Claim: Artificial Intelligence can Improve Climate Models

Essay by Eric Worrall

If you have a possible missing variable problem, the solution is to add more arbitrary adjustments to your model?

AI Exposes Accelerated Climate Change: 3°C Temperature Rise Imminent

BY IOP PUBLISHING

AI-enhanced research shows regional warming will exceed critical thresholds faster than expected, with most regions surpassing 1.5°C by 2040. Vulnerable areas like South Asia face heightened risks, urging swift adaptation actions.

Three leading climate scientists have analyzed data from 10 global climate models, utilizing artificial intelligence (AI) to enhance accuracy. Their findings indicate that regional warming thresholds are likely to be reached sooner than previously estimated.

Elizabeth Barnes says: “Our research underscores the importance of incorporating innovative AI techniques like transfer learning into climate modeling to potentially improve and constrain regional forecasts and provide actionable insights for policymakers, scientists, and communities worldwide.”

Read more: https://scitechdaily.com/ai-exposes-accelerated-climate-change-3c-temperature-rise-imminent/

The referenced study;

Combining climate models and observations to predict the time remaining until regional warming thresholds are reached

Elizabeth A Barnes*, Noah S Diffenbaugh and Sonia I Seneviratne

Published 10 December 2024 • © 2024 The Author(s). Published by IOP Publishing Ltd
Environmental Research LettersVolume 20Number 1 Citation Elizabeth A Barnes et al 2025 Environ. Res. Lett. 20 014008DOI 10.1088/1748-9326/ad91ca

Abstract

The importance of climate change for driving adverse climate impacts has motivated substantial effort to understand the rate and magnitude of regional climate change in different parts of the world. However, despite decades of research, there is substantial uncertainty in the time remaining until specific regional temperature thresholds are reached, with climate models often disagreeing both on the warming that has occurred to-date, as well as the warming that might be experienced in the next few decades. Here, we adapt a recent machine learning approach to train a convolutional neural network to predict the time (and its uncertainty) until different regional warming thresholds are reached based on the current state of the climate system. In addition to predicting regional rather than global warming thresholds, we include a transfer learning step in which the climate-model-trained network is fine-tuned with limited observations, which further improves predictions of the real world. Using observed 2023 temperature anomalies to define the current climate state, our method yields a central estimate of 2040 or earlier for reaching the 1.5 °C threshold for all regions where transfer learning is possible, and a central estimate of 2040 or earlier for reaching the 2.0 °C threshold for 31 out of 34 regions. For 3.0 °C, 26 °C out of 34 regions are predicted to reach the threshold by 2060. Our results highlight the power of transfer learning as a tool to combine a suite of climate model projections with observations to produce constrained predictions of future temperatures based on the current climate.

Read more: https://iopscience.iop.org/article/10.1088/1748-9326/ad91ca

If I have understood correctly, they are essentially using the AI as a complex black box polynomial correction to their rather imprecise climate models, to try to squeeze out better answers. The polynomial is trained by comparing observed temperature data to model output, then the resultant amalgamation of climate models and AI polynomial corrections is extrapolated to try to predict future events.

The problem with this approach is it creates the illusion of accuracy, without actually knowing if greater accuracy has been achieved. An AI used in this way applies complex arbitrary “corrections” to input data, to generate a near perfect match to any data used to train that AI. But the AI knows nothing about the underlying physical phenomena. The AI might be able infer physical phenomena if it has enough data – or the AI could just make stuff up, especially if unknown critical input data is missing from the set of data which is used to train the AI.

AI does have a role in scientific analysis. In fields like drug discovery and complex optimisation problems, AI can produce excellent results.

But AI also has a well known tendency to go off the rails, to “hallucinate” false results.

An AI malfunction is not a problem if you can test the quality of the AI results immediately. But using AI to try to figure out how to correct climate models, where nobody will know for years or decades whether the AI got it right, then using those AI corrections to project future events, this seems a dubious use of artificial intelligence.

5 17 votes
Article Rating

Discover more from Watts Up With That?

Subscribe to get the latest posts sent to your email.

84 Comments
Inline Feedbacks
View all comments
December 31, 2024 2:05 pm

The word “intelligence”, artificial or otherwise…

… does not belong in a sentence with the words “climate model”.
____

comparing observed temperature data”

Do they mean data from urban tainted, unfit-for-purpose, mal-adjusted, and agenda-manufactured surface sites??

Reply to  bnice2000
December 31, 2024 3:48 pm

Has A.I. Looked at the Data on Hurricanes, Typhons, Tornadoes , Floods, Droubts, Sea level rise, ETC. they are not getting worse!!!

Rick C
Reply to  purecolorartist@gmail.com
December 31, 2024 7:19 pm

Forget all that – apply their method to the stock market! If they’re so good at predicting the future they should all be billionaires in a couple years. If they’re still pandering for research grants in 2027 we’ll know that their methods are crap.

Anyone who thinks that somehow AI will achieve accurate predictions of future events should bet their future on it. Coming soon – AI predicts the outcome of the next PowerBall Loto drawing.

Reply to  purecolorartist@gmail.com
January 1, 2025 5:26 am

Yes, AI has all that data about how hurricanes, and tornadoes and floods and droughts are not getting worse. It says so right in the IPCC documents. Surely the AI has perused the IPCC documentation. I would think that is one of the first things the programmers would feed the AI.

It’s all a Fraud. The whole Human-cause Climate Change narrative is a fraud.

Sparta Nova 4
Reply to  Tom Abbott
January 3, 2025 11:00 am

They selected 10 climate models. Out of how many? Seems like is 37 or something like that. What criteria were used to choose which models to use?

Lots of questions. No anweres.

Reply to  bnice2000
December 31, 2024 9:22 pm

The abstract is a meaningless word salad, full of hype and fuzzy jargon:

specific regional temperature thresholds are reached

What committee defines these “thresholds”? (Dare I ask.)

Tom Halla
December 31, 2024 2:11 pm

Large Language Models, miscalled AI, so far seem to be an idiot savant version of the programming team. Someone gave the program the training data, so it comes with the same sort of prejudices as Wikipedia.
What I expect are more subtle versions of Google’s AI, when asked to show 1943 German troops, showed East Asians and Blacks in Waffen SS uniforms.

Rud Istvan
Reply to  Tom Halla
December 31, 2024 2:28 pm

At least Google’s AI was DEI.

Reply to  Tom Halla
December 31, 2024 3:36 pm

GIGO is always a problem and currently no AI can discern what’s garbage and what isn’t.

abolition man
Reply to  More Soylent Green!
December 31, 2024 8:36 pm

More Garbage, Piled higher and Deeper Upon the Garbage In, Garbage Out!
MGPhDUGIGO for government applications!

leefor
Reply to  Tom Halla
December 31, 2024 7:06 pm

So has AI got beyond the 3-year-old child mentality yet?

Tom Halla
Reply to  leefor
December 31, 2024 7:11 pm

More like an adolescent with really profound autism spectrum. Machine vision is also something the programmers have not settled yet, as the programs could not solve the “rotate these stacks of blocks” type puzzles.

Reply to  Tom Halla
January 1, 2025 6:24 am

AI is pattern recognition software. Therefore it is going to respond with the consensus version of whatever part of “the cloud” it searched. We have so many cases of the consensus being shown to be wrong after a few years that it can’t be considered anything other than a cursory review….and letting AI make consequential decisions could well be worse than allowing politicians to make them….

December 31, 2024 2:11 pm

comment cancelled because I didn’t mean to post it yet.

Reply to  bnice2000
December 31, 2024 9:42 pm

What I was going to say is that climate models are built on erroneous not-real physics from the ground up. (see Shula/Ott video further down)

Doesn’t matter if they try to implement AI… just compounding the junk that is already there.

Reply to  bnice2000
January 1, 2025 5:38 am

Digesting the bogus, bastardized Hockey Stick data will start AI down the wrong track.

AI will be trying to make sense from bogus, made-up temperature data. And, as we see, if it is projecting a 3C rise in temperatures, then it is not dealing with reality already. It’s been fed a bunch of crap and then it comes to erroneous conclusions.

Reply to  bnice2000
January 1, 2025 5:33 am

If you write a comment that you decide not to post, you can just erase the comment from the comment box, and then click on “reply” and the comment box will close and nobody will be the wiser. 🙂

Reply to  Tom Abbott
January 1, 2025 11:50 am

I tried that and it kept coming back.. wordpress can be odd sometimes.

Scarecrow Repair
December 31, 2024 2:16 pm

My mind glazes over reading such gobbledegook

My limited experience with neural nets (6 months) tells me they should have trained it on old models predicting their future, our past, and tested it similarly (we used a random half do train it, the other half to test it, and ran it many times with different random halves). My brief skim saw nothing about what they trained and tested it on. I did note this confusing line:

we include a transfer learning step in which the climate-model-trained network is fine-tuned with limited observations, which further improves predictions of the real world.

“Limited” observations? They threw in arbitrary cherry-picked corrections? I have no idea what it really means, but it sure doesn’t sound objective.

Then there’s this:

Using observed 2023 temperature anomalies to define the current climate state

More cherry picked data, and only temperatures? What about humidity, rainfall, and other aspects of weather?

Eh, it’s all pointless anyway. They start with the assumption that something bad is going to happen, and only ask “how soon?”

However, despite decades of research, there is substantial uncertainty in the time remaining until specific regional temperature thresholds are reached

Reply to  Scarecrow Repair
December 31, 2024 2:45 pm

2023 just happened to have a strong, persistent El Nino event, exacerbated by the HT effect and decreased cloud leading to increase absorption of solar energy.

Do you think they programmed that in !!!

Robert Cutler
Reply to  Scarecrow Repair
December 31, 2024 5:39 pm
  1. No self-respecting data scientist would call a convolutional neural network AI. It’s simply machine learning.
  2. No self-respecting data scientist would attempt to train a neural net on the limited, low-accuracy climate data. There’s not enough data to split into training and validation datasets. Weather data, absolutely. Climate data, never.
  3. To predict into the future the model will have to be fed estimates of the future values of parameters in the training data. Where will those come from? More models? In short, there are as many control knobs as input parameters which allow the output to take on any desired value.
  4. The problem with non-parametric models is that you can never explain how they work. Well, perhaps that’s an advantage because there’s no longer a need to estimate troublesome parameters like ECS.
  5. Now, if they trained the model using sunspot data as one of the inputs (rather than low-ball estimates of TSI), the model might reproduce the same results as my 99-year moving average model, but they’d never be able to figure out why, and they’d lose interest when they found the model had the greatest sensitivity to that input.
  6. This is clearly part of a propaganda campaign intended to ride on the current AI coattails. In other words, attempting to add public credibility to meritless predictions.
Reply to  Robert Cutler
January 1, 2025 5:40 am

“This is clearly part of a propaganda campaign intended to ride on the current AI coattails.”

Yes, it is.

Reply to  Scarecrow Repair
December 31, 2024 7:33 pm

“‘Limited’ observations? They threw in arbitrary cherry-picked corrections?”

It sounds suspiciously like the tuning to agree with historical data, which is already done.

Reply to  Clyde Spencer
January 1, 2025 5:42 am

There’s historical data, and then there’s “historical data”.

Some of that historical data has been bastardized to the point of being unrecognizable.

Peter Barrett
December 31, 2024 2:17 pm

I am reminded of my first interaction with a computer, a vast mainframe housed in a university building basement behind a glass panelled wall through which the operators, fully gowned and hairnetted could be observed. We mere mortal students could only approach as far as the hatch through which we passed our stacks of punched cards, reporting back the following day to collect the inevitable list of programming errors.

Late sixties, of course, and our lecturer started every talk by writing on the blackboard (yes, blackboard!) “GARBAGE IN, GARBAGE OUT”. Some things are constant.

Reply to  Peter Barrett
January 1, 2025 5:44 am

Don’t drop that stack of punch cards!

I did that once. I learned my lesson. 🙂

Reply to  Tom Abbott
January 1, 2025 6:29 am

I recall the hard-core computer jockeys always being armed with a big fat Magic Marker for marking the tops of decks.

Also remember going to the Student Center to buy cards — the serious players bought them in the big cardboard boxes. Don’t remember how much they cost.

old cocky
Reply to  karlomonte
January 1, 2025 12:03 pm

I recall the hard-core computer jockeys always being armed with a big fat Magic Marker for marking the tops of decks.

The diagonal lines, Vs and Ws in Texta along the edges of the cards were a good idea, but had problems if you had to insert lines of code.

That’s why BASIC had line numbers. The first few columns of FORTRAN were reserved for comments, which tended to be used for card numbers as well. I think COBOL may have done the same. And you numbered the original cards in increments of 10 so you could add extra lines.

Then, of course, there were the prefix job control cards before getting into the program itself.

We thought we’d died and gone to heaven when we got to use the teletypes on the PDP minicomputer.

But you try and tell the yoong people today that; and they won’t believe ya.

Rud Istvan
December 31, 2024 2:27 pm

What this new paper proves is that climate researchers are devoid of intelligence, either artificial or natural.

AI cannot ‘improve’ climate models easily shown wrong both in ‘best tuned parameter’ hindcasts, and by predicting a non-existent tropical troposphere hotspot. AI trained on past falsehoods merely reproduces those falsehoods. A ‘convolutional neural network’ just reproduces convoluted falsehoods.

Convoluted has two textbook definitions:

  1. Extremely complex and difficult to follow.
  2. Intricately folded and twisted.

Both apply here.

Izaak Walton
Reply to  Rud Istvan
December 31, 2024 2:54 pm

And convolution has a technical meaning that has nothing to do with the “textbook definitions” above. See
https://en.wikipedia.org/wiki/Convolution

Reply to  Izaak Walton
December 31, 2024 6:58 pm

convolution has a technical meaning that has nothing to do with the “textbook definitions” above.

Whilst this is true. Its also true to say

AI cannot ‘improve’ climate models easily shown wrong both in ‘best tuned parameter’ hindcasts, and by predicting a non-existent tropical troposphere hotspot.

In principle, AI is great at interpolation but not extrapolation. There are exceptions of course, but not with GCMs and climate. It just not that kind of problem.

In other news, ChatGPT retires after winning Lotto.

Bryan A
Reply to  Izaak Walton
December 31, 2024 7:52 pm

Con-volution: A Con used to support a Socialist Revolution

Reply to  Izaak Walton
December 31, 2024 7:55 pm

Please explain how the mathematical definition of ‘convolution’ as provided by you applies to the use in association with “convolutional neural network,” or as used by the referenced article, “incorporating innovative AI techniques like transfer learning into climate modeling.”

Reply to  Clyde Spencer
December 31, 2024 8:11 pm

Please explain how the mathematical definition of ‘convolution’ as provided by you applies to the use in association with “convolutional neural network,”

Yes, good point. I’d taken Izaak’s statement “nothing to do with the “textbook definitions” above” at face value and didn’t look at his link. It turns out neither of them is even close.

https://en.wikipedia.org/wiki/Convolutional_neural_network

Izaak Walton
Reply to  TimTheToolMan
December 31, 2024 9:26 pm

The wikipedia article states that “convolutional neural network(CNN) is a regularized type of feed-forward neural network that learns features by itself via filter (or kernel) optimization.” The use of a filter means that essentially the output can be considered as a convolution between the input and the filter or kernel. The machine learning comes from the fact that the precise properties of the filter are unknown to begin with and has to be set by training the neural net. It is not identical to a mathematical convolution but similar enough.

Reply to  Izaak Walton
December 31, 2024 11:41 pm

It is not identical to a mathematical convolution but similar enough.

Its not even close Izaak. That’s like saying neural networks are nothing but mathematics. Its so far away, its not even wrong.

Izaak Walton
Reply to  TimTheToolMan
December 31, 2024 11:59 pm

Of course neural networks are nothing but mathematics. They only exist inside a computer program. All programs are nothing but mathematics. Turing showed that decades ago.

Reply to  Izaak Walton
January 1, 2025 12:51 am

Of course neural networks are nothing but mathematics.

Congratulations. You have exceeded my expectations.

Izaak Walton
Reply to  TimTheToolMan
January 1, 2025 2:18 pm

And would care to explain why programs are not mathematics?

Reply to  Izaak Walton
January 1, 2025 4:55 pm

Of course neural networks are nothing but mathematics.

Firstly your brain is a neural network and its not mathematics in a computer. Secondly this was about describing what a convolutional neural network is and its not even slightly described as just “mathematics” even though there is mathematics used in its creation.

But now you have a new claim that “programs” are mathematics and that’s not true either.

You may as well claim a flower pot is mathematics because it looks like anything with mathematics describing it or underlying its creation is just “mathematics” in your definitional world.

Izaak Walton
Reply to  TimTheToolMan
January 1, 2025 11:04 pm

Tim,
it is well known that programs are independent of the hardware running them. And the algorithms themselves are all instances of Turing machines. As such they correspond to functions from the integers to the integers and nothing more. And as Turing machines are definitely part of mathematics so is all of computing.

Similarly the brain can be thought of as a computer and again the Church-Turing thesis applies, computability means able to be computed using a Turing machines. Thus the human mind is nothing more than a particularly complicated Turing machine. Alternatively the physical processes that occur within a neuron can be simulated to an arbitrary degree of accuracy using a digital computer and thus again the entire human brain could in principle be simulated and so is nothing more than a Turning machine.

Reply to  Izaak Walton
January 1, 2025 11:38 pm

And a flowerpot can be represented mathematically to whatever level of detail you require but it doesn’t make it a mathematical construct.

A trained neural network, whether that be a convolutional neural network or biological one, contains information which is abstractly represented and an emergent property.

Describing the “convolutional” part of a CNN separately with its mathematical meaning is about as useful as defining yellowcake as being uranium and trying to define cake in any common usage.

In the case of CNNs, convolutional is part of the name but only very obliquely has anything to do with mathematics.

Reply to  TimTheToolMan
January 2, 2025 7:22 am

How do you describe “intuition” using mathematics? How do you describe “creativity” using mathematics?

The human brain may be partly a neural network but it is *more* than that. Show me a Turing machine that can “create” something new.

Reply to  Izaak Walton
December 31, 2024 11:57 pm

A euphemism for data torture….. OK!

So it is going to be useful as a “climate” thing !

All climate data has been “convoluted”…

… that’s where GISS et al come from.

Reply to  Izaak Walton
January 1, 2025 12:12 am

Izaak,

You need to reread Rud’s comment and not mix up “convolution” with “convoluted”

DMA
December 31, 2024 2:48 pm

AI cannot improve models that are using erroneous physics. Here is a link to a video where Tom Shula falsifies the greenhouse effect hypothesis and shows a good analysis of the processes in the atmosphere that have produced the data that, even by the best, has been misinterpreted as evidence of that effect.

Reply to  DMA
December 31, 2024 3:57 pm

It really is a very good idea to watch this video at least a couple of times. !

Reply to  bnice2000
December 31, 2024 6:34 pm

I second that.

Reply to  DMA
December 31, 2024 10:52 pm

At an average radiating power of 240W/m^2, the radiating temperature of water is 260K. At that average temperature, the vast majority of the water is ice.

These guys do a good job on demising the “greenhouse effect” but talking about water vapour and gases is not very relevant to the radiation heat balance in Earth’s atmosphere. No one can grasp what is happening in the climate system unless they first understand the basics of ice formation; whether that be on land; on water or in the atmosphere.

Ice knocks out 29% of the incoming solar EMR before it even gets a change to thermalise. Ice is far more important than any of the gasses.

Giving_Cat
December 31, 2024 2:55 pm

AI heterodynes AI results as if it were data.

Ed Zuiderwijk
December 31, 2024 2:56 pm

0 x N = 0, however big, large or intelligent N is.

CFM
December 31, 2024 3:11 pm

Nice picture

December 31, 2024 3:15 pm

What would be hilarious is if the AI analyzed the data, set TCR=ECR=1.00, faithfully modeled temperatures and declared no problem.

Yeah, like that solution would ever have a chance of seeing the light of day.

December 31, 2024 3:16 pm

It has long been my hope that true artificial intelligence will blow the lid off this entire climate scam.

Clarky of Oz
December 31, 2024 3:26 pm

Seems appropriate seeing the whole model thing is artificial anyway. The only thing I would question is the use of the word “intelligence”

December 31, 2024 3:31 pm

Gee, I thought climate models already were artificial intelligence

Reply to  wilpost
December 31, 2024 3:54 pm

NO, they may be totally artificial…..

… but they are also totally NON-intelligent.

(like their authors)

David Wojick
December 31, 2024 3:32 pm

Also they used a year that is the tip of a big transient El Niño driven temp spike to define the climate state. The models do not see such spikes so the AI will “correct” these already unrealistically hot models even hotter. No wonder they get unrealistically fast warming.

son of mulder
December 31, 2024 3:34 pm

Climate is chaotic. They can’t possibly give reasonable, long range, regional predictions no matter how much AI is used. It’s 100% Snake oil.

December 31, 2024 3:39 pm

The models use unproven assumptions on how the climate system works and the datasets are incredibly dirty. AI can’t overcome that.

Ex-KaliforniaKook
December 31, 2024 3:47 pm

The expression “Echo Chamber” comes to mind. The AI learns from the internet, which prioritizes climate alarmism, and viola! provides predictions of calamity.

Who da thunk?

John Hultquist
December 31, 2024 3:50 pm

Hasn’t the 1.5 C degree already been surpassed? Looking at a chart of Little Ice Age (estimated) temperatures makes it look that way. For what it is worth!

Reply to  John Hultquist
January 1, 2025 5:55 am

Current temperatures are cooler than both the 1998 high point and the 2016 high point.

comment image

Bryan A
December 31, 2024 3:52 pm

Looking at the UAH_LT chart on the sidebar displaying UAH Global Temperatures,
comment image?resize=1320%2C594&ssl=1
From whatever 30 year period the baseline is set to, back in 1984 the low was -0.68°C below the baseline and the 2024 high point was +0.95°C above the baseline. So from 1984 to 2024 (40 years) Global temperatures increased a total of 1.63°C. The increase would be well over 2°C since pre industrial times…AND…no tipping points passed the world goes on and in fact temperatures dropped From 0.95°C to 0.64°C (3/10)
Still no tipping
No runaway Warmageddon
Guam is still upright
The oldest living US president is still in office (somewhere)
Nancy Lugosi still sits in dark corners during the day and is seen out at night

abolition man
Reply to  Bryan A
December 31, 2024 8:43 pm

“The oldest living US president is still in office;”
I think they have him propped up in the janitor’s closet until they need him for a photo op!

Bob
December 31, 2024 3:58 pm

Very nice Eric. A perfect misuse of AI. We start out with the last miracle tool the climate model. A system so complex and dealing with so much information only the newest most powerful computers can be expected to do a credible job. But even these powerhouses can’t produce results that agree with one another or accurately match observations. Clearly the information input is not correct or the algorithms aren’t up to snuff. Rather than admit their theory may be mistaken or change the input or change the algorithm these geniuses clamber onto the newest computer wonder, AI. They figure if they add some observations and AI to their models the combination will predict when the answers they want may come true. This is a stinking pile of bull feces.

December 31, 2024 4:17 pm

Three leading climate scientists have analyzed data from 10 global climate models, utilizing artificial intelligence (AI) to enhance accuracy.

Accuracy. I don’t think that word means what they think it means…

Chris Hanley
Reply to  Zig Zag Wanderer
December 31, 2024 5:03 pm

Also climate models do not produce data.

December 31, 2024 5:02 pm

Chuckle.
Next, we will be told the heat islands surrounding AI centers are the cause.

Reply to  Eric Worrall
December 31, 2024 8:13 pm

43 is good enough for government work.

Editor
December 31, 2024 5:08 pm

If AI is instructed to show future global cooling, it will do just that – with ease.

ScienceABC123
December 31, 2024 5:32 pm

There have already been instances of A.I.’s showing bias. So I’m sure they can support any climate position their programmers put into them. The real question is – Why would we listen to them?

Verified by MonsterInsights