When The Model Models Itself

Guest Post by Willis Eschenbach

Eric Worrell posted an interesting article wherein a climate “scientist” says that falsifiability is not an integral part of science … now that’s bizarre madness to me, but here’s what she says:

It turns out that my work now as a climate scientist doesn’t quite gel with the way we typically talk about science and how science works.

…

1. Methods aren’t always necessarily falsifiable

Falsifiability is the idea that an assertion can be shown to be false by an experiment or an observation, and is critical to distinctions between “true science” and “pseudoscience”.

Climate models are important and complex tools for understanding the climate system. Are climate models falsifiable? Are they science? A test of falsifiability requires a model test or climate observation that shows global warming caused by increased human-produced greenhouse gases is untrue. It is difficult to propose a test of climate models in advance that is falsifiable.

Science is complicated – and doesn’t always fit the simplified version we learn as children.

This difficulty doesn’t mean that climate models or climate science are invalid or untrustworthy. Climate models are carefully developed and evaluated based on their ability to accurately reproduce observed climate trends and processes. This is why climatologists have confidence in them as scientific tools, not because of ideas around falsifiability.

For some time now, I’ve said that a computer model is merely a solid incarnation of the beliefs, theories, and misconceptions of the programmers. However, there is a lovely new paper called The Effect of Fossil Fuel Emissions on Sea Level Rise: An Exploratory Study in which I found a curious statement. The paper deserves reading on its own merits, but there was one sentence in it which struck me as a natural extension of what I have been saying, but one which I’d never considered.

The author, Jamal Munshi, who it turns out works at my alma mater about 45 minutes from where I live, first described the findings of other scientists regarding sea level acceleration. He then says:

This work is a critical evaluation of these findings. Three weaknesses in this line of empirical research are noted.

First, the use of climate models interferes with the validity of the empirical test because models are an expression of theory and their use compromises the independence of the empirical test of theory from the theory itself.

Secondly, correlations between cumulative SLR and cumulative emissions do not serve as empirical evidence because correlations between cumulative values of time series data are spurious (Munshi, 2017).

And third, the usually held belief that acceleration in SLR, in and of itself, serves as evidence of its anthropogenic cause is a form of circular reasoning because it assumes that acceleration is unnatural.

Now, each of these is indeed a devastating critique of the state of the science regarding sea level acceleration. However, I was particularly struck by the first one, viz:

… the use of climate models interferes with the validity of the empirical test because models are an expression of theory and their use compromises the independence of the empirical test of theory from the theory itself.

Indeed. The models are an expression of the theory that CO2 causes warming. As a result, they are less than useful in testing that same theory.

Now, the scientist quoted by Eric Worrell above says that scientists believe the models because they “accurately reproduce climate trends and processes”. However, I see very little evidence of that. In the event, they have wildly overestimated the changes in temperature since the start of this century. Yes, they can reproduce the historical record, if you squint at it in the dusk with the light behind it … but that’s because they’ve been evolutionarily trained to do that—the ones that couldn’t reproduce the past died on the cutting room floor. However, for anything else, like say rainfall and temperature at various locations, they perform very poorly.

Finally, I’ve shown that the modeled global temperature output can be emulated to a very high degree of accuracy by a simple lagging and rescaling of the inputs … despite their complexity, their output is a simple function of their input.

So … since:

•  we can’t trust the models because their predictions suck, and

•  we can emulate their temperature output with a simple function of their input forcing, and

•  they are an expression of the CO2 theory so they are less than useful in testing that theory …

… then … just what is it that are they good for?

Yes, I’m aware that all models are wrong, but some models are useful … however, are climate models useful? And if so, just what are these models useful for?

I’ll leave it there for y’all to take forwards. I’m reluctant to say anything further, ’cause I know that every word I write increases the odds that some charming fellow like 1sky1 or Mosh will come along to tell me in very unpleasant terms that I’m doing it wrong because I’m so dumb, and then they will flat-out refuse to demonstrate how to do it right.

Most days that’s not a problem, but it’s after midnight here, the stars are out, and my blood pressure is just fine, so I’ll let someone else have that fun …

My regards to everyone, commenters and lurkers, even 1sky1 and Mosh, I wish you all only the best,

w.

My Usual Request: Misunderstandings start easily and can last forever. I politely request that commenters QUOTE THE EXACT WORDS YOU DISAGREE WITH, so we can all understand your objection.

My Second Request: Please do not stop after merely claiming I’m using the wrong dataset or the wrong method. I may well be wrong, but such observations are not meaningful until you add a link to the proper dataset or an explanation of the right method.