Climatically driven fluctuations in Southern Ocean ecosystems

Determining how climate fluctuations affect ocean ecosystems requires an understanding of how biological and physical processes interact across a wide range of scales. Here we examine the role of physical and biological processes in generating fluctuations in the ecosystem around South Georgia in the South Atlantic sector of the Southern Ocean. Anomalies in sea surface temperature (SST) in the South Pacific sector of the Southern Ocean have previously been shown to be generated through atmospheric teleconnections with El Niño Southern Oscillation (ENSO)-related processes. These SST anomalies are propagated via the Antarctic Circumpolar Current into the South Atlantic (on time scales of more than 1 year), where ENSO and Southern Annular Mode-related atmospheric processes have a direct influence on short (less than six months) time scales. We find that across the South Atlantic sector, these changes in SST, and related fluctuations in winter sea ice extent, affect the recruitment and dispersal of Antarctic krill. This oceanographically driven variation in krill population dynamics and abundance in turn affects the breeding success of seabird and marine mammal predators that depend on krill as food. Such propagating anomalies, mediated through physical and trophic interactions, are likely to be an important component of variation in ocean ecosystems and affect responses to longer term change. Population models derived on the basis of these oceanic fluctuations indicate that plausible rates of regional warming of 1oC over the next 100 years could lead to more than a 95% reduction in the biomass and abundance of krill across the Scotia Sea by the end of the century.


Supplementary Methods and Analyses
Physical data. The timing of the variables used in the analyses are illustrated schematically in supplementary table 1.
Biological data. Demographic data on Antarctic fur seals (termed fur seals) from Bird Island, South Georgia include estimates of the number of pups produced each year during the breeding season and the mass at which they were weaned. Pup production is considered to be the best index of interannual changes in predator performance while weaning mass gives an index of performance during the early part of the season (Forcada et al. 2005). Weekly data on the mean length of krill (in mm) consumed by fur seals at South Georgia were collected since 1991. The mean length of krill in the diet during the last 3 wks in March is used as an index of krill population changes (Reid et al. 1999). Acoustic data providing short-term (~1 wk) estimates of local (within < 100 km distance of Bird Island, 80 x 80 km region) krill biomass were collected irregularly between 1981 and 1994 and yearly between November and January since 1995.

Time series analyses.
To account for the autocorrelation structure for cross-correlation analyses shown here, the monthly SST and sea ice data were analysed with a seasonal decomposition based on a semiparametric regression with loess smoothing (Cleveland et al. 1990;Forcada et al. 2005). The seasonal component was removed from each monthly series and the new sub-series was smoothed to find long-term trends through several iterations. The residuals of the seasonal and trend fits were also removed, leaving a smoothed trend filtered for residual noise. The filtered series showed the succession of anomalies over time and was used in subsequent cross-correlation analyses (supplementary figure 1).
The ice-edge position at 45.5 o W (15% concentration) was used as an index of sea ice cover in this area (supplementary figure   2).

Pacific and South Atlantic variation in SST
Further to the analyses given in the main text (see paper figure 2) the detailed analyses of the interannual changes in SST at South Georgia and their relationship with the south-east Pacific sector and equatorial region SST variation are presented in supplementary table 2 and are plotted in supplementary figure 2. Smoothing the data to remove monthly variation indicated that 49% of the variation in SST anomalies at South Georgia (SG) was explained by a model including both the SST anomaly data from the Amundsen-Bellingshausen Sea (AB) region 12 months previously and the Nino-3 series 6 months previously (Based on smoothed 6-month moving average series; SG 0 = 0.020 + 0.382 AB -12 -0.160 Nino-3 -6 ; p < 0.0001). Scotia Sea ice extent shows marked inter-and intra-annual variation, which is closely correlated with changes in SST (supplementary figure 3). As noted in the main text the Scotia Sea sea ice variation leads the South Georgia SST fluctuations by < 6 months. Variables are SST anomaly for the Bellingshausen Sea (BS) region for 10 months earlier; Nino-4 series 4 months earlier, Nino-3 series 4 months earlier and the SAM monthly series with no lag. ΔAIC c values are given for comparable models and scaled from the lowest value of AIC c . n adj = adjusted sample number allowing for series autocorrelation; Spearman Rank Correlation (R S ). All regressions are significant (< 1% probability level) as are correlation coefficients (including autocorrelation effects).

Population dynamics of krill in the Scotia Sea
Regression modelling. Linear regression models were estimated for all possible combinations of variables and the 5 models with the lowest Akaike Information Criteria (AIC ) c were reported in the paper observations was also adjusted for autocorrelation using the equation: n e = n/(1 + 2 r a1 r b1 + 2 r a2 r b2 ) , where r a1 = series a correlation lag 1, r a2 = series a correlation lag 2, r b1 = series b correlation lag 1, r b2 = series b correlation lag 2 (Ottersen & Loeng 2000;Ottersen & Stenseth 2001). For the relationship with SST in the previous season, the relative abundance (RI t ) of the juvenile cohort that has survived the first winter (here termed year class 1+) was estimated as a function of the SST anomaly (T t ) at the end of December of the previous season and is expressed as: The second function relates recruitment of the larval cohort (0+ age group) to the SST anomaly (T t ) at the end of December 2 years before the observed South Georgia recruitment and is expressed as: Where β is a constant and T d is the SST anomaly that results in a relative abundance index of 0.5. Changes in the population were then calculated in both cases based on all year classes 2 yrs or older. The functions then produce either a 1 or 2 -year lag between the environmental effect of SST on the size of the recruiting age class and its effect on the population as the 2 yr age class.
Changes in the mean size of each krill year class were modelled as a seasonal von Bertalanffy function. A Normal distribution (mean = 0, sd = 2.5) was used to derive a cohort size distribution (2 mm length resolution) for each year (Murphy & Reid 2001).
The complete length frequency was then derived as the sum across all year classes. The mean length was derived from the combined length distribution (Reid et al. 1999).
Multiple simulations were produced in a Monte Carlo analysis to estimate the best fit

Climate interactions in ocean ecosystems.
Climate impacts in ocean ecosystems. Further to the discussion given in the paper, here we summarise the links between climate related fluctuations in the Pacific and the ecosystem impacts in the South Atlantic. Large-scale climate-related processes in the equatorial Pacific generate anomalies in SST across the South Pacific sector of the Southern Ocean that propagate in the Antarctic Circumpolar Current. This physical variation can be further modified by direct short term (< 6 month) ENSO and SAM related atmospheric effects. This variation impacts krill population processes in the southern Scotia Sea. This generates biological fluctuations that in turn propagate across the Scotia Sea and up to higher trophic levels in the regional ecosystem. The nature of the relationship is such that positive SST anomalies in the ENSO regions occur around 2 -3 years prior to positive anomalies in the southwest Atlantic; this variability then generates a biological response-wave in the regional ecosystem (see supplementary figure 4). When the Scotia Sea is in a cold phase strong recruitment occurs which generates a peak in krill abundance and dispersal of older year classes 1 -2 years later, and a maximum krill biomass in the north after a further 1 -2 years. The physical signal is sufficiently strong in the southeast Pacific 1 year before it enters the Scotia Sea that it will be a useful basis for prediction. We can therefore predict that warm conditions (positive anomalies) in the southeast Pacific to the west of the Peninsula will lead the Scotia Sea region by 1 year, and hence pre-empt periods of very low krill biomass and poor predator breeding performance across the northern Scotia Sea by 2 years.
Forecasting the effects of climate change. To consider the potential effects of regional warming on krill population dynamics we firstly undertook Monte Carlo simulations of future change scenarios. The simulations were based on the multiple regression models shown in the main table 1. For these long-term projections we used the lowest AIC c values models which included a zero-lag population term and a single environmental variable. These relate the rate of population growth for biomass with B t and SST (in the spring) (paper table 1, M3), and for density include terms N t and N t-1 and ice conditions with a 2-year lag (paper table 1, M9).
We assume that the models represent the major factors controlling krill density and biomass in the Scotia Sea.
Applying these models in this way does not take account of the effects of other long-term physical and biological interaction effects, such as changing circulation patterns, food-availability, competition or predation. The models were used to estimate the change in the population growth rate (based on density and biomass) and hence population size over the next 100 years. The SST anomaly was represented as a Normal Where R t is the population growth rate (ln[B t+1 /B t ]) at time t, BB t is the biomass and α, β 1 & β 2 are the model parameters (paper table 1, M3) and ε t is the Normal variate representing the SST variation. Model M9 included a delay term (N t-1 ) and a lag effect of sea-ice. For each set of runs we estimated the probability of the population size (density or biomass) being reduced and remaining at a level less than 5% of the start value during the simulation. The results were not sensitive to the threshold level selected. The probability of decline of the krill population was estimated as the proportion of the 1000 runs in which the decline occurred. For these runs we also estimated the mean time at which the decline below 5% occurred. We note that in this case the reduction refers to local decline of krill across the Scotia Sea/South Atlantic sector of the Southern Ocean and not to the whole krill population. However, as this region currently contains ~50% of the circumpolar population of krill such a change would have profound consequences.
Further to the points made in the main text, a simplistic extrapolation of the declining trend in the abundance series, if it continues, would suggest an absence of krill from the Scotia Sea in about 25 to 30 years. Developing more detailed predictions of biological responses to global climate change are limited by uncertainty in both physical system changes and the biological processes involved. Climate models generally predict warming and sea ice reductions in the Southern Ocean over the next 100 years, but the magnitude of these predicted changes is highly variable between models (Clarke et al. 2007). Current coupled ocean-atmosphere models do not predict well the rapid regional atmospheric warming that has occurred around the west Antarctic Peninsula over the last 50 years.
An increase in frequency of years of low production, through a change in the periodicity and/or an increase in mean SST, would tend to increase the potential for locally significant reduction in krill abundance. The demographic age-structured model generates a very similar effect. The form of the recruitment functional relationship (we consider the 2-year-lag model, see above) indicates that an increase of 1 o C would lead to consistently low recruitment and reduce abundance by > 95% in < 100 years.
Derived relationships of predator performance and SST also suggest that changes of only ~1 o C would have a significant negative impact on predator breeding performance (Forcada et al. 2005;Trathan et al. 2006 ).  propagate from the Pacific sector of the SO into the Atlantic sector of the SO in association with the Antarctic Circumpolar Current (ACC) over a period of 2 to 3 years, with other atmospheric forcing acting on the anomalies as they propagate. Regional changes in SST are linked to fluctuations in winter sea ice in the southern Scotia Sea and around the Antarctic Peninsula. The distribution of sea ice affects the recruitment, survival and dispersal of Antarctic krill. Physically driven variation in krill population dynamics and abundance, in turn affects the breeding success of seabird and marine mammal predators that depend on krill as food. There is no lag between SST variations in the ENSO region and the south west Pacific sector of the SO (yellow arrow). Once generated, anomalies in the southwest Pacific take around 2-3 years to propagate to the southwest Atlantic, with significant further atmospheric interaction occurring as they progress (especially in the southeast Pacific, yellow arrows). Colder SST periods in the Atlantic sector associated with more extensive winter sea ice (light blue arrows) enhance Antarctic krill recruitment in this region (i.e. more 0-age class individuals) and generates higher winter krill-survival rates over the following 1 or 2 winters. More northward extent of sea ice also enhances the dispersal of these krill further across the Scotia Sea towards South Georgia (SG) (red arrows) where they become the dominant size class in the diet of Antarctic fur seals (Arctocephalus gazella) in the following season. Depending on the growth conditions experienced after spawning, these krill may be 2 to 3 years old by the time they reach South Georgia. Numbers in red refer to the age class of krill in the various regions.