Stepping stones towards Antarctica: Switch to southern spawning grounds explains an abrupt range shift in krill

Poleward range shifts are a global‐scale response to warming, but these vary greatly among taxa and are hard to predict for individual species, localized regions or over shorter (years to decadal) timescales. Moving poleward might be easier in the Arctic than in the Southern Ocean, where evidence for range shifts is sparse and contradictory. Here, we compiled a database of larval Antarctic krill, Euphausia superba and, together with an adult database, it showed how their range shift is out of step with the pace of warming. During a 70‐year period of rapid warming (1920s–1990s), distribution centres of both larvae and adults in the SW Atlantic sector remained fixed, despite warming by 0.5–1.0°C and losing sea ice. This was followed by a hiatus in surface warming and ice loss, yet during this period the distributions of krill life stages shifted greatly, by ~1000 km, to the south‐west. Understanding the mechanism of such step changes is essential, since they herald system reorganizations that are hard to predict with current modelling approaches. We propose that the abrupt shift was driven by climatic controls acting on localized recruitment hotspots, superimposed on thermal niche conservatism. During the warming hiatus, the Southern Annular Mode index continued to become increasingly positive and, likely through reduced feeding success for larvae, this led to a precipitous decline in recruitment from the main reproduction hotspot along the southern Scotia Arc. This cut replenishment to the northern portion of the krill stock, as evidenced by declining density and swarm frequency. Concomitantly, a new, southern reproduction area developed after the 1990s, reinforcing the range shift despite the lack of surface warming. New spawning hotspots may provide the stepping stones needed for range shifts into polar regions, so planning of climate‐ready marine protected areas should include these key areas of future habitat.


| INTRODUC TI ON
Temperature is a major driver of marine species distribution at global scales (Beaugrand et al., 2019;Burrows et al., 2011Burrows et al., , 2019Pinsky et al., 2013) and, accordingly, one of the 'universal' responses to climatic warming is a poleward shift in distribution (Chen et al., 2011;Parmesan & Yohe, 2003). This accords with the general principle of thermal niche conservatism, a central concept of species distribution models, which project distributions under future climatic states based on the assumption that the relationship of a species to its environment (including temperature) remains unchanged (Brun et al., 2016).
The urgent need to gauge species-and ecosystem-level responses to a warmer world has led to a rapid increase in such projections.
These are valuable, for example, in providing broad-brush, large-scale approximations of end-of-century distributions (Cheung et al., 2009).
Despite their clear value at large scales, the utility of such projections at smaller scales has been questioned on several counts. They can be limited in projecting the distributions of individual species, particularly over the shorter timescales (years to a few decades) required by resource managers (Brun et al., 2016;Fernandes et al., 2020). Further, the underpinning assumptions for the concept of niche conservatism (Crisp et al., 2009) are also under scrutiny, given the potential for genetic adaptation, phenotypic variability, life stage-specific responses and other compensation mechanisms (Dam, 2013;Pinsky et al., 2020). Indeed, range shifts among individual species are highly variable and can be much faster, slower, unresponsive or even opposite in direction to the general poleward shifts of isotherms in a warming environment (Chen et al., 2011;Chivers et al., 2017;Fuchs et al., 2020;Tarling et al., 2018). Beaugrand and Kirby (2018) have proposed a theoretical framework to understand the reasons behind this apparent lack of congruence, in terms of interacting processes at multiple levels of spatial and biological organization.
Polar ectotherms are temperature sensitive and yet their habitats are warming faster than the global average rate, so we urgently need to understand and project range shifts to gauge their future roles in ecosystem functioning, food provision and biogeochemical cycling.
With some notable exceptions in parts of the Arctic and subarctic (Dalpadado et al., 2020;Edwards et al., 2021;Ershova et al., 2021;Fossheim et al., 2015), polar data sets rarely have sufficient temporal and spatial coverage to understand how rapid polar warming has translated into range shifts (Wassmann, 2011). Poor understanding of the mechanisms behind past change challenges our confidence in the future projections .
Within the Southern Ocean, Antarctic krill (Euphausia superba) play a central role in the food web, and the spatio-temporal dynamics of this single species impact both higher and lower trophic levels Schmidt et al., 2016). For this reason, studies are increasingly projecting krill distribution (Mackey et al., 2012;Piñones & Fedorov, 2016) or growth potential (Hill et al., 2013;Murphy et al., 2017;Veytia et al., 2020) based on the untested assumption that the relationship of krill with its environment remains fixed. Fortunately, this assumption is testable, since extensive abundance data have been collected over the last century. For this study, we built a longterm, large-scale database for krill larvae to complement an existing one on adults (Atkinson et al., 2017). The overall range of these adult stages has previously been shown to have contracted southwards over the last century . By adding the new larval data set, we show here that the range contraction is actually a step change, which is decoupled from the pace of climate warming.
The larval and adult databases allow us to hypothesize a mechanism, based on spawning dynamics, to explain such abrupt shifts.

| Source krill data
The data set on post-larval krill (i.e. all juvenile and adult krill, hereafter termed collectively as 'adults') was obtained from KRILLBASEabundance, which is described in Atkinson et al. (2017). This is a multinational compilation of all available data on adult abundance (no. niche conservatism. During the warming hiatus, the Southern Annular Mode index continued to become increasingly positive and, likely through reduced feeding success for larvae, this led to a precipitous decline in recruitment from the main reproduction hotspot along the southern Scotia Arc. This cut replenishment to the northern portion of the krill stock, as evidenced by declining density and swarm frequency. Concomitantly, a new, southern reproduction area developed after the 1990s, reinforcing the range shift despite the lack of surface warming. New spawning hotspots may provide the stepping stones needed for range shifts into polar regions, so planning of climate-ready marine protected areas should include these key areas of future habitat.

K E Y W O R D S
abrupt community shift, Antarctic krill, ecosystem shift, euphausiid, management, marine protected areas, range shift, recruitment, spawning m −2 ) of E. superba, based on net samples. It spans the period 1926-2016 and, while circumpolar in scope, has most data in the Atlantic sector. For this study, we also generated a large, parallel database for larval E. superba abundance for the Atlantic sector only (KRILLBASElarvae). An earlier version of this composite database was analysed by Perry et al. (2019) but we have since added several thousand extra net hauls. This larval database, like the one on adults, is a compilation of individual survey data transcribed by one of us (AA) from catch notebooks, appendices of publications or sent directly by the data originators. The original data supplied were presented as numbers of larvae per m 3 or numbers per m 2 , based on a volume filtered estimated typically by multiplying the mouth area of the net by the distance towed. In common with the adult database, we have converted all densities to numbers per m 2 . With 9497 records, KRILLBASElarvae builds on an original database of around 1000 records for the Atlantic sector, compiled by Siegel and Watkins (2016). Table S1 lists the main sources of data that we used for KRILLBASE-larvae.
None of the data in the adult or larval abundance databases are based on sampling targeted on krill aggregations. Both databases include a variety of sampling nets, depths and times of year. Because of the ability of adult krill to escape nets, we base our analyses here on adult densities that have been numerically standardized to a single, relatively efficient net sampling method (a night-time RMT8 net fishing from 0 to 200 m on 1 January). The standardization methods used are detailed in Atkinson et al. (2017). The larval data have not been standardized due to their smaller size, and thus reduced ability to evade nets. However, we checked that the main trends observed were followed independently by both component larval stages, namely calyptopes and furcilia, which differ greatly in size and vertical distribution and thus in their catchability by nets. The sampling coverage for each life stage is shown in Table 1.
For most of the analyses, we divided the sampling period into three eras of approximately 20 years each, with the exact era boundaries selected to be consistent with analyses in Atkinson et al. (2019).
Thus, era 1 was 1926-1939 (The 'Discovery era'), era 2 was 1976-1995 and era 3 was 1996-2016. Table 1 shows the station coverage from larval and adult databases for each era, after the data screening described in the next section.

| Krill data screening
From the complete larval and adult krill databases, we first performed screening procedures to exclude outlying hauls which provided poor or unbalanced representations of krill density or distribution (Table 1). For the adults, these were based on previous screenings according to sampling depth of nets and time of year of sampling so as to remain consistent with the methods used in Atkinson et al. (2008Atkinson et al. ( , 2019. The same screening was applied to larvae, except the additional months of October and November were removed due to the rarity of larvae at that time of year. The full larval database includes density (no. m −2 ) data on eggs, nauplii plus metanauplii, and each of the three calyptope and six furcilia stages. To maximize sample size, we pool all calyptope and furcilia into one single category entitled 'larvae'. The exceptions are Figures S1 and S4, which compare trends in calyptopes and furcilia separately to investigate any biases related to net mouth area and mesh selection.

| Environmental data
Most records of the two krill databases did not have parallel environmental data, so we needed to add them separately. GEBCO bathymetry data were added to each data record as described in Atkinson et al. (2017). To provide an estimate of sea surface temperature at the time of sampling, we used the Extended Reconstructed Sea Surface Temperature (ERSST) data set (Huang et al., 2018; https://www.ncdc.noaa.gov/data-acces s/marin eocea n-data/exten ded-recon struc ted-sea-surfa ce-tempe ratur e-ersst -v5). The data set is derived from a reanalysis based on the most recently available International Comprehensive Ocean-Atmosphere Data Set (ICOADS). Improved statistical methods have been applied to produce a stable monthly reconstruction, on a 2° × 2° spatial grid, based on sparse data (Smith et al., 2008). An annual average was calculated for each geographical cell (between 52 and 74°S and 80 and 20°W) and each year . These values were extracted in Matlab and then each larval or adult station was interpolated onto this grid to extract an annual mean sea surface TA B L E 1 Screening of the full larval and adult krill KRILLBASE databases for our analysis. Further spatial screening applied to some of the subsequent analyses. Sampling eras are described in Section 2.1, with numbers of sampling stations after screening presented for each era temperature relevant to its year of sampling. We chose to extract annual averages for this analysis, rather than those months specific to sampling to maximize the ERSST data input for this highlatitude region. To further account for incomplete spatio-temporal coverage of temperature data, the 2° × 2° grid cells were analysed here based on their average annual values over the whole ~20- year period encompassing each era.
Sea ice is important for the krill life cycle and its winter coverage provides an index of long-term environmental change independent from that of the ERSST data set. The most reliable sea ice index in this sector that spans a century is the South Orkney Fast Ice Series

| Multiple approaches to range shift analysis
Since our databases do not provide an evenly weighted spatial and temporal coverage, we approached the range shift analysis of larval and adult krill in six different ways, which contrast greatly in the way they combine and analyse the data. These were: first, simple gridding and visualization of the data, based both on three sampling eras To examine the 'centre of gravity' range shift in terms of latitude, we first averaged, for each era, the available means for the respective grid cells within each of the nine latitudinal bands. This had the effect of stratifying the data coverage (such that each of our comparable grid cells was given the same weight). This method was chosen for all our range shift analysis here, to account for uneven sampling distribution, with some cells sampled much better than others. We then calculated the centre of gravity as the sum of the product of band mean density and mean latitude of the band, divided by the sum of the densities in each band. Similar calculations of the centre of gravity were performed for the longitudinal bands.
To calculate range shifts in terms of cumulative percentiles ( Figure 3), we took the mean densities in each latitudinal band, as  The trends in sea ice concentration over the last century at the South Orkneys showed a broadly similar pattern to those of sea surface temperature (Figure 1). There was a reduction in winter F I G U R E 4 Larval and adult krill have not maintained a fixed thermal niche during periods of warming and cooling. Change in annual mean sea surface temperature inhabited by larval and adult krill over the three eras. As for Figure 3, we only included grid cells sampled in each era. Grid cells were ordered in increasing temperature and cumulative percentiles and centre of gravity were calculated as per Figure 3, weighting according to krill density in each cell. Because of sparse data coverage at the thermal extremes, coupled to the coarse resolution of Extended Reconstructed Sea Surface Temperature, we were not able to provide a reliable picture of outer thermal limits (i.e. 10% and 90% thermal quantiles)  To obtain more detail on the timing of range shifts, we divided the time period into five eras (each of around a decade) rather than the three periods as used above. This analysis ( Figure S4) showed fairly unchanging latitudinal distributions of larvae and adults until the mid-1990s, and it was only in the last two of the decades that the centre of gravity of their distributions moved substantially to the south.

| Range shifts of larval and adult krill
This finer time resolution clearly shows the out-of-step relationship between the main environmental warming and ice loss (before the 1980s) and the krill range shift (after the 1980s). Importantly, this shift of the centre of gravity of the population did not simply reflect a decline in krill densities in the far north, since densities of larvae increased substantially in the southernmost latitudes ( Figure S4a).

| Shifts in the thermal range occupied by krill
A common method of assessing range shifts is to compare them to the pace of isotherm movement over the corresponding period (Chivers et al., 2017). However, since isotherms run broadly SW-NE in this region, parallel to the range shifts themselves (rather than running across them) and also broadening in extent from west to east (Hofmann & Murphy, 2004), this metric was not used in the present study. Instead, each station was allocated a sea surface temperature value based on a gridded ERSST v5 product (see Sections 2.3 and 2.4.3), allowing an analysis of how the thermal habitat of krill has changed over the study period (Figure 4).
During the rapid warming of the whole sector between the first and second eras, the spatial distributions of both larvae and adults were fairly static (Figure 3), which means that they were inhabiting increasingly warm water. Figure 4 shows that this increase is typically 0.5-1.0°C; similar to the temperature increase in specific locations ( Figure 1). The changes in thermal regime during the cooling period between eras 2 and 3 are more varied, but frequently show the larvae inhabiting increasingly cool waters. This variation reflects a combination of a cooling of waters over this time and variable degrees of range shift (into cooler regions) between leading and trailing edges. Therefore, despite some radical changes in thermal regime and krill distribution over the last century, the net result is that the centre of larval distribution nowadays is in water about 0.5°C warmer than it was 90 years ago, while that of adults is in water of broadly similar temperature. Mann-Kendall test on equalized sample sizes shows consistently more negative z-scores in the northern part of the range than the south, indicating steeper decline trends. These declines were seen for both swarm frequency and for mean density, supporting previous findings that changes in mean krill density may actually be related to changing numbers of swarms in any given area (Brierley & Cox, 2015).

| Causes of the abrupt distributional shift during the warming hiatus
Taken together, the evidence suggests that in this northern portion of the SW Atlantic, there has been a sharp downward trend in density of all three major krill life stages, namely calyptopes, furcilia and adults, since the 1970s. These trends are reflected also in declining frequency of swarms or dense aggregations and likely result from declining recruitment from the major spawning and nursery areas adjacent to the Southern Scotia Arc. This in turn reflects the SAM index becoming increasingly positive during this period, diminishing the importance of this northern spawning and nursery area, concomitantly with the increase in importance of a southern spawning area.

| DISCUSS ION
Our six main approaches to analysing the distributions of calyptope, furcilia and adult stages of krill all show that, during the last 90 years, their distributions have shifted substantially (~1000 km) within the SW Atlantic sector. While a range contraction of adults has been reported previously in this sector , our new larval database now shows how this shift occurred. Understanding the mechanism is important for several reasons. First, the sheer speed of the recent shift is alarming. Assuming that it started from the 1990s ( Figure S4c), it is a jump of ~500 km per decade-far faster than most observed and projected values . Second, and despite much speculation on Southern Ocean range shifts, their existence is not a general phenomenon and so demands a mechanistic explanation. A suite of copepod species, broadly similar in size and swimming ability to krill larvae, showed no evidence for a range shift last century; they maintained their distributions within the SW Atlantic while their environment warmed (Tarling et al., 2018). By contrast, the macroplanktonic salps expanded their leading range edge southwards in a similar way to krill (Atkinson et al., 2004;Pakhomov et al., 2002).
These contrasting examples for and against plankton range shifts in the Southern Ocean differ from the increasing evidence for northward range shifts in the Arctic (Campana et al., 2020;Ershova et al., 2021;Fossheim et al., 2015;Møller & Nielsen, 2019). Is a range shift into the high-latitude Southern Ocean impeded by the particular challenges of bathymetry and hydrography? While the Arctic Ocean is a deep basin supplied by localized northward inflows from the Atlantic and Pacific Oceans (Wassmann, 2011), the other pole comprises a central landmass encircled by a continuous, powerful

Antarctic Circumpolar Current (ACC). Establishment in a warming
Arctic Ocean relies on the reproduction of already advected populations, while occupation of the high-latitude Southern Ocean requires not only successful reproduction but also an additional transport mechanism across or against the currents. Indeed, the SW range shift of krill is actually in the reverse direction to the prevailing ACC flow and its northward surface Ekman component (Hofmann et al., 1998;Hofmann & Murphy, 2004).
For this reason, we need to examine whether we are really observing a range shift related to warming, or instead population expansions and contractions that seem to have the effect of moving the range. For instance, the 'basin model', as applied to small pelagic fish, suggests that in periods of strong recruitment and high total abundance, the range expands, while density within the range centre increases (Barange et al., 2009). This type of basin model may indeed apply to krill; for instance, adults extended almost to the Polar Front after exceptional recruitment years 1981 and 1996, and the rapid retreat of the northern range edge of adults since the 1980s coincided with a major decline in their density and swarm frequency ( Figures S4 and S6). However, the southern (leading) edges of both larvae and adult distributions have also moved south over the last century, and larvae have increased greatly in the far south ( Figure S3). This strongly suggests that we are observing a temperature-induced range shift, in addition to a range contraction as abundance has declined. In the future, this southern area may become the main successful spawning area of krill in this sector since a series of habitat models, both of larvae (Thorpe et al., 2019) and adults (Hill et al., 2013;Piñones & Fedorov, 2016;Veytia et al., 2020), project increasingly favourable conditions for krill towards the south.
Our ability to observe both multiple life stages of krill and their recruitment over time and space sheds light on the mechanisms behind range shifts. The movements do not comply in a simple manner with thermal niche theory: between the 1920s and the 1990s, the central and northern edges of adults and larvae remain in place while their environment warmed. This parallels the observations of Scotia Sea copepods over a similar time span (Tarling et al., 2018), but runs counter to many projection models of species distributions, such as Mackey et al. (2012), which assume that species track thermally optimum isotherms poleward. With a non-continuous time series between eras 1 and 2, we can only speculate why this is the case. One explanation relates to the depth distribution of krill. In common with most other long timeseries analyses and projections, we were forced by data availability to use surface values for our temperature index. However, krill distribution extends to depth (Marr, 1962;Schmidt et al., 2011;, in waters insulated from the rapid changes at the surface (Meredith & King, 2005;Whitehouse et al., 2008) and this may provide a more geographically stable thermal environment for krill.
This similarity in distributions between eras 1 and 2 amid a warming surface environment contrasts with the major (~1000 km) jump to the SW during the 25-year warming hiatus from the 1990s. Could this rapid shift also be interpreted in terms of differential warming of surface and deep waters? The hiatus in surface warming (Turner et al., 2016) and more stable sea ice conditions (Henley et al., 2019) appear to be marked by a general cooling of the topmost ~50 m layer only; this being suggested to be related to strengthening winds and associated sea ice transport (Haumann et al., 2020). Below this near-surface layer, at the 55-65°S latitudes important for krill, there has been a continuation of the longer term warming since the 1980s, albeit with little change below 200 m (Haumann et al., 2020). There is certainly no evidence for a major acceleration in subsurface warming in recent decades that would account for such an abrupt shift in krill distribution.
Neither would such an explanation fit with the fact that during the previous decades of warming of surface and subsurface waters (Gille, 2002;Whitehouse et al., 2008), the krill distribution was fairly static.
To explain this abrupt, non-linear range shift of krill, we instead suggest a putative model ( Figure 6) in which two separate mechanisms interact. First is the underlying change in thermal regime, and this warming allows the extension of the adult range down the WAP last century. The second effect is indirect: from climatically driven changes in the food environment that affect krill recruitment.
Winter sea ice was initially shown to be an important predictor of recruitment (Atkinson et al., 2004;Loeb et al., 1997;Siegel & Loeb, 1995), but work since has shown that recruitment is better predicted statistically by climatic indices which relate not only to winter sea ice, but which also modulate food supply for larvae in the ice-free season (Loeb & Santora, 2015;Saba et al., 2014). Thus, at the wholesector scale of our study, the SAM has a dominant role in driving recruitment . Here (Figure 5a,b) we pinpoint its effect as being specifically in the main reproductive and nursery grounds which are along the Southern Scotia Arc (Perry et al., 2019).
How does SAM affect krill recruitment? The increasingly positive SAM is thought to relate to warmer, windier and more unsettled weather at the northern Antarctic Peninsula; unfavourable conditions for larval fitness (Loeb et al., 1997;Quetin et al., 2007;Saba et al., 2014). We suggest that this combination of conditions has reduced krill replenishment to the main northern (Scotia Sea) part of the range ( Figure 6). This is evidenced by the sharp declines in larvae and adult density (Figure 5c,d), swarm frequency ( Figure S4) and recruitment indices .
SAM also has an effect on recruitment in the WAP populations, but here its influence is not nearly so strong, with El Niño Southern Oscillation (ENSO) also having a major effect (Saba et al., 2014). Consequently, we suggest that reproductive success has been more stable in the southern part of the range. There is a subdecadal periodicity in adult density but little directional change evident since the early 1990s (Steinberg et al., 2015; see also Figure 5e,f; Figure S6) and increasing larval densities at the leading range edge ( Figure S4a). Together, this has contributed to a shift of the overall range to the south ( Figure 6).
Observations of phytoplankton distributions along the WAP which has experienced the rapid decline in krill densities. The mechanisms behind these changes are likely to be complex , include changes in food quality as well as quantity (Schofield et al., 2017) and act on separate spawning locations along the Antarctic Peninsula (Conroy et al., 2020). This situation is similar to that in the rapidly warming Arctic, where poleward range shifts have been interpreted as a combination of direct temperature effects and superimposed effects from food web interactions (Dalpadado et al., 2020;Ershova et al., 2021).
Whatever the mechanisms, our results emphasize the decoupling between the range shifts of krill and the pace of climatic warming.
At the largest scale, krill have broadly maintained their thermal niche because, in the last few decades, their adults are centred in a similar thermal regime as they were 90 years ago, with this regime now found at higher latitudes. However, they have not smoothly tracked these thermal changes, withstanding about 1°C of warming in situ before undergoing an enormous shift during the warming hiatus now to occupy a new spawning ground. This has major implications for future projections, since processes are strongly non-linear, with the possibility of further abrupt shifts unrelated to temperature but more strongly related to climatic controls in localized recruitment hotspots.
Differential range shifts among species, particularly if they involve key species, can reconfigure food web interactions (Wallingford et al., 2020). Krill are important biogeochemically (Schmidt et al., 2016), supporting an iconic food web of penguins, seals, fish, flying seabirds and whales (Constable et al., 2014) and an expanding fishery (Meyer et al., 2020). Some of these dependent predators, for example gentoo penguins, have themselves expanded their ranges southwards (Korczak-Abshire et al., 2021). Given the pace of previous warming, there is widespread concern over future trends, particularly over the possibility of step changes or other nonlinear responses to climatic change. Range shifts of krill can lead to substantial increases or decreases in availability to krill-dependent predators, already suggested through reductions in the contribution of krill to the diet of a krill specialist, the gentoo penguin (McMahon et al., 2019) and reduced fur seal pup birth weights (Forcada & Hoffman, 2014). The great speed of the recent larval and adult poleward shift, at a time when parts of the region were actually cooling, is a warning of more surprises in store.
Designing 'climate-ready' areas for protection that help key, vulnerable or exploited species such as krill needs to account for future climate moving their distribution (Queirós et al., 2016;Visalli et al., 2020). To project the future, species distribution models are particularly valuable at larger scales, including over 50-year time frames and global scales. At these scales, for example, they help gauge vulnerable biomes and potential hotspots of change (e.g. Jones & Cheung, 2015). Likewise, mapping of climate change velocities may help conservation planning in many areas of the world ocean where biological data are sparse (Brito-Morales et al., 2018;Burrows et al., 2011). All of these approaches carry limitations, especially for certain taxa (Brun et al., 2016), for short projection periods or for limited geographical areas (Cheung et al., 2016;Fernandes et al., 2020). In this context, a ~1000 km jump in krill distribution, in two decades and under stable surface temperature, provides a sobering reminder about projection at managementrelevant timescales. Krill provide a case study of the mechanisms by which such surprises can occur, for which a dogged maintenance of time series has proved essential. Understanding the drivers of step changes is critical (Beaugrand, 2012;Beaugrand et al., 2019;Conversi et al., 2015), since they can herald system reorganizations that impact on food web structure, biogeochemical function and fisheries management. Equally important is the protection of habitats, such as the new southern krill reproduction area, that sustain populations as distributions change.

ACK N OWLED G EM ENTS
This study was based on a large number of samples collected over the last 90 years and the authors are indebted to the various crews and scientists not only for obtaining these data but also for presenting it in a form suitable for reuse years later. They thank in particular Volker Siegel and Peter Ward who provided a large amount of data to get the larval database started as well as Project SYM-PEL (NE/S002502/1).

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Most data supporting this paper are published in Atkinson et al. (2017;ESSD) and the remaining data will be available from the BAS Polar Data Centre on publication.

F I G U R E 6
Hypothesized mechanism for the stepwise range shift. The schematic depicts the warm (red) to cool (blue) temperature gradient across the SW Atlantic sector in the warming era and the subsequent warming hiatus, with main population centres of larvae (red) and adults (blue), based on Figure 2. The hypothesized mechanism for this shift is based on the adults moving their southern range edge south during the warming period, supporting spawning off Marguerite Bay (WAP, Western Antarctic Peninsula), which then fuels increasing adult stocks in this southern area. Meanwhile deteriorating conditions for recruitment in the former Scotia Sea spawning stronghold, due to an increasingly positive Southern Annular Mode ( Figure 5), cuts the recruitment of adults to the northern part of the range. We suggest that this differential recruitment process drives the main range shift which is seen, surprisingly, during the surface warming hiatus O RCI D