AMOC To Collapse–Part 98

A call went out from the CO2 Coalition regarding this paper last week, which induced me to have a look. The discrepancy between its title and what was actually done in the research is a bad indicator.

Let’s just begin with the title the Guardian chose to use:

“Critical Atlantic current significantly more likely to collapse than thought”

and now the actual research title…

“Observational constraints project a ~50% AMOC weakening by the end of this century”

Now compare motivation for the effort to what was actually done:

Motivation:

Many different climate models are used to make projections of future climate; in particularly the strength of the AMOC. The outcomes of these are highly variable and the best indication of this is from the abstract from the research itself, not the Guardian interpretation.

“Climate models suggest a reduction in AMOC strength of 32 ± 37% by 2100 (90% probability, SharedSocioeconomic Pathways 2-4.5 scenario, Coupled Model Intercomparison Project Phase 6).”

What this says in plain English is that the average projection from a group of 32 models is that the AMOC is expected to weaken 32% from its current value, but the variability of this result is plus or minus 37%. In other words, within the 90% confidence interval of model outcomes the AMOC could be expected to weaken, but considering the big plus or minus value, it could also not weaken at all (0%) or even strengthen (32%-37% is -5%). 

Now variability of projections this large is not helpful to policy makers. Thus the researchers looked at ways to constrain the model outcomes to reduce this variability. 

What they did:

What the research actually investigated was methods, several different ones, of reducing the variability of projected changes to the Atlantic Meridional Overturning Circulation some 70 years in the future from computer models of climate.

As the Guardian article states:

“The new research, published in the journal Science Advances, explored four different ways of using real-world observations to assess the models. They found a method called ridge regression, which had been little used in climate science before now, provided the BEST results.”

I have emphasized the word “best” in this quotation because its meaning is not clear. The word “best” could mean “more accurate” or it could mean something else like “making all the models agree with one another better”. In this case it only means reducing the variability of the models by some means.

The means employed here are rather complex. However, an honest description of the means leading to the greatest reduction in variance boils down to this: We add bias to the models in order to reduce variance. This is not something new. A person could ask, “Does the added bias, in addition to reducing variance among models, also push their average result toward a larger reduction of the AMOC?” If so, this looks a bit like an exercise in circular reasoning.

Explaining the full effort involves too much background explanation to be useful here. Yet, what results is that they are looking not directly at the processes that lead to the strength of the AMOC, but rather at the output of a limited number (32) computer models, and their measure of best is least variance, while their assessment of credibility involves using the output of the models themselves.

Their effort is mainly statistical data massaging but the description suggests a definitive conclusion drawn from first principles of AMOC dynamics.