, Station Marine, Université des Sciences et Technologies de Lille 1, Lille 1 BP 80, Wimereux 62930, France

Google Scholar

Find this author on PubMed

,
A. Conversi

A. Conversi

Institute of Marine Sciences, National Research Council of Italy, Forte Santa Teresa, Loc Pozzuolo, Lerici, La Spezia 19032, Italy

SAHFOS, Sir Alister Hardy Foundation for Ocean Science, The Laboratory, Citadel Hill, The Hoe, Plymouth PL1 2PB, UK

Centre for Marine and Coastal Policy Research, Marine Institute, Plymouth University, Plymouth PL4 8AA, UK

[email protected]

Google Scholar

Find this author on PubMed

,
S. Chiba

S. Chiba

RIGC, JAMSTEC, 3173-25 Showa-machi, Kanazawa-ku, Yokohama 236-0001, Japan

Google Scholar

Find this author on PubMed

,
M. Edwards

M. Edwards

Institute of Marine Sciences, National Research Council of Italy, Forte Santa Teresa, Loc Pozzuolo, Lerici, La Spezia 19032, Italy

SAHFOS, Sir Alister Hardy Foundation for Ocean Science, The Laboratory, Citadel Hill, The Hoe, Plymouth PL1 2PB, UK

Google Scholar

Find this author on PubMed

,
S. Fonda-Umani

S. Fonda-Umani

Department of Life Sciences, University of Trieste, v. Giorgieri, 10, Trieste, Italy

Google Scholar

Find this author on PubMed

,
C. Greene

C. Greene

Ocean Resources and Ecosystems Program, Cornell University, Ithaca, NY, USA

Google Scholar

Find this author on PubMed

,
N. Mantua

N. Mantua

Southwest Fisheries Science Center, National Marine Fisheries Service, 110 Shaffer Road, Santa Cruz, CA 95060, USA

Google Scholar

Find this author on PubMed

,
S. A. Otto

S. A. Otto

Stockholm Resilience Centre, Stockholm University, Kräftriket 2B, Stockholm 106 91, Sweden

Institute for Hydrobiology and Fisheries Science, Center for Earth System Research and Sustainability (CEN), KlimaCampus, University of Hamburg, Grosse Elbstrasse 133, Hamburg 22767, Germany

Google Scholar

Find this author on PubMed

,
P. C. Reid

P. C. Reid

SAHFOS, Sir Alister Hardy Foundation for Ocean Science, The Laboratory, Citadel Hill, The Hoe, Plymouth PL1 2PB, UK

Centre for Marine and Coastal Policy Research, Marine Institute, Plymouth University, Plymouth PL4 8AA, UK

Marine Biological Association of the UK, The Laboratory, Citadel Hill, Plymouth PL1 2PB, UK

Google Scholar

Find this author on PubMed

,
M. M. Stachura

M. M. Stachura

School of Aquatic and Fishery Sciences, University of Washington, Box 355020, Seattle, WA 98195, USA

Google Scholar

Find this author on PubMed

,
L. Stemmann

L. Stemmann

LOV, Observatoire Océanologique de Villefranche-sur-Mer, Sorbonne Universités, UPMC Univ Paris 06, France

Google Scholar

Find this author on PubMed

and
H. Sugisaki

H. Sugisaki

Fisheries Research Agency, 2-3-3, Minatomirai, Nishi-ku, Yokohama, Japan

Google Scholar

Find this author on PubMed

    Abstract

    Regime shifts are characterized by sudden, substantial and temporally persistent changes in the state of an ecosystem. They involve major biological modifications and often have important implications for exploited living resources. In this study, we examine whether regime shifts observed in 11 marine systems from two oceans and three regional seas in the Northern Hemisphere (NH) are synchronous, applying the same methodology to all. We primarily infer marine pelagic regime shifts from abrupt shifts in zooplankton assemblages, with the exception of the East Pacific where ecosystem changes are inferred from fish. Our analyses provide evidence for quasi-synchronicity of marine pelagic regime shifts both within and between ocean basins, although these shifts lie embedded within considerable regional variability at both year-to-year and lower-frequency time scales. In particular, a regime shift was detected in the late 1980s in many studied marine regions, although the exact year of the observed shift varied somewhat from one basin to another. Another regime shift was also identified in the mid- to late 1970s but concerned less marine regions. We subsequently analyse the main biological signals in relation to changes in NH temperature and pressure anomalies. The results suggest that the main factor synchronizing regime shifts on large scales is NH temperature; however, changes in atmospheric circulation also appear important. We propose that this quasi-synchronous shift could represent the variably lagged biological response in each ecosystem to a large-scale, NH change of the climatic system, involving both an increase in NH temperature and a strongly positive phase of the Arctic Oscillation. Further investigation is needed to determine the relative roles of changes in temperature and atmospheric pressure patterns and their resultant teleconnections in synchronizing regime shifts at large scales.

    1. Introduction

    Regime shifts are sudden, substantial and temporally persistent changes in the state of communities/ecosystems [1,2]. They involve large-scale reorganizations in the structure and function of the biological components in ecosystems [35]. Regime shifts are often examined by investigating a few key indicators of ecosystem state. Important examples of regime shifts in large marine ecosystems have been reported for both the Northern Hemisphere (NH) [611] and the Southern Hemisphere [12,13]. As studies focusing on pelagic regime shifts around the world increased and became more global, researchers began to examine potential synchronicities in the variation of species and ecosystems that are often separated by great distances, and are even in different ocean basins [1,8,1425].

    Three periods of apparent synchronicity stand out when reviewing the literature of pelagic marine regime shifts: the late 1970s, the mid- to late 1980s and the late 1990s. Regime shifts have been reported for the North Pacific (1976/1977 and 1989) [6,2629], specifically the northeast Pacific (1976/1977) [30,31], (1998/1999) [32,33] and the northwest Pacific (1976/1977, 1988/1989 and 1998) [3436]; the Humboldt Current in the southern Pacific (1968/1970 and 1984/1986) [37,38]; and the northwest Atlantic (1989/1990) [3943]. In addition, regime shifts were observed in the late 1980s around all European seas [8]: the North Sea [44,45], the Baltic Sea [9,46], the northwest European shelf seas (1987/1990) [47,48], the western [49] and eastern (Adriatic) Mediterranean Sea [8,50,51] and the Black Sea [52,53]. Other shifts have been observed in the late 1990s in the northeast Atlantic and its adjacent seas, in the Black Sea, and in San Francisco Bay [5457] at the time of a global shift in temperature [19]. Note that the late 1970s regime followed an abrupt change in the large-scale boreal winter circulation pattern over the North Pacific during the mid-1970s, which affected the thermal regime of these oceanic regions [58,59].

    The drivers of the regime shifts are also under debate. Investigating the late 1980s regime shifts reported in most European seas, Conversi et al. [8] suggested that such apparent synchronicities in widely separated marine systems might be explained in two ways: (i) due to random coincidence or (ii) due to regional-, basin- or hemispheric-scale manifestations of large-scale climatic patterns. In particular, they proposed that the synergistic effects of changes in temperature and atmospheric pressure patterns could provide the means for transmitting the effects of atmospheric regime shifts and climate change to ecosystems in the pelagic realm.

    Möllmann & Diekmann [11] in their review of the NH regime shifts conclude that multiple drivers, such as climate and overfishing, may interact in triggering ecosystem regime shifts, while other studies explain planktonic shifts as stochastic noise resulting from the biological integration of the external physical variability [60,61]. Reid & Beaugrand [19] observed that in many cases the reported ecological regime shifts coincided with major temporal changes seen in marine temperature anomalies and suggested that temperature may be an important synchronizing agent.

    Here, we further explore the hypotheses associated with synchronicity, by analysing the timing of regime shifts for the first time on a hemispheric scale, using the same methodology on marine zooplankton and fish data from ecosystems ranging from the Mediterranean Sea to the western Pacific Ocean, and by addressing two questions:

    1. Are there quasi-synchronous ecological shifts across different regions worldwide?

    2. Are there large-scale mechanisms that can force synchronous ecological reorganizations around the NH?

    To answer these questions, we assembled datasets from 11 regions across the NH, collected over multiple decades (1960–2005) in two oceans, the Atlantic and the Pacific, and three seas, the North, Baltic and Mediterranean Seas (figure 1 and table 1; the electronic supplementary material). We then first quantitatively characterized long-term patterns, including regime shifts, in the multivariate biological states of each region. This biological state is inferred from zooplankton time series in all systems investigated, with the exception of the eastern and western North Pacific Rim where it is characterized from shifts in species of salmon. The time series chosen for study were the only ones available starting before the 1980s, i.e. of appropriate length for our multi-decadal investigation. Even if the data used in this investigation represent mostly one trophic level, it has to be noted that other studies have identified ecosystem-wide regime shifts in most of these areas [6,8,9,11,36,43,44,47].

    Table 1.Summary of biological multi-decadal time series used in this study. Detailed information per basin is given in the electronic supplementary material.

    basin area period frequency gaps source variables notes
    North Sea central North Sea 1958–2007 monthly no SAHFOS http://www.sahfos.ac.uk zooplankton CPR
    Baltic Sea central Baltic Sea 1959–2008 seasonal no Otto zooplankton Juday net
    western Mediterranean Sea Ligurian Sea (Villefranche) 1974–2003 seasonal no Stemmann zooplankton Zooscan imaging technique
    eastern Mediterranean Sea northern Adriatic Sea (Gulf of Trieste) 1970–2005 monthly yes (1981–1985) Fonda-Umani, Conversi zooplankton WP2 net
    western Atlantic northern area 1977–2011 seasonal no http://osprey.bcodmo.org/dataset.cfm?flag=view&id=13684 zooplankton Marmap bongo data
    western Atlantic southern area 1977–2011 seasonal no http://osprey.bcodmo.org/dataset.cfm?flag=view&id=13684 zooplankton Marmap bongo data
    eastern Pacific California Current 1951–2009 seasonal no California Cooperative Oceanic Fisheries Investigations: http://data.calcofi.org/zooplankton.html zooplankton CalCOFI
    eastern Pacific eastern Pacific Rim 1952–2005 annual no http://hdl.handle.net/1773/16262 [62] wild salmon
    western Pacific western Pacific Rim 1952–2005 annual no http://hdl.handle.net/1773/16262 [62] wild salmon
    western Pacific Oyashio Current 1960–2002 seasonal (April–June and July–September) yes (1969) Sugisaki, Chiba http://tnfri.fra.affrc.go.jp/eindex.html zooplankton NORPAC ring net
    western Pacific transition zone 1960–1999 seasonal (April–June and July–September) yes (1992) Sugisaki, Chiba http://tnfri.fra.affrc.go.jp/eindex.html zooplankton NORPAC ring net
    Figure 1.

    Figure 1. Geographical location of the marine ecological time series analysed in this study. (1) North Sea (CPR collection); (2) central Baltic Sea (Gotland basin, Latvian time series); (3) Ligurian Sea, western Mediterranean (Point B time series); (4) northern Adriatic Sea, eastern Mediterranean (Gulf of Trieste time series); (5) western Atlantic—northern area (Gulf of Maine and Georges Bank regions); (6) western Atlantic—southern area (New England and mid-Atlantic regions); (7) eastern Pacific—California Current (CalCOFI data); (8) eastern Pacific (Pacific Rim salmon data); (9) western Pacific (Pacific Rim salmon data); (10) western Pacific—Oyashio Current (ODATE collection); (11) western Pacific—Transition zone (south of the Oyashio Current; ODATE collection). (Online version in colour.)

    Second, we identified temporal shifts in all these systems, individually and in aggregate, and examined their relationships to (i) large-scale climatic indices (Northern Hemisphere temperature (NHT) anomalies, Arctic Oscillation (AO), Atlantic Multi-decadal Oscillation (AMO) and Pacific Decadal Oscillation (PDO)), and (ii) spatial patterns of change in sea surface temperature (SST) and sea-level pressure (SLP) over the NH.

    2. Data and methods

    (a) Data

    The biological time series used in this study covered a total of 11 regions located in two oceans, the Atlantic and the Pacific, and three regional seas, the North, the Baltic and the Mediterranean Seas. Figure 1 shows their geographical locations, and table 1 summarizes information on the time series. Detailed information on the marine systems and their associated time series is provided in the electronic supplementary material.

    We used four large-scale temperature- or pressure-based hydro-climatic (or climatic) indices to investigate the relationships between long-term biological changes and climate. NHT anomalies originated from the Hadley Centre (www.cru.uea.ac.uk) (HadCRUT3: Hadley land and sea combined temperature anomalies). The AO index signifies the strength of the polar vortex, with positive values signifying anomalously low pressure over the Arctic and high pressure over the Pacific and Atlantic at a latitude of roughly 45° N [63]. This index is constructed by projecting the daily 1000 mb height anomalies north of 20° N, and was obtained from the Climate Prediction Center (http://www.cpc.noaa.gov/). The PDO index is defined as the first principal component of North Pacific monthly SST variability north of 20° N (http://www.jisao.washington.edu/pdo/). The AMO is an index of long-term (50–80 years) climate variability based on temperature in the range of 0.4°C in the North Atlantic [64]. We used the Extended Reconstruction sea surface temperature (ERSST) data averaged for the area 25–60° N and 7–75° W minus a regression of global mean temperature (National Climatic Data Center; http://www.esrl.noaa.gov/psd/data/timeseries/AMO/). Each of these indices track phenomena that have been assumed to play a role in some of the reported regime shifts [6,8,19,43,65].

    The SST was extracted from the ERSST_V3 dataset, which is derived from a reanalysis of the most recently available International Comprehensive Ocean–Atmosphere Data Set (ICOADS) SST data. Improved statistical methods were used to produce a stable monthly reconstruction from relatively sparse data [66]. For this work, we constructed a spatially gridded (2° latitude × 2° longitude) annual average dataset for the NH, corresponding to the period 1960–2005.

    The SLP was extracted from the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) Reanalysis project (http://www.esrl.noaa.gov/psd/data/gridded/reanalysis/). NCEP uses a climate model that is initialized with observations originating from a variety of sources (e.g. ships, planes and satellite observations). A spatially gridded (2.5° latitude × 2.5° longitude) annual average SLP dataset for the NH was constructed corresponding to the period 1960–2005.

    (b) Statistical analyses

    The statistical procedures used in this study are summarized in figure 2. We first performed a standardized principal component analysis (PCA) on each time series, applying an algorithm for missing data [67], as gaps were present in some time series. We retained the first three principal components from each PCA. This first analysis was performed on matrices (tables A1 to A11, see scheme in figure 2) with different observations (temporally or spatially structured) and biological variables (zooplankton or fish), with or without seasonal information (see the electronic supplementary material). The results from these analyses provided a three-dimensional summary of the biological state of the ecosystems represented by the time series. The examination of more dimensions did not provide substantial additional information.

    Figure 2.

    Figure 2. Schematic showing the main statistical analyses performed in this study. PCA, principal component analysis; PC, principal component; SST, sea surface temperature; SLP, sea-level pressure; r, correlation analyses. (Online version in colour.)

    We then combined the first three principal components (PCs) for each of the 11 ecosystems into a table corresponding to the 46 years (1960–2005) of this study × 33 PCs (11 time series × 3 PCs; scheme in figure 2 and details in electronic supplementary material, table B). The period 1960–2005 was selected to be certain that no year had more than 50% missing data. For each column of this matrix, we performed Taylor's [68] change-point analysis to detect abrupt shifts in each ecosystem and PC. First, the cumulative sums of the time series were calculated [69,70]. Second, the amplitude of the cumulative sums was estimated. Third, a Monte Carlo test was conducted and the probability of a shift was determined based on the number of times the simulated amplitude was superior to the observed amplitude. Approximately 100 000 runs were performed and the simulated time series were retained if their order-1 autocorrelation was superior or equal to the one observed in the original time series. As change-point analysis cannot be used on missing data, we interpolated the gaps by using eigenvector filtering [71]. These results are detailed in the electronic supplementary material and are synthesized in table 2 and in figure 3 to examine the issue of synchronicity.

    Table 2.Marine areas analysed, with number of variables used, percentage of variance explained by the first three principal components and year(s) of regime shift in each principal component. Variables are either species or taxonomic groups. The number of descriptors used in the PCA can be directly calculated by multiplying the number of time periods or areas by the number of variables. N, no shift detected at the threshold probability of 0.05.

    marine area year in PC1 (%) year in PC2 (%) year in PC3 (%)
    central North Sea (1958–2007)(14 variables) 27.96%1987 27.15%N 13.91%N
    central Baltic Sea (1959–2008) (4 seasons × 10 variables) 23.08%1988,1997 11.12%1972 8.66%N
    Ligurian Sea (1974–2003)(4 seasons × 6 variables) 30.53%N 13.86%N 11.21%N
    northern Adriatic Sea (1970–2005)(4 seasons × 12 variables) 17.70%1989 12.63%N 9.62%N
    western Atlantic (North) (1977–2011)(6 2-month periods × 24 variables) 15.01%19941998 10.57%N 8.05%1989
    western Atlantic (South) (1977–2011)(6 2-month periods × 17 variables) 19.31%19892001 11.07%N 7.98%N
    California Current (1951–2009)(4 seasons × 1 variable) 64.98%1989 21.08%N 9.67%19781992
    eastern Pacific Rim (salmon) (1952–2005)(8 areas × 3 species) 30.31%1979 13.04%N 9.97%1991
    western Pacific Rim (salmon) (1952–2005)(4 areas and 3 species) 27.77%19671984 20.63%N 15.94%195619701984
    Oyashio Current (1960–2002)(2 seasons × 32 variables) 17.32%1980 11.64%1976 8.89%N
    transition zone (1960–1999)(2 seasons × 67 variables) 15.26%1980 13.76%N 7.70%1979
    Figure 3.

    Figure 3. Summary of the results from change-point analysis performed on each of the first three principal components for each ecosystem (a total of 33 principal components). The number of significant shifts (p < 0.05) for each year is shown in the blue bar plot. The red curve above the bars indicates the number of shifts for a sliding 3-year period. For example, there were seven significant shifts centred around 1989–1990. (Online version in colour.)

    We also performed a second standardized PCA on table B of the electronic supplementary material, to reveal the main long-term patterns in the 11 ecosystems and to examine synchronicity (normalized eigenvectors or the correlation between each variable and the PCs; table 3).

    Table 3.Normalized eigenvectors from a standardized PCA performed on a table of 46 years (1960–2005) × 33 principal components (electronic supplementary material, table B). Numbers in bold are above 0.5. Because eigenvectors were normalized, they reflect the correlation between the long-term changes in the biological state of each ecosystem and the corresponding principal component.

    systems PC eigenvector 1 eigenvector 2 eigenvector 3
    central North Sea 1 −0.7853 −0.3958 −0.0997
    2 0.0283 −0.3241 −0.089
    3 −0.3243 0.5863 0.3985
    central Baltic Sea 1 0.8360 0.0389 0.1028
    2 −0.1922 −0.2157 0.8013
    3 0.1448 −0.0686 −0.2308
    Ligurian Sea 1 −0.4544 0.5534 0.4857
    2 −0.124 −0.3148 0.0239
    3 −0.4594 −0.1758 −0.341
    northern Adriatic Sea 1 −0.8765 −0.0403 −0.1619
    2 0.0372 0.7049 −0.1473
    3 0.201 −0.1633 0.0368
    western Atlantic (North) 1 0.6602 −0.239 0.5743
    2 0.4183 −0.1119 −0.1897
    3 −0.3891 0.197 0.0665
    western Atlantic (South) 1 −0.7445 0.4332 0.0169
    2 −0.2854 −0.2008 −0.6136
    3 −0.1894 −0.0167 −0.3479
    California Current 1 −0.7548 0.0623 0.2861
    2 0.1366 −0.4541 −0.0826
    3 −0.1129 0.5652 −0.0235
    eastern Pacific Rim (salmon) 1 0.5185 0.0239 0.6275
    2 0.5968 −0.2407 −0.2755
    3 −0.5999 −0.5649 0.0654
    western Pacific Rim (salmon) 1 0.557 0.6611 −0.0477
    2 −0.4932 0.1497 0.1901
    3 −0.5704 0.2834 −0.1808
    Oyashio Current 1 0.6194 0.0305 −0.2569
    2 −0.0735 0.7621 −0.408
    3 −0.2338 −0.3483 −0.0969
    transition zone 1 0.1969 0.3671 −0.0288
    2 −0.0494 −0.247 0.3093
    3 0.7144 0.2441 −0.2473

    We considered the PCs of this PCA to be indicators of the biological state (BioPC) of the NH. Relationships between the first three BioPCs and some key large-scale hydro-climatic indices, NHT anomalies, the AO index, the AMO index and the PDO index (figures 4 and 5; electronic supplementary material, table C), were investigated by means of a linear correlation analysis. The correlation probabilities were corrected to account for temporal autocorrelation by adjusting the degrees of freedom [72]. However, as this technique can be overly conservative, we also examined the uncorrected probabilities. The large difference between uncorrected (p) and adjusted (pACF) probabilities suggests that the correlation between two variables is mainly related to low-frequency (long-term) variability [73].

    Figure 4.

    Figure 4. Long-term changes in the first three NH principal components and in large-scale climatic indices for the period 1960–2005. (a) First bio-principal component and NHT anomalies. (b) Second bio-principal component and the PDO. (c) Third principal component and the AMO. The timing of three shifts, determined by a cluster analysis (figure 6), is superimposed on the figure (light grey).

    Figure 5.

    Figure 5. Long-term changes in the first principal component BioPC2 in relation to the PDO and the AO. A change in the correlation between BioPC2 and the oscillations appeared at the time of the late 1980s regime shift. The PDO and AO were correlated prior and after the shift, respectively.

    We used a cluster analysis (figure 6) to identify relatively homogeneous time periods and associated shifts in a table (electronic supplementary material, table C) containing the first three BioPCs and the four large-scale hydro-climatic indices (NHT, AO, PDO and AMO). We used the hierarchical flexible agglomerative clustering method proposed by Lance & Williams [74]. By fixing the values of the four parameters αj, αm, β and γ [74], it is possible to go from a single to a complete linkage. Here, αj was fixed to 0.625, αm to 0.625, β to −0.25 and γ to 0 so that the method was close to the unweighted centroid clustering (also called unweighted pair-group centroid method, see [75] for further details). This analysis was based on the Euclidean distance after standardization of each of the seven variables of table C in the electronic supplementary material between 0 and 1.

    Figure 6.

    Figure 6. Dendrogram on years originating from a cluster analysis performed on a matrix 46 years × 7 variables (variables of electronic supplementary material, table C, standardized between 0 and 1) using the Euclidean distance. Four periods were found at the threshold of 2: 1960–1976, 1977–1988, 1989–1998 and 1999–2005. (Online version in colour.)

    To understand the underlying mechanisms behind any potential synchronicities, we calculated the spatial point-by-point correlation patterns between gridded SSTs and the first three biological state PCs (figure 7). We also performed the same analysis with SLPs (figure 8).

    Figure 7.

    Figure 7. Spatial correlations between long-term changes in SSTs and changes in the first three biological state principal components for the period 1960–2005. (a) First bio-principal component. (b) Second bio-principal component. (c) Third bio-principal component. (Online version in colour.)

    Figure 8.

    Figure 8. Spatial correlations between long-term changes in SLPs and changes in the first three biological state principal components for the period 1960–2005. (a) First bio-principal component. (b) Second bio-principal component. (c) Third bio-principal component. (Online version in colour.)

    3. Results

    The PCAs were performed separately on data from all 11 regions, and the first three PCs were retained from each. The variance explained by each of these PC is shown in table 2. These percentages should be interpreted with caution as they also depend on the number of descriptors, which vary among regions (table 2). Change-point analyses were carried out for each PC in each region. A total of 23 shifts were detected in the 33 time series (11 regions × 3 PCs; table 2 and figure 3) tested. No shifts were identified for the Ligurian Sea (western Mediterranean Sea). A regime shift that took place mainly at the end of the 1980s (1987–1990) was detected across most regions (figure 3 and table 2): North Sea (1987); central Baltic Sea (1988); northern Adriatic Sea (1989), western North Atlantic, both southern and northern areas (1989); California Current (1989); and eastern North Pacific Rim (salmon) (1991). Another regime shift was detected near the end of the 1970s, ca 1976–1980, in the Pacific regions: California Current, Oyashio Current, Transition zone (south of the Oyashio Current) and eastern North Pacific Rim (salmon) time series. The earliest regime shifts were observed during the early 1970s in the western North Pacific Rim (salmon) (1967 and 1970) and in the central Baltic Sea (1972). Other shifts were observed in the late 1990s in other regions (central Baltic Sea (1997); western Atlantic—northern (1998) and southern (2001) areas), but were not synchronous.

    We next examined common long-term patterns in the biological state of all regions (figure 4 and table 3). A standardized PCA based on the first three PCs of each time series (a total of 33 PCs; figure 2) was applied to determine their combined main variability. This analysis provided three BioPCs, which summarize the combined biological state (data available in electronic supplementary material, table C).

    Long-term changes in the first BioPC (23.08% of the total variance) were correlated positively with NHT anomalies (r = 0.72; d.f. = 44; p < 0.01; pACF = 0.10). The large difference between uncorrected (p) and adjusted (pACF) probability suggests that the correlation between the two variables was mainly related to low-frequency (long-term) variability. This PC showed a pronounced shift that took place ca 1988 (figure 4a). All areas, except the Ligurian Sea, were related to the first BioPC (table 3).

    The second BioPC (13.22% of the total variance) showed long-term changes that were overall positively correlated to the PDO index (r = 0.46; d.f. = 44; p < 0.01; pACF = 0.11; figure 4b). However, a closer look shows that this correlation varied substantially over time. Prior to the 1988 shift, the correlation was significant (r = 0.76; d.f. = 27; p < 0.01; pACF < 0.05), while afterwards it disappeared. From 1988 onwards, the second BioPC was instead significantly correlated to the AO (r = 0.66; d.f. = 17; p < 0.01; pACF < 0.05; figure 5). The North Sea (PC3), Ligurian Sea (PC1), northern Adriatic (PC2), California Current (PC3), eastern (PC3) and western (PC1) Pacific Rim and Oyashio Current (PC2) were related to the second BioPC (table 3).

    The third BioPC (9.66% of the total variance) showed a pseudo-periodic signal that paralleled low-frequency changes in the AMO index (figure 4c). Correlations were however weak and not significant (r = 0.28; d.f. = 44; p = 0.06; pACF = 0.29). The central Baltic Sea (PC2), the western North Atlantic northern (PC1) and southern (PC2) areas, and the eastern North Pacific Rim (PC1) were related to the third BioPC (table 3).

    A cluster analysis (figure 6) was performed to identify common time periods for the seven variables identified above (first three BioPCs, NHT, AO, AMO and PDO; electronic supplementary material, table C; figures 2 and 6). The first partition was found at a Euclidean distance of greater than 2, identifying two time periods: (i) 1960–1988 and (ii) 1989–2005. At a Euclidean distance of 2, the period 1960–1988 was subdivided into 1960–1975 and 1976–1988 and the period 1989–2005 was subdivided into 1989–1998 and 1999–2005. The three breakpoints were, therefore, from the strongest to the weakest: 1988/1989, 1975/1976 and 1998/1999. Two of these breakpoints correspond to the two (1978 and 1989/90) main regime shifts observed for the individual PCs of each region (figure 3), and also to periods that have been widely reported in other regime-shift studies (see Introduction). The cluster analysis did reveal a further shift in 1998/1999, also documented in the Atlantic Ocean [55,56].

    Both long-term trends and regime shifts paralleled changes in large-scale climatic indices based on SST or SLP. We therefore investigated the spatial correlation patterns between SST, SLP and the first three BioPCs to try to elucidate mechanisms that could modulate NH-scale ecological reorganizations.

    We found that long-term changes in annual hemispheric SST were positively correlated to the first BioPC, over most of the investigated marine areas (figure 7a). The hemispheric relationship between SST and BioPC1 suggests common long-term patterns of change, and possibly a common response to forcing agents. It confirms the results from the positive correlation between BioPC1 and NHT, seen in figure 4a, and suggests that temperature is a synchronizing agent for BioPC1.

    The spatial patterns of the correlation between the second BioPC and SST, on the other hand, highlight the importance of oceanic circulation for the biological state: the correlation patterns appear in fact to mirror the main circulation systems in the North Atlantic and North Pacific Oceans (figure 7b).

    In addition, both BioPC1 and BioPC2 are strongly negatively correlated to SLP over the Arctic Sea (figure 8a,b). These correlations might indicate a specific role of the Arctic area, which has been shown to influence both North Pacific and North Atlantic ecosystems [76,77].

    The relationship between SST, SLP and BioPC3 is less obvious (figures 7c and 8c). It should be noted that only four areas are related to this BioPC, so it is less interesting for our investigation on multi-area synchronicities.

    4. Discussion

    In this paper, we applied for the first time the same set of statistical tools to a number of different marine ecosystems (using mainly zooplankton data) to investigate whether shifts occur at similar times around the world. We specifically addressed two questions.

    (a) Question 1: Are there quasi-synchronous ecological shifts across different regions worldwide?

    Our analyses indicate that there is some synchronicity between ecosystem changes (as inferred from zooplankton/salmon data), together with a large degree of regional variation. With the exception of the Ligurian Sea (western Mediterranean), where no shifts were detected, all regions exhibited one or more abrupt shifts in their PCs, and these shifts co-occurred in two or more time series in almost all cases (table 2). When all shifts were combined together (figure 3), three periods were evident: (i) between 1976 and 1978; (ii) between 1988 and 1990; and (iii) ca 1997. The most conspicuous was the shift about 1988.

    Slight changes in timing were evident from one analysis to another (figures 3, 6 and table 2). Such differences in timing may be real or artefacts of the statistical techniques employed. For example, a cluster analysis is meant to reveal time periods characterized by similar attributes and relatively little change, whereas a change-point analysis focuses on abrupt transitions between such periods in a time series. Therefore, the timing of regime shifts revealed by these techniques may often be slightly different. Similarly, visual inspection of a principal component (or a time series) is also subjective, and results may vary depending on whether one looks at the beginning of the new regime, the end of the previous regime or the middle of the transition [47].

    The late 1980s stands out as a time period when regime shifts were detected in many basins that are geographically spread far apart, including the Mediterranean Sea (northern Adriatic only), the North Sea, the Baltic Sea (1991), as well as the western Atlantic, and the Pacific Ocean. While the timing of the detected regime shifts is not identical, the quasi-synoptic occurrence of regime shifts in seven out of the 11 ecosystems analysed is remarkable (figure 3 and table 2). This finding provides some circumstantial evidence in support of the hypothesis that across-system synchronicity exists, embedded within the background of local variability.

    The late 1970s, especially 1976–1979, is another time period when regime shifts were detected. However, of the 11 systems analysed, only four exhibited regime shifts in this time period, and all were located in the Pacific Ocean. Therefore, there were far too few shifts to make a case for hemispheric-scale synchronicity. In the Pacific [26,30,58,78] the late 1970s regime shift involved both biological and climate variables, and was associated with climate forcing, while the late 1980s regime shift was found only in the biological variables [6].

    Few regime shifts were identified during the late 1990s in the datasets we analysed. This result is surprising as changes at this time have been reported from the North Atlantic and its adjacent seas in terms of the phytoplankton, zooplankton, fish and seabirds [55,56,79] and for fish and invertebrates in the eastern North Pacific (San Francisco Bay) [54]. These ecosystems, with the exception of the North Sea, were however not analysed in this study. These widespread changes have been attributed to changes in circulation and temperature associated to the strength of the subpolar gyres [54,79] and to global-scale changes in temperature at the end of the 1990s [19].

    The lack of shifts in the western Mediterranean in the late 1980s contrasts with what was observed by Molinero et al. [49], who found reduced concentrations of all zooplankton groups, except gelatinous plankton, after 1989 and attributed these changes to possible hydro-climatic changes triggered by changes in the North Atlantic Oscillation. On the other hand, our findings for this site are consistent with the observations of decadal oscillations in zooplankton found by García-Comas et al. [80] and Vandromme et al. [81]. They proposed that the long-term warming of seawaters in the northwest Mediterranean Sea was not a sufficient driver to trigger ecological shifts. By contrast, the nearly decadal-scale changes in surface salinity and/or cool winters found during ‘dry years’ allowed strong convection and nutrient input into the photic zone in the 1980s and after the 2000. These conditions enabled zooplankton communities to recover to previous levels, resulting hence in decadal oscillations rather than long-term shifts.

    Our analyses highlight a quasi-synchronicity in regime shifts in unconnected basins at the end of the 1980s, where 'quasi-synchronicity' means that many, but not all ecosystems present a shift at a similar, but not identical time. This synchronicity, although not perfect, is still remarkable if one takes into consideration the many differences in available data, sampling protocols and ecosystem variables. (i) Local hydro-climatic variability or anthropogenic effects (e.g. fishing) may exacerbate or mitigate the influence of large-scale forcing [19,82]. (ii) Statistical techniques used to detect regime shifts are very sensitive to sampling bias or analytical errors, such as taxonomic misidentification [7]. Sampling bias may arise from many factors such as uncertainties in the spatial (vertical or horizontal) and temporal locations of samples or the amount of seawater filtered [47,83]. For example, some of the time series used in this study are averaged from samples taken over large, offshore areas (e.g. CPR in the North Sea; ECOMON in the western Atlantic; CalCOFI in the California Current area), while others were collected from small, inshore areas, such as a coastal marine station (e.g. the Mediterranean time series). (iii) A third factor is related to mixing different taxonomic groups, with different life-history traits (e.g. plankton and fish), spatial distributions and ecological preferences. For example, species with long generation times may exhibit time lags in responses whereas those with fast generation times are more likely to display rapid responses to environmental forcing [47]. (iv) A further factor, explained by Beaugrand [84], is related to the nonlinear interaction between environmental change and the ecological niche of a species. For example, if the thermal regime of a region is close to the optimal part of a species thermal niche, then the sensitivity of that species to climate change will be small. On the other hand, if the thermal regime is close to the edge of the thermal niche of the species, the species will be sensitive to climate-induced temperature change and exhibit a rapid response [85,86]. This phenomenon explains why all species in an ecosystem do not exhibit a shift [45,87], and, if a limited number of species is analysed, it may also explain a shift in timing or a lack of response to a common large-scale hydro-climatic factor.

    (b) Question 2: Are there large-scale mechanisms that can force synchronous ecological reorganizations around the Northern Hemisphere?

    To address this question, we analysed the first three PCs originating from the PCAs performed for each ecosystem to reveal the NH biological state (BioPC), together with climate indicators (NHT, AO, PDO and AMO) and with long-term spatial variations in NHTs and SLPs. These variables were chosen because previous studies have suggested that temperature may affect the biogeographical distribution of many pelagic organisms [7,69], while changes in the main pressure centres can influence ocean circulation and the biological processes [41,43,8890], as well as local temperature [6].

    We searched for potential lags in the responses of all regions to large-scale hydro-climatic forcing but no lag was observed using cross-correlation analysis and cross-correlograms.

    The results of our analyses indicate multi-level and complex relationships between ecosystem state and NHTs. We note that the first BioPC of the ecosystem state, which captures the shifts in the mid- to late 1970s, late 1980s and end 1990s, and represents 23.08% of the total variance, is highly correlated with NHT anomalies (figure 4a). This result suggests a relationship between biological variability and hemispheric warming. All the regions but one (Ligurian Sea) were related to this component (table 3). As global temperature did not rise continuously, but increased in a stepwise manner in 1977, 1987 and 1997 [19], with an especially large rise in the late 1980s [91], it could be the primary driver behind all observed shifts. The spatial correlations between this PC and hemispheric SSTs (figure 7a) support the idea that hemispheric temperature trends may serve as a large-scale synchronizing agent. In fact, SSTs are highly and positively correlated with this PC in all basins. Such geographically extended correlations strengthen the case for teleconnections across ocean basins.

    How subtle changes in temperature may involve such substantial biological reorganizations has been investigated, and recent studies suggest that such shifts can arise from the nonlinear interaction between the ecological species niche (sensu Hutchinson) and temperature-induced environmental changes [84,9294]. Simulating a species assemblage having all possible thermal niches in the North Sea, these studies showed that when temperature increases (or decreases) rapidly, such changes lead to an abrupt community shift (ACS); the magnitude of the shift is higher when temperature changes are greater and when the degree of stenothermy of the species is higher. Such shifts should also be more likely at the boundary between biomes or provinces and near critical thermal boundaries [7]. This can be easily understood at the species level. At this organizational level, temperature is likely to trigger major changes in abundance when the regional thermal regime is at the edge of the thermal niche. For example, temperature change is likely to trigger a major change in the abundance of cod in the North Sea where the thermal regime is at the edge of the species' thermal niche, whereas only major temperature changes may trigger a shift in Iceland where the thermal regime corresponds to the centre of the species' thermal niche [85]. At the community level, it is likely that ACSs are more prominent at the boundary between biomes because there are often associations with rapid spatial temperature changes. In addition, changes in temperature have also an effect on phytoplankton dynamics (e.g. the onset of the spring bloom and the timing of peaks of individual phytoplankton species), which can affect zooplankton species differently. Hence, for some zooplankton species slight changes in temperature can be magnified via the phytoplankton.

    In addition to the role of temperature, both atmospheric and oceanic circulations have been shown to play a key role in the long-term changes of marine populations [6,43,79,95], and may also influence the regime dynamics of the areas we studied. Our results support the hypothesis that climate dynamics may have a relevant role. For example, the second component of the biological state (13.22% of the total variance), although not correlated with any single large-scale climatic index throughout its entire time series, was positively correlated with the AO after the late 1980s shift (figure 5). These results are similar to the correlation between the AO and the Arctic Ocean Oscillation, which became significant from the beginning of the 1980s onwards [42]. The spatial correlation patterns between SLP and the first two BioPCs (figure 8) also indicate an association between the biological state and the Arctic climate system.

    The links between interdecadal variability in the Arctic climate system, upper-ocean circulation patterns in the Arctic and Atlantic Oceans, and ecosystem regime shifts in the North Atlantic during the past three decades have been reviewed recently [42,43,96]. Changes in Arctic climate at the transitions between the decades of the 1980s and 1990s and between the decades of the 1990s and 2000s resulted in abrupt changes in the export of freshwater from the Arctic Ocean to the North Atlantic. The first of these resulted in the Great Salinity Anomaly of the 1990s, characterized by the discharge of relatively low-salinity water from the Arctic Ocean into the North Atlantic through the Canadian Arctic Archipelago and Fram Strait (between Greenland and Svalbard) [41,76]. For example, feedbacks from Arctic atmospheric climate have led to changes in the upper-ocean circulation of the Arctic Ocean between the late 1980s and early 1990s, which in turn have resulted in changes in North Atlantic circulation (increased discharge of relatively low-salinity water into the North Atlantic, and redirection of the shallow, low-salinity outflow from the Arctic Ocean mainly through the Labrador Sea [41,76]).

    The Pacific Ocean circulation is also influenced by the Arctic circulation [97] and by the AO. In fact, the AO is the major factor of SST increase over the subarctic North Pacific in 1988/1989 [98]. Its influence is more conspicuous in the western subarctic North Pacific than in the eastern North Pacific, affecting precipitations over Japan [99]. Some studies have reported AO-related SST change in western North Pacific ecosystems, for example shifts of spawning ground and biomass of the Japanese common squid (Todarodes pacifics) in the Japanese Sea [77].

    Finally, changes in the Arctic climate system, especially those affecting sea ice extent, have the potential to influence global climate through the Atlantic Meridional Overturning Circulation [96] and middle-latitude North Atlantic climate and weather systems through recently recognized atmospheric teleconnections [100102].

    The spatial correlations between BioPC2 and hemispheric SST suggest also a link with circulation, but at a regional scale. Figure 7b displays correlations with the main circulation patterns in both oceans: the Gulf Stream and the Labrador Current systems in the North Atlantic, and the Kuroshio Current and Alaska gyre systems in the North Pacific. In this case, SST should be considered a proxy for oceanic circulation at a regional scale, rather than for global warming. These findings highlight the complex associations between temperature and biological variability.

    These results lead to the question of what happened in the hemispheric climate (including temperature and atmospheric circulation) in the late 1980s. Large-scale changes took place during this period. Lo & Hsu [91] show that widespread abrupt warming occurred in the late 1980s in the extratropical NH, and that this warming was associated with a change in the relative influence of the PDO-like pattern and the AO-like pattern. In particular, the AO-like pattern had a dominant influence on NHT since the late 1980s, whereas the influence of the PDO weakened [91] (see also figure 5). Here we have shown that the NHT, AO, AMO and PDO indices are associated with the first three components of the biological state. It is thus reasonable to propose that they can act as synchronizing agents for biological variations at hemispheric scales.

    A final point regards the role of trophic cascades as drivers of regime shifts. Although ecological shifts have been attributed in some ecosystems to trophic cascade alterations owing to overfishing [39], it is unlikely that the widespread late 1980s regime shift is due to the indirect effect of this human activity on zooplankton in all regions. Indeed, it is difficult to envision that exploitation changed radically for different stocks in different systems at a similar time. However, as we had very limited fishery data available for this study, we cannot at this stage rule out the role of fishing on ecosystem shifts, especially at the single-basin scale, for example, for setting the preconditions for a shift.

    We also note that the first three biological components together account for approximately 47% of the total variability of the biological state, so the synchronization at best can only be partial. We propose that part of the remaining variability is probably associated with local factors (e.g. local hydrodynamics and anthropogenic impacts), which could explain the differences between basins. Noise associated with sampling may also be important.

    5. Conclusion

    This study provides evidence for quasi-synchronicity of ecological shifts both within and between ocean basins, although these shifts lie embedded within considerable regional variability at both year-to-year and lower-frequency time scales. We believe these shifts were mediated by an abrupt warming seen in the climate [19,91]. For example, a regime shift during the late 1980s was detected in many large marine regions around the NH, although the exact year of the observed shift varied somewhat from one basin to another. Regime shifts at other times (e.g. late 1970s and late 1990s) appear in some large marine systems, but they are not so widely detectable in this dataset.

    Relatively strong correlations are detected between long-term changes in a biological synthesis index (first BioPC; extracted mostly from zooplankton taxa and salmon for the Pacific Rim) and changes in NHT, SST and SLP. These correlations suggest that temperature may be the most important factor explaining the quasi-synchronicity of the shifts. From an ecological point of view, temperature affects both the ecological niche of the species and their biogeographical distributions; hence, changes in temperature at the hemispheric scale may lead to hemispheric-scale changes in the biota.

    Correlations with other parameters such as the AO and SLPs over the Arctic are weaker, but still significant. They do suggest that the Arctic climate system is also driving synchronous changes in the biology on a hemispheric scale. From an oceanographic perspective, it has been shown that changes in the Arctic climate system are affecting atmospheric and oceanic circulation in both the Atlantic and the Pacific Oceans, which in turn determine the thermal regime of an oceanic region. In addition, it has been reported that the Arctic circulation and pressure system affects ecosystems in the North Atlantic and Pacific. Therefore, it is plausible that these changes in Arctic climate play an important role in driving the dynamic regime of NH marine ecosystems.

    The fact that widely separated marine ecosystems respond quasi-synchronously (at least in the late 1980s) to climatic change indicates that there may be teleconnections between them. We propose that the late 1980s shift could represent the variably lagged biological response in each ecosystem to a large-scale, NH change of the climatic system, involving both an increase in NHT and a strongly positive phase of the AO [91].

    We believe that quasi-synchronicity in zooplankton/fish shifts observed in many oceanic basins can have major scientific implications. First, identifying biological teleconnections could certainly shed light on the extent of climate impacts on marine ecology and on physical circulation, and in particular would indicate that large-scale drivers (e.g. climate pressure variations, climate warming) have an important role on the systems, as opposed to local drivers (fishing, eutrophication, pollution) or stochastic noise. Second, teleconnections in the biological realm across marine systems are still not yet considered a possibility in mainstream science, and therefore are not incorporated in any biological or physical models, but they might be in the future.

    This initial study proposes hypotheses that will need further investigation with more biological systems and larger datasets, possibly including also fishery data. The continuity of long-term monitoring programs will also be essential for providing us with more insight into the mechanisms shaping ecosystems in a global change perspective.

    Acknowledgements

    A.C. is very thankful to Prof. Martin Attrill and Dr. Tony Walne for essential logistic support. We are particularly thankful to the agencies and people who provided us with data and valuable information: Eastern AtlanticSAHFOS CPR data: David Johns, of the Sir Alister Hardy Foundation for Ocean Science (SAHFOS), http://www.sahfos.ac.uk. Western AtlanticNOAA ECOMON data: Dr Jonathan Hare, Oceanography Branch Chief, NOAA Fisheries Service. This study utilized data collected by the NOAA's Northeast Fisheries Science Center as part of an ongoing mission to monitor and assess the Northeast Continental Shelf ecosystem. The data are available from The Biological and Chemical Oceanography Data Management Office (BCODMO) site, http://osprey.bcodmo.org/dataset.cfm?flag=view&id=13684. Western PacificODATE: The Odate collection of the Tohoku National Fisheries Research Institute, Fisheries Research Agency, and the fund ‘Global Environmental Research Fund of the Japanese Ministry of the Environment’. Central Baltic Sea: We would like to thank all personnel from the Department of Fish Resources Research, Latvian Institute of Food Safety, Animal Health and Environment in Riga, involved in the set-up of the databases.

    This work has relied on real data obtained through decades of effort around the world. We want to express our gratitude to all funding agencies who have through time understood the importance of funding long-term environmental collection, to the visionary people who began such programmes, and to the people who have subsequently maintained such time-series collections. Long-term collections are key tools for investigative research questions, they are irreplaceable, and as such they are invaluable.

    Funding statement

    (A.C.) This research has received funding from the European Union Seventh Framework Programme (FP7/2007–2013) under grant agreement no PIEF-GA-2010–275790, Marie Curie project SYNRESH.

    Footnotes

    †These two authors have contributed equally to the study.

    One contribution of 16 to a Theme Issue ‘Marine regime shifts around the globe: theory, drivers and impacts’.

    References