25-year Climatological state and long-term change
Based on the 148 repeats of the same section, we construct a summer temperature climatological mean over the 25 years, which shows the main Southern Ocean water-masses and the fingerprints of the main fronts associated with the Antarctic Circumpolar Current (Fig. 1b; see Methods). The warmest water-masses on the section, the Subtropical Water (STW) and SubAntarctic Mode Water (SAMW) are located in the northern part of the transects. Their southern extent is limited by the Subtropical Front (11°C at 150m16) and the
Subantarctic Front (strongest temperature gradient between 3-8°C at 300 m depth17, respectively. SAMW is found down to 600 m depth, beneath the summer mixed layer, consistent with previous studies18,19. Antarctic Surface Waters (AASW) are located in the upper 250 meters of the Southern Ocean and south of the Polar Front (most northern extent of the subsurface 2°C water20). AASWs are composed of a remnant subsurface tongue of cold water produced in winter21,22 (Winter Water), and warmer surface waters produced in summer23,24. Below the Winter Water tongue lies the Upper Circumpolar Deep Water (UCDW), slightly warmer and saltier water than Winter Water, that are advected at depth around the Southern Ocean, partly originating from the North Atlantic Deep Water25.
We are interested in how this temperature structure is changing over time on a multi-decadal timescale. Over the past decades, the temperature has been warming overall across the section, but with a structure showing marked patterns, which are related to the different water- masses of the region. The largest warming reaching 0.4 to 0.8 °C per decade is observed on the northern end of the section, north and within the ACC (region A in Fig. 2b) in the subtropical waters and subantarctic Mode Waters. In contrast, on the southern end of the section, a significant cooling of 0.1 to 0.3°C per decade, extending from the surface to about 200 m is observed in the coolest water-mass of the region (region B in Fig. 2b). Hints of cooling trends are apparent in the surface layer further north, but the trends are not significant (lower than their standard error) north of ~61°S in the surface layer. At deeper depth, in the Upper Circumpolar Deep Water layer (region C in Fig. 2b), subtle but significant (greater than their standard error) warming trends of around 0.05°C per decade are observed from - 62.5°S to -52°S.
The temperature change structure shown across the section concurs well with past studies that have investigated long-term temperature trends in the Southern Ocean (ref 6, and references therein). Here, we however bring an important step forward in our understanding of past changes by showing for the first time that Southern Ocean water-mass temperature trends is robust over a 25-year period, relevant in a climate change context (i.e. beyond decadal timescales). But more importantly, we are able to estimate the typical interannual variability (referred to as noise) to better interpret the observed trends over a 25-year period (referred to as signal; see Methods). In other words, for the first time from observations in the Southern Ocean, we are able to estimate whether the signal of temperature change has emerged above the interannual variability noise. A latitude-vertical section of this trend signal-to-noise ratio is shown in Fig 2c. The three regions highlighted above with significant trends clearly stand out, experiencing temperature changes that emerge above the background interannual variability over the past 25 years. Counter-intuitively, it is in the Upper Circumpolar Deep Water layer, where the long-term change amplitude is the lowest of the section, that the signal-to-noise ratio is the largest because interannual variability is actually very weak. This clearly pinpoints that, while subtle, the observed temperature increase in the Upper Circumpolar Deep Water represents a radical deviation from its mean state. In other water- masses with a more recent surface connection, the 25-year trends are weaker compared to the typical interannual variability. A signal-to-noise ratio lower than one does not mean trends are insignificant, rather it remains unclear whether the measured long-term change reflects a robust change departing from its typical interannual variability. A robust long-term trend might be hidden behind a low signal-to-noise ratio, but one would have to accumulate more years of repeat observations to observe its emergence above the interannual noise.
Water-mass temperature time-series and forcing
We next compute time-series and associated trends, averaged over the three regions identified above where trends overcome both their standard error, and the typical interannual variability: in the subantarctic and subtropical region north of 52.5°S (region A); in the near-surface subpolar region, in the upper 200 m, south of 61°S (region B); and in the subsurface Upper Circumpolar Deep Water, deeper than 250 m, and between 62.5°S-55°S (region C).
When averaged over the entire Subantarctic and Subtropical Mode Water region (region A), the temperature has increased overall by 0.29±0.09 °C per decade, with a 25-year signal to noise ratio of 2.45, indicating a signal much greater than the estimated interannual noise (Fig 3a). Locally the trend can be as high as 0.8°C per decade (Fig. 2b), with the strongest warming organized in deep-reaching localized cells. Based on a shorter 13-yr time-series, ref 26 proposed that this warming was due to the southward movement of both the STF and the SAF, reflecting the consensus when the study was published that ACC fronts were shifting southward. After a decade of scientific debate, a new consensus emerges that on a circumpolar average, the SAF has been shown to be stable and not moving meridionally in the last decades2,27 and that the warming might instead be due to increased heat uptake from the ocean surface19,28. While the warming trend is relatively constant over the 25-year period, there are periods of distinct cooling, for example in 1996 and 2005, and stronger warming in 2001-2002 and in 2014-2016. Similar interannual variability is also evident in the sea-surface temperature fields, with a correlation of 0.64, and a slightly lower 25-yr trend of 0.15+- 0.09°C per decade, consistent with the trend distribution within the zone (Figure 2b). Part of the observed interannual variability might be due to intermittent incursions of subtropical waters carried by the Tasman Sea extension south of Tasmania, impacting the volume of STW, as well as local eddy activity around the SAF29-31 (See Supplementary Information S6).
The overall cooling in the surface subpolar waters close to Antarctica, from the surface to 200 m and from 66°S to 61°S (region B), is -0.07±0.04°C per decade, with a signal-to-noise ratio of 0.97 (Fig. 3b). The cooling appears mostly associated with the coolest waters in the regions (Fig 2b); when isolating only data points cooler than 0°C, the cooling is slightly more marked (-0.08±0.05°C per decade, signal-to-noise ratio of 1.08; Fig. 4a). This cooling of subpolar waters is also accompanied by a freshening of the surface waters over the same period, as well as an increase in sea-ice cover5. Region B has a lower signal-to-noise, and the interannual variability in temperature, SSS and sea-ice is impacted by local coastal circulation changes and increased ice flow from 2011 onwards, following the Mertz Glacier calving just upstream32-34. Such high-latitude cooling is also consistent with local sea surface cooling observed from satellite observations (Figure 2c, correlation r=0.80), and more generally with the surface cooling of a large part of the Southern Ocean that have been observed from observations in the subpolar waters over the past three decades 10,13,35.This cooling might be explained by the increased stratification associated with freshening of the surface layer which would tend to reduce mixing with the slightly warmer underlying Lower and Upper Circumpolar Deep Water4,10,36-38. Locally, ref 5 found a link between the freshening of the subpolar waters near 140°E, the sea-ice cover and a large-scale northward shift of the zero- zonal wind position from 1999 onwards.
Interestingly the winter water tongue extending further north does not show a similar cooling. Small pockets of cooling exist but the WW trend signals are dominated by interannual variability (0.28 signal to noise ratio). When focusing only on the temperature of the core of the Winter Water layer, defined as the layer with temperature colder than 2°C between 55°S and 61.5°S, the large interannual variations overwhelm any long-term change, with peak-to peak temperature ranging from 0.40 to 0.65°C (Fig. 4b). These temperature variations within the Winter Water core are positively correlated (r=0.68) with the sea surface temperature of the previous winter further upstream in the subpolar Australian-Antarctic basin (120-145°E; 57-61°S) (Fig. 4b), where the Winter Waters were modified at the surface (See Supplementary Information S8).
The Upper Circumpolar Deep Water from 62.5°S to 55°S, and over 300-800m depth (region C exhibits a small overall warming trend of 0.04 ± 0.01 °C per decade, associated with a high signal to noise ratio of 3.27. Consistently, the time-series show relatively weak interannual variability, but a steady warming of the layer. The maximum temperature increase sits directly below the seasonally variable surface layer, in the upper and warmer part of the water-mass around 300-500 m (Fig 2b). When the temperature time-series is computed in this core of temperature maximum, the warming trend is even greater, reaching 0.05±0.01°C per decade, with a signal to noise ratio of 4.09 (when excluding 2012 which appears as a clear warm outlier, the trend is slightly lower: 0.048°C per decade instead of 0.054°C per decade; we give the rounded value of 0.05±0.01°C per decade). Previous authors have suggested the warming of the Upper Circumpolar Deep Water might be driven by increased stratification at the base of the Winter Water layer due to freshening, which would reduce mixing between the two layers and heat removal from the Upper Circumpolar Deep Water to the atmosphere10,39,40. Since we have only temperature profiles, the role of the salinity stratification cannot be verified directly. However, in accordance with this hypothesis, we observe larger warming in the upper part of the layer, directly underlying a near-surface water mostly affected by interannual variability (Fig 4b) but with a few hints of local cooling (Fig 2b). In addition to the warming of Upper Circumpolar Deep Water, the depth of the core of maximum temperature is observed to shallow at a rate of 39±11 m per decade (Fig 4d, three to ten times higher than previously reported (5-10 m per decade11), and within the error envelope of the rate observed in West Antarctica (50±18 m per decade11). The shallowing of the layer, which has a strong trend signal emerging from the interannual variability noise (signal to noise ratio of 2.24), might be driven by large-scale atmospheric pattern changes, driving stronger upwelling favorable winds. Indeed, there is a small trend in negative wind stress curl (upwelling favorable) although the standard error around the trend is large (Figure S7). The cause of the CDW shallowing remains an open question, as a dedicated study investigating the mechanistic understanding of subsurface temperature depth change is still missing11.