Is the Antarctic-driven abyssal ocean overturning doomed in 2050?

From Climate Etc.

by Frank Bosse

Probably not, in spite of the recent headlines.

A recent article in Nature Abyssal ocean overturning slowdown and warming driven by Antarctic meltwater by England et al. (hereafter E23) caused quite a stir in the media.  The BBC wrote:Antarctic Ocean currents heading for collapse – report.

E23 built a model to describe the formation and behavior of abyssal water masses around Antarctica.  The Antarctic abyssal waters are important due to its impact on the overturning circulation (AOC) – the lower cell of the Meridional Overturning Curculation (MOC) – which overturns heat, freshwater, oxygen, carbon and nutrients in the abyssal ocean.  The AOC directly influences warming and the availability of nutrients to support marine life near the surface of the ocean.

Here is a schematic of the global MOC:

Fig.1.: The global MOC, a reproduction of Fig.1 of Marshall / Speer (2012).  The Antarctic Bottom Water (AABW) is shown on the left side in descending blue arrows.

E23 concludes:

“In particular, a net slowdown of the abyssal ocean overturning circulation of just over 40% is projected to occur by 205”

According to E23, this would also have some impact on the Atlantic Meridional Overturning Circulation (AMOC), which is responsible for the vast majority of the northward heat transport on earth:

“As the meltwater release from Greenland and Antarctica increases over time, the AABW overturning and AMOC strength both weaken by 2050.” (AMOC by 19% shows Fig. 2).

The cause is the additional meltwater from the Antarctic ice shelves, which has a widespread impact on the Antarctic Bottom Water (AABW):

“First, the projected addition of Antarctic meltwater causes an anomalous freshening . . . which produces fresher and less dense AABW, and eventually reduced AABW volume, after the 2030s.”

The key figure of E23:

Fig.2: A reproduction of Fig. 3a, b in E23. The Antarctic melting will lead to a reduction under the influence of the anthropogenic forcing (aka “Climate Crisis”) of the AABW of 42% (a) in 2050, shown in red. In black: without this forcing.

In Fig. 2 (b) the AMOC shows a robust downward trend over 2004-2020; this is not the case in the observations of “Rapid” at 26.5N;  there is much internal variability, with a dip in 2010 and thereafter a slightly recovery.

Figure:  Observations of the AMOC 2004 to 2020 of “Rapid” at 26,5°N:  Source.

Let’s now have a look how the authors calculated the melting up to 2050, which is a crucial input of the described model for the AABW. From the Methods section of E23:

“…and the multi-model ensemble mean of CMIP6 models under a high- anthropogenic-emissions scenario, Shared Socioeconomic Pathway 5-8.5 (SSP5-8.5), for the future climate component from 2020 until 2050.”

In a twitter thread the lead author stated (and provided a “SharedIt” link to read the full paper, thanks for this):

“…our projections were run under a ‘business as usual’ scenario. Deep and urgent emissions reductions will give us a chance of avoiding an ocean overturning collapse.”

Is SSP5-8.5 (or RCP 8.5 in IPCC AR5) “Business as usual”? Not so, stated this comment, also in “Nature”:

“Stop using the worst-case scenario for climate warming as the most likely outcome”.

Its projections of future greenhouse gas emissions are generally acknowledged to be unrealistic even on pessimistic assumptions.

Furthermore: is the Multi -Model Ensemble mean (MME) of the CMIP6-models appropriate for this approach? No, the MME mean is skewed high owing to a high climate sensitivity due to some models running much too hot. Gavin Schmidt:

“The default behavior in the community has to move away from considering the raw model ensemble mean as meaningful.”

This leads to an urgent need of a discussion of the choice of SPS5-8.5 and the CMIP6 ensemble mean in E23. Unfortunately the paper doesn’t do this, so I will do it in this blog post.

What effect does the choice of the projected temperatures in the Antarctic for 2020 to 2050 have, which in turn influences the expected melting?

With the help of the KNMI Climate explorer I investigated the expected trends, first for the settings used in E23:

Fig.3: The linear temperature trends in and around Antarctica for SPS5-8.5 and the MMM of the CMIP6’s, as it was estimated in E23.

In comparison, the not-so-skewed CMIP5’s MME mean for the more likely RCP4.5 scenario:

Fig.4: The linear temperature trends for RCP4.5 in and around Antarctica.

Note that the trend slopes in the crucial melting areas of the western Antarctic (including the Ross Sea and the Weddell Sea) are nearly 50% steeper in the Fig. 3 than in Fig. 4. This results in a warming in this area from 2020 to 2050 of 0.6 K (Fig.4) based on RCP 4.5 and the CMIP5 MME mean; in Fig. 3 it results in 1.3 K based on SSP5-8.5 and the CMIP6 MME mean.

However, these are climate model simulation results. Let’s compare the spatially resolved linear trends of grid cells for the area 60°S to 90°S in the time span 1990 to 2021, virtually the same length as the 30 years long time span 2020 to 2050 in E23 for the observations (GISS) and the “not so hot case” of Fig.4:

Fig.5: The spatial trend slopes of the gridded data for the CMIP5 models with the scenario RCP 4.5 (left) and observations, GISS (right) for the time span 1990-2021 (“Hindcast”)

Not only do the simulations warm far too quickly in Antarctica and its environs over the last 30 years,
but the modeled warming is poorly correlated with observed warming in most grid cells (Fig.5)

In wide and crucial areas for the forming of the AABW especially on the coastlines (with the exception of the Ross Sea on the bottom) the observed and modelled trends are quite different – in the observations the trends are near zero.

The more realistic scenario CMIP5 RCP4.5 shows a twice as fast warming in the Antarctic (60°S-90°S) as the observations, and the CMIP 6-SSP5-8.5 mean scenario shows an almost 3 times faster warming in the “hindcast” period 1990 to 2021 despite the fact that there is relatively little difference in greenhouse gas emissions and changes in other drivers of climate change between the SSP5-8.5 and RCP4.5 scenarios and observations during that period.

In the light uncertainty of the spatial resolved trends in the observations, I use the relation of the trends of the entire Antarctic region, estimating that the warming bias in models will persist to 2050. This would lead to an additional warming of only 0.3K for 2020 to 2050 in the Western Arctic, 23% of the estimation in E23.


Neither the sole warming scenario nor the multi model CMIP6 ensemble mean used by E23 to estimate the melting in Antarctica up to 2050 is appropriate. The resulting MME mean heavily overestimates the likely surface warming and hence the melting, making it “not meaningful” (see Gavin Schmidt’s citation) as input for the AABW-model used in E23.

For the crucial regions, the trend slopes 1990 to 2021 in the observations are only about one third of the simulations used by E23. Moreover, projected future greenhouse gas emissions and levels are unrealistically high in SSP5-8.5 scenario used by E23. This suggests that future surface warming in and around Antarctica is likely to be far lower than E23 assumes, which in turn means ice melting and hence the slowing of the abyssal ocean overturning would be much less than E23 projects.

E23 moreover concluded that the ocean freshening due to melting near parts of the western Antarctic (namely Ross Sea and Weddell Sea) will lead to the described reduction of the AABW within 30 years to 2050. I had a look at the observational “Argo” data, provided by the “Marine Atlas”. Until December 2021 there is no trend in the salinity data, here shown for the average 0-2000 m depth in the Weddell Sea:

Fig. 6: “Argo” observations of the ocean salinity near the Weddell Sea. In the area of the Ross Sea, there is also no trend (not shown) . The figure was generated with the “Marine Atlas”.

The problems with this paper are: reliance on the implausible SSP5-8.5 emissions scenario, use of the CMIP6 multi-model ensemble mean which is running too hot, and failure to critically evaluate the model simulations using recent observations.  Further failures by Nature’s review and editorial process, combined with uncritical and amplified media promotion,  have unnecessarily confused the science and public.

Acknowledgement: I thank Nic Lewis for very helpful comments on earlier draft versions.