2 royalsocietypublishing.org/journal/rsta Phil. Trans. R. Soc. A 381: 20220193
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1. Introduction
In August 2021, Working Group 1—The Physical Science Basis—of the Intergovernmental Panel
for Climate Change published its 6th Assessment Report (IPCC AR6 WG1 [1]). With it came
the reiteration that the oceans were changing due to climate change—it is ‘virtually certain’ that
ocean heat content increased, and there was ‘high confidence’ that sea levels had risen. Climate
models—in this case the climate models of the Coupled Model Intercomparison Project Phase 6
(CMIP6)—showed these trends would continue along trajectories governed by greenhouse gas
emissions, with ‘high confidence’ that already emitted greenhouse gases will continue to warm
the ocean, and it being ‘virtually certain’ that sea levels will continue to rise. However, not all icons
of ocean change had such a certain affirmation—in particular, statements of confidence about
observed changes in the Atlantic Meridional Overturning Circulation (AMOC) were weakened
relative to the preceding IPCC Special Report on the Ocean and Cryosphere in a Changing
Climate [2] in 2019.
The AMOC is a system of ocean currents that moves warm shallow water northwards and
returns cold deep water southwards (figure 1). The Gulf Stream is part of this system (hence
popular use of the ‘Gulf Stream System’), but the AMOC is a much broader system of ocean
circulation. The AMOC receives much attention due to its outsized role in the climate of the
Northern Hemisphere and of Europe, in particular. Atlantic maritime climates in northwestern
Europe are about 5°C warmer on average than Pacific maritime climates at similar latitudes [3].
The port city of Murmansk, north of the Arctic circle, remains ice free year round due to warm
Atlantic waters. The popularization of ‘the Gulf Stream’ as the reason for Europe’s mild climate
dates from the racist [4] Matthew Fontaine Maury’s Physical Geography of the Seas [5] and has
permeated to the point that David Ellett memorably stated that in ‘the west of Scotland, chance
remarks to elderly ladies upon the warmth of a winter’s day invariably bring the reply “Aye, it’s
the Gulf Stream, you know”’ [6].
The idea that the AMOC will weaken due to climate change has a long history also. Since the
mid-1980s, the earliest climate models first predicted that an AMOC slowdown would result from
an increase in atmospheric CO2 [7,8]. This projection has been repeated in successive IPCC reports
in the intervening 40 years. A large AMOC slowdown is projected to bring a cooler, stormier
climate, drier summers and wetter winters, to millions of people, especially in Europe [9,10]. In
the most extreme scenario of a complete AMOC collapse where no heat is delivered by the ocean
to the northern North Atlantic, a cascade of climate impacts could ensue, accelerating ice loss in
West Antarctica, Amazon dieback, and a collapse of arable farming in northwest Europe [11,12].
Whether the AMOC will weaken in the future, or perhaps has already begun to do so, is an
important question. However, this question cannot be answered with certainty, which is reflected
in the different assessments of the state of the AMOC as given in the IPCC reports over the last
decade. While the IPCC AR5 report concluded that there was no observational evidence of a
long-term AMOC decline, in 2019, the IPCC SROCC report stated with medium confidence that
the AMOC had weakened relative to 1850–1900. However, by 2021, the IPCC AR6 report went
back to stating there was low confidence in a twentieth century AMOC decline. How did this
happen? In 2018, two independent studies were published in Nature that concluded that the
AMOC had weakened by about 15% over the course of the twentieth century [13,14]. While the temporal evolution suggested by these proxies (Caesar and Thornalley indices in figure 2) did not fully agree with the latest climate models at that time (CMIP5) that suggested only a weak downward trend [23,24], they at least agreed on the sign of the change. Yet this changed as the latest generation of climate models (CMIP6) was released that showed, in the multi-model ensemble mean, a slight strengthening of the AMOC over the historical period [25].
In this paper, we will look at possible reasons for the model-observation discrepancy of the past AMOC evolution. First, we look at the twentieth century AMOC evolution in observational data and in the CMIP5 and CMIP6 ensembles, with a focus on the multi-model ensemble mean to mirror the approach of the IPCC. We consider how observations and models are/can be compared and discuss three possible reasons for the existing model-observation discrepancy, namely: (i) the models are wrong, (ii) the observations are wrong, (iii) models and observations estimate different quantities and are not expected to agree.

...
Comparing the models’ AMOC evolution to the observational data (figure 3, lower panel), we find that neither the CMIP5 nor the CMIP6 ensemble mean are successful at representing the observational AMOC data.

But, if these models cannot reproduce past variations, why
should we be so confident about their ability to predict the future?