From the COLUMBIA UNIVERSITY SCHOOL OF ENGINEERING AND APPLIED SCIENCE and the “learn garbage in, get garbage out” department.
Machine learning may be a game-changer for climate prediction
New York, NY–June 19, 2018–A major challenge in current climate prediction models is how to accurately represent clouds and their atmospheric heating and moistening. This challenge is behind the wide spread in climate prediction. Yet accurate predictions of global warming in response to increased greenhouse gas concentrations are essential for policy-makers (e.g. the Paris climate agreement).
In a paper recently published online in Geophysical Research Letters (May 23), researchers led by Pierre Gentine, associate professor of earth and environmental engineering at Columbia Engineering, demonstrate that machine learning techniques can be used to tackle this issue and better represent clouds in coarse resolution (~100km) climate models, with the potential to narrow the range of prediction.
“This could be a real game-changer for climate prediction,” says Gentine, lead author of the paper, and a member of the Earth Institute and the Data Science Institute. “We have large uncertainties in our prediction of the response of the Earth’s climate to rising greenhouse gas concentrations. The primary reason is the representation of clouds and how they respond to a change in those gases. Our study shows that machine-learning techniques help us better represent clouds and thus better predict global and regional climate’s response to rising greenhouse gas concentrations.”
The researchers used an idealized setup (an aquaplanet, or a planet with continents) as a proof of concept for their novel approach to convective parameterization based on machine learning. They trained a deep neural network to learn from a simulation that explicitly represents clouds. The machine-learning representation of clouds, which they named the Cloud Brain (CBRAIN), could skillfully predict many of the cloud heating, moistening, and radiative features that are essential to climate simulation.
Gentine notes, “Our approach may open up a new possibility for a future of model representation in climate models, which are data driven and are built ‘top-down,’ that is, by learning the salient features of the processes we are trying to represent.”
The researchers also note that, because global temperature sensitivity to CO2 is strongly linked to cloud representation, CBRAIN may also improve estimates of future temperature. They have tested this in fully coupled climate models and have demonstrated very promising results, showing that this could be used to predict greenhouse gas response.
###
About the Study
The study is titled “Could Machine Learning Break the Convection Parameterization Deadlock?”
https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2018GL078202
Authors are: P. Gentine1 , M. Pritchard2 , S. Rasp3 , G. Reinaudi1, and G. Yacalis2 (1Earth and Environmental Engineering, Columbia University, New York, NY, USA, 2Earth System Science, University of California, Irvine, CA, USA, 3Faculty of Physics, LMU Munich, Munich, Germany).
Wouldn’t it be a hoot if they created this amazing AI and fed in all of the climate nonsense and it came back and told them they were full of crap?
Anthony,
You wrote, “… (an aquaplanet, or a planet with continents)…” Should that be “a planet without continents”?
Machines can just do a better job of curve-fitting than climate modelers can do. The physics of clouds is no better understood by machines than it is by humans. Using machines to deploy the same flawed physics is just a different way to make the same mistakes.
There’s this weird almost subconscious idea spreading around that using computers to make decisions for humans somehow removes bias and prejudice. It doesn’t. All computers do is make prejudice dispassionate.
I don’t recall who said this, but the heart of the matter is that where there’s artificial intelligence, there is also artificial stupidity. Given the way things work AS will be more prevalent than AI.
Yes, that was my immediate reaction to the Article before I even started to read the comments. This won’t be “Artificial Intelligence,” this will be “Artificial Stupidity,” because they’ll just be feeding the newfangled machine the same old stupid input assumptions, and at the end of the day, no matter how “sophisticated” the computer, GIGO still applies.
I majorly concur.
I guess machine-garbage-out is somehow more acceptable, because then you can blame stupid machines instead of stupid humans. But no, stupid humans TAUGHT the machines to be stupid.
You know what I find so amazing? It is that 100% of the “physics” of purported global warming (caused mostly by CO₂ it is said, but also by CH₄ (methane), N₂O (nitrous oxide), O₃ (ozone), (CHCl)^ⁿ (chlorocarbons), (CFCl)^ⁿ (chlorofluorocarbons CFCs), and certain rare but potent industrial gas byproducts) can be computed to at least 2 digits of precision … using the CPU of a laptop.
Computing this doesn’t require a floor-full of hyper-parallel processors, billions if gigabytes of memory, the power consumption of a small town or the investment of tens — nay, hundreds — of millions of dollars in machinery, operators, programmers and public relations sycophants.
Indeed: there is not ONE extant piece of documentation showing that the results painstakingly derived from these giant sub-billion dollar pterodactyls … has resulted in even 3 sig-figs of precision in prediction, or for that matter, any unusual result neither previously predicted, or entirely unexpected.
And THAT is the point.
IF (as there must, at some level) there is anthropic greenhouse gas “global warming” enhancement of the climate, THEN it has mostly been sussed out in the late 19th and early 20th centuries. Which is sobering, at any level.
The amazing thing is the duplicity, the rhetorical chest beating, gnashing of teeth, ostracizing and denigration of the World Powers who haven’t the temerity to outright mandate the cut and bleed of their economy sovereignty by way of carbon-fueling their energy needs.
Yet, on this blue-green gem of a planet, we are adding a measurable and substantial amount of CO₂ to the atmosphere every year. I don’t think anyone here seriously considers this to be fiction. It is measurable and measured — Moana Loa famously — but also at hundreds of other observation sites globally. It is real, and it is more-or-less in lockstep with the collated global production numbers for coal, natural gas and petroleum. We don’t even need “consumption figures”.
Point tho is: the effect is rather small considering the absolute degree of hyperventilation going on in the pseudo-science community. The “Save the Trees, eat a Beaver” crowd desperately need a religious iconography to worship. It used to be (embarrassingly) boreal forests disappearance (which unhelpfully have been growing faster, thicker and more solidly with increased CO₂), it used to be glowing orange forests because of toxic acid-rain.
Now it is supercomputers, billion dollar “research and grant” budgets.
And frankly, I think its just gawdawfully mendacious. Because year-over-year, the people of the planet, still very much RAISING THEMSELVES UP BY THEIR BOOTSTRAPS, are digging up more coal to burn, more petroleum to refine, ore natural gas to cook, heat and produce avidly wanted electricity. Cars go, mopeds zoom, lights turn on, smart phones are recharged, and work, the Internet allows one-and-all to learn of the whole world’s trends, knowledge, fandom and hypocrisies.
Energy.
More every year.
Yielding, inevitably, more CO₂, CH₄ and rare-but-potent byproduct greenhouse gasses.
Just mendacious to think it needs billion-dollar research (accomplishing exactly naught).
GoatGuy
And to reduce the worry even further, even if the consumption of fossil fuels seems to coincide with the CO2 increase measured at Moana Loa, it may be just coincidence. Last report I read, of the CO2 in the atmosphere, the amount contributed by humans may be as low as 4%. Further, the economic depression of 2008 resulted in a dip in fossil fuel consumption, but Moana Loa did not record a corresponding decline in the rate of CO2 increase. So, burn baby burn! Be efficient, the less money you pay to power your production the better it is for the bottom line, but otherwise, burn all you need!
Don’t forget to train it to bully anyone who questions its assumptions, methodologies, or conclusions. and throw punch cards in the air for theatrics
The proposition to use AI to improve long-term climate predictions sounds like a non-starter. The project begins with the assumption that CO2 concentration in the atmosphere is a significant cause of global warming, i.e., another “What if?” study. The researchers say the model has been “tested” with good results. What does that mean? Good results might be to match GCMs, which have all been failures in predicting short-term temperatures and untested by long-term, real world data.
Cloud formation is a chaotic process. Will AI find order in a chaotic process? A shortcoming of AI in oil exploration is that it removes “anomalies” from databases. The “anomalies” in databases might be important in locating oil accumulations. They may also be important in climate studies.
I view a project like this one as a perpetual money sinkhole, a never-ending project going in the wrong direction.
RIGHT… neural networks always in training, ever tweaking, with massively parallel current states that are best described as “um, er, just like this, see here” at any given moment because they are self-modifying… are sure to do a better job any cell based deterministic model. This is obvious because when ever we have a tough job that has to be done right, we always ask a schizophrenic to do it. /s after witnessing personality simulations
[not /s, Q:] What exactly does “global temperature sensitivity to CO2 is strongly linked to cloud representation” mean?
“ What exactly does “global temperature sensitivity to CO2 is strongly linked to cloud representation” mean?”
It means when modelers change their parameter sets to some equally plausible set (i.e., within the limits of uncertainty), their model produces a different warming trend for the identical trend in CO2.
The cloud types, cover, and persistence change with the parameter sets, no matter that the forcing trend was the same.
Changing parameters to see what the model does is called a sensitivity analysis, or more generally, a “perturbed physics” test.
Although never interpreted this way in the field, the results show that, as regards CO2 and climate, no one knows what they’re talking about.
One grain of sense in the article that should be at the top of every report on climate predictions. “We have large uncertainties in our prediction of the response of the Earth’s climate to rising greenhouse gas concentrations. The primary reason is the representation of clouds and how they respond to a change in those gases.”
They’re a bit late in trying to hitch their research to the CLimate Grant Boondoggle. There’s a new sheriff in town and he ain’t impressed.
Not yet.
The people with the most money to spend on this right now, is Google.
Their biggest motivation and market is targeting advertising to consumers with AI/machine learning.
When I searched for X-ray spectrophotometers, I suddenly found myself receiving adverts for transparent underwear.
That is the current level of AI/machine learning.
“A major challenge in current climate prediction models is how to accurately represent clouds and their atmospheric heating and moistening. ”
And I thought that the climate science was settled. Stupid me.
I’ll trust the output of an uneducated Commodore 64 over anything Michael Mann produces.
Having just come back from a conference on AI, I understand that what makes AI work is realism – deep learning allows for ever closer and more accurate representations of the structures of reality. The more accurate the representation, the better the predictive capability.