Mike Jonas
… and wave as hard as you like. There’s no need to worry about disrupting Earth’s climate.
It has been a long time coming, but I have had another paper on Climate Modelling published – Chaos Theory is Void. It addresses the well-known and generally-accepted Butterfly Effect, aka Chaos Theory. I and others have felt for a long time that the Butterfly Effect exists only in models, and that Earth’s self-correcting weather/climate is too stable for the Butterfly Effect to exist to any meaningful extent in the real world. But how to prove that?
Two years ago, in GCMs Cannot Predict Climate, I described how GCMs (General Circulation Models) cannot predict climate because their iterative mechanism fails after a few weeks. Clearly the iterative mechanism itself is the problem. But how to prove that?
It all stayed on the back-burner for a while, but then I came across a paper that criticised an AI-based model for not being able to replicate the Butterfly Effect. In my previous paper I had argued that the structure of GCMs was upside-down: instead of working bottom-up from weather to climate, a climate model should work top-down with climate. Now, here was AI doing exactly what I argued for and being criticised for not having the same problem as GCMs!
That got me thinking about exactly how I could argue in favour of the AI-based model. The Butterfly Effect is difficult to argue against because it compares what is with what might have been. Since you can never tell what might have been, you can never disprove it.
And then the thought occurred to me: How large would a cloud have to be, to deliver the same global average temperature change as the adjustments to initial conditions used in the Kay et al paper (“Kay”) that demonstrated the Butterfly Effect and which I had cited in my previous paper. The answer: 1000 square metres. In just one hour, a single cloud, 25 metres by 40 metres, could change Earth’s global average surface temperature by as much as all the adjustments added together that Kay used to demonstrate the Butterfly Effect. Now we all know that the Butterfly Effect is not literally about a butterfly flapping or not flapping its wings, but this is a very small cloud (25×40 metres is about 30×45 yards). That butterfly is intangible, but Kay has given me a number to work with. Anyway, one thought led to another, and to the conclusion (carefully argued in the paper) that for all real world climate purposes, Chaos Theory can safely be ignored. It is scientifically inoperative. It is Void.
The two thoughts that led to this conclusion are:
1. There are a very large number of opportunities for a very small cloud to exist or not exist. We are not talking about just one cloud, we are talking about many millions of very small clouds that sometimes exist and sometimes don’t. Even if one very small cloud or non-cloud really can make Earth’s climate change as much as the Kay model does, it doesn’t have the planet to itself. There are all these millions of other potential clouds, and some of them, by existing or not existing, will pull Earth’s climate in one direction while other clouds and absent clouds will pull in the opposite direction. Maybe they all just cancel each other out?
2. GCMs try to model climate in small time steps through the daily cycle. At the end of each day, they return to a state that is often very close to the state of the day before. Any small model error during the day becomes a large error, relatively, at the end of the day. So you cannot model multiple days without knowing what makes the weather change over multiple days – after a few days, the GCM must fail. This is why weather forecasts become unreliable after a few days. The general wisdom is that there is an absolute 2-week limit for weather forecasts.
But what if we realise that the 2-week limit for weather forecasting is imposed by the models and not by the weather? There is then no intrinsic reason why a longer useful forecast cannot be made. Climate models can be improved similarly.
In fact, this is already happening. AI is increasingly being used for both weather and climate forecasting. AI is working top-down, and getting better results. We may soon be getting useful long term (eg, a month or two) weather forecasts. With top-down climate modelling and the death of RCP8.5 we might even get better climate predictions.
It probably can never be fully proved, but clearly the Butterfly Effect is a feature of GCMs that does not exist in the real world. Well, not quite “not exist”, but close enough. It can safely be ignored.
The Abstract of the paper:
This paper demonstrates that in the real world, Chaos Theory is Void. “Void” is used to mean that the theory is logically coherent and perhaps even true, but it can never contribute to scientific enquiry because it cannot be tested or applied. A Void theory is not necessarily false, it is just scientifically inoperative. Chaos theory does apply in General Circulation Models, but in this context there is no useful relationship between General Circulation Models and the real world. This has positive implications for both climate prediction and weather forecasting: by using a more appropriate model structure than General Circulation Models, it may be possible to improve climate prediction and to extend weather forecasting beyond the generally-accepted two-week prediction horizon.
I also introduce this general principle:
Any predictive system constructing cycles of variable duration and/or varying amplitude will fail after a few cycles if the mechanisms behind the variability are not fully understood and/or cannot be accurately replicated.
Before I wrote the paper, I used AI to do a thorough search of the scientific literature for anyone saying the things that I say in the paper. To my surprise I came up with nothing. The nearest was Krishnamurthy (2019) which said that “slowly varying components [..] provide a basis for predicting certain aspects of climate at longer range“. It got agonisingly close to limiting Chaos Theory to models, saying “imperfections in the models limit reliable predictability“, but there was still implicit acceptance that Chaos Theory applied in the real world. So I had to write my paper.
Footnote: I needed a clear and preferably short word to describe the non-applicability of Chaos Theory. I searched, and Grok searched, and we came up blank. “not falsifiable” is equated with “not science” by Karl Popper, “null” as in “a null hypothesis” is inaccurate, and so on. Then I thought of the legal expression “Null and Void” – “Null” and “Void” must have different meanings, otherwise they would only use one of them. Silly me, the law is not that logical, but in law there actually is a meaning of “Void” (without legal force or binding effect) which is analogous, so I used “Void”.
PS: I thought for a while about how to keep the paper’s title as short and clear as possible. At just 4 words, had I created the shortest scientific paper title ever? Alas, no, the record is still held by Professor Doron Zeilberger for a 2007 paper title 0 characters long. (The paper was about Nothing). That record will be hard to beat.
