Machine learning may be a game-changer for climate prediction

June 20, 2018, Columbia University School of Engineering and Applied Science
Credit: CC0 Public Domain

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, 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) , with the potential to narrow the range of prediction.

"This could be a real game-changer for ," 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 . 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 '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 , 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.

Explore further: New approach to global-warming projections could make regional estimates more precise

More information: P. Gentine et al, Could Machine Learning Break the Convection Parameterization Deadlock?, Geophysical Research Letters (2018). DOI: 10.1029/2018GL078202

Related Stories

Study could mean greater anticipated global warming

November 22, 2010

Global climate models disagree widely in the magnitude of the warming we can expect with increasing carbon dioxide. This is mainly because the models represent clouds differently. A new modeling approach successfully simulates ...

Tagging tiny particles in turbulent clouds

May 30, 2014

(Phys.org) —Hitching tiny atmospheric particles to cloud formation enables climate models to represent the particles' effects on convective storm systems. Scientists at Pacific Northwest National Laboratory and collaborators ...

Recommended for you

New insight into Greenland's melting glaciers

July 17, 2018

New research into Greenland's glaciers will help bring accurate sea level rise forecasts – which are crucial in preparing for the impacts of climate change—a step closer.

Thawing permafrost microbiomes fuel climate change

July 16, 2018

A University of Queensland-led international study could lead to more accurate predictions or the rate of global warming from greenhouse gas emissions produced by thawing permafrost in the next 100 years.

15 comments

Adjust slider to filter visible comments by rank

Display comments: newest first

eric96
1 / 5 (1) Jun 20, 2018
AI and statistics (rather than people) should be exclusively used to predict weather; however it should be city specific, and a prediction should be made every 15 min. It should also be able to filter bad data. That is if a specific source of data is nearly always wrong -> exclude that source. And all this should be updated to an app on our phone. Of all the tech professions, meteorology should be the first to go. To accomplish all this requires a substantial investment in supercomputer power. The benefit is, the weather reports will be very consistent and should in theory on get better and better. There would still be a need for meteorology researchers; to detect new variables or weather phenomena happening on earth. All this should have been implemented in the year 2000; however the cost of the computing power at the time was too prohibitive. However, a supercomputer is only as good as the people who program them.
NoStrings
1 / 5 (1) Jun 20, 2018
Yes, we can't do it. So let the buzzword black box do it.

Anyone remembers 1990th Expert Systems? They were supposed to answer all our questions >25 years ago. And here we are again.

And back comes Vanilla Sky too, how fitting.

grandpa
1.8 / 5 (5) Jun 20, 2018
That is all fine and good if someone want to improve of predictions of global climate change. It is however likely that there are too many variables to make accurate predictions. It is obvious that technology is changing rapidly where fossil fuels and nuclear fuels will be obsolete in 25 years. It is also obvious that the earth was heading toward ice earth and human activity by fossil fuel burning have probably allowed earth to head off this disaster. The last 3 million years have been rapid cycles of heating and cooling earth. Not good. More carbon dioxide will make this situation more stable. Almost like humans are a gift to earth. Also humans have been increasing the introduction of new species to areas which accelerates the introduction of stronger species and creates a greener earth. Both of these should reverse the desertification of earth. Almost like humans were sent to earth to save it.
antialias_physorg
5 / 5 (2) Jun 20, 2018
Anyone remembers 1990th Expert Systems?

Experts systems are just one form of machine learning. Actually they are the lowest form because an expert system just defines a set of 'if-then' connections and then let's the system figure out which set of such rules are appropriate to a given set of input parameters. If you know how the programming language Prolog works then that's basically the setup of expert systems.

That an expert system is the 'lowest form of machine learning' doesn't mean it isn't powerful. But today there are a lot more powerful sets of patterns within the machine learning domain available.

No, machine learning does not answer all questions (no one besides some populist journals on the level of National Enquirer or SciFi writers ever claimed this). But it can make sense of very complicated patterns that human minds aren't equipped to handle.
Gigel
not rated yet Jun 20, 2018
@grandpa: warm-cold cycles are really a puny problem for the environment compared to some real ones that will happen given enough time:
- a large asteroid striking Earth
- a wandering star going close to the Solar System
- a supernova close by
- the Sun's old age warms the Earth, which drives higher rock erosion and with it the absorption of all the CO2 in the atmosphere, with a total loss of plant and most known life; this will probably happen in a few hundred million years
- and the last one, the Sun frying completely Earth

Humans are the only species that can detect these dangers and with enough progress learn to avert them. There may not be enough time for another intelligent species to arise; it took 500 million years from first skeletal fossils to us. So we may really be Earth's last chance to thrive. It will take some tens of years of progress at current rate for us to reduce our influence on the environment. Current problems will be mostly gone by then.
BobSage
1 / 5 (3) Jun 20, 2018
So, it appears the old computer models actually weren't very accurate at all. Yet they are the basis for the predictions of global warming.

Regardless of what kind of computer models are created, they can be written to reach whatever conclusion the designer requires.
antialias_physorg
5 / 5 (2) Jun 20, 2018
So, it appears the old computer models actually weren't very accurate at all.

I think you have a bit of a reading comprehension issue. Nowhere is this stated.
antialias_physorg
not rated yet Jun 20, 2018
double post
Shootist
2.3 / 5 (3) Jun 20, 2018
learning to predict a random system. let's try craps next, eh? you can pay your carbon tax that way.
antialias_physorg
4.2 / 5 (5) Jun 20, 2018
Where do you get that it is a random system? Climate follows the laws of physics like everything else.

Or are you confused about the difference between the words "probabilistic" and "random"? There are books called 'dictionaries' that will help you out.
grandpa
1 / 5 (2) Jun 20, 2018
Gigel, Mostly a good point. A total iceball earth we may have been heading to would be bad too.

From Gigel, " @grandpa: warm-cold cycles are really a puny problem for the environment compared to some real ones that will happen given enough time:
- a large asteroid striking Earth
- a wandering star going close to the Solar System
- a supernova close by
- the Sun's old age warms the Earth, which drives higher rock erosion and with it the absorption of all the CO2 in the atmosphere, with a total loss of plant and most known life; this will probably happen in a few hundred million years
- and the last one, the Sun frying completely Earth

"
ddaye
not rated yet Jun 20, 2018
the words "probabilistic" and "random"?
It may seem obvious to you, but I haven't ever seen these 2 terms used to clarify the nature of processes like climate or evolution in mainstream US political debates over contentious topics. This country at least is very under supplied with science advocates that can and will clarify such basic concepts in public policy discussions that reach average people.
SamB
1 / 5 (1) Jun 20, 2018
Where do you get that it is a random system? Climate follows the laws of physics like everything else.


Yes, I would suggest that climate does follow the laws of physics. However, the 'physics' changes so very rapidly in such irregular ways that it appears 'random' to us and therefore is useless for predictions. A computer can not duplicate any climate system in any truly predictable way. (no pun intended). Scientists who 'tweak' computer programs to 'predict' what they expect to see are just kidding themselves and the public too.
antialias_physorg
5 / 5 (1) Jun 21, 2018
However, the 'physics' changes so very rapidly in such irregular ways that it appears 'random' to us and therefore is useless for predictions

It doesn't change and there is nothing random about it - only probabilistic. The two are entirely different critters.

A computer can not duplicate any climate system in any truly predictable way.

And yet they have been doing just that for the past decades.

Scientists who 'tweak' computer programs to 'predict' what they expect to see are just kidding themselves and the public too.

Incorporating an hitherto not considered factor is not 'tweaking'. Tweaking would be changing stuff to get a desired result. (And no, refining such factors as cloud production does not radically change the outcomes)
antialias_physorg
5 / 5 (1) Jun 21, 2018
Here's a short primer on the difference between random and probabilistic

Flip a fair coin. The outcome is random.

Flip a fair coin a thousand times and add up the number of times you get 'heads'. "Random" would be if every number were equally likely. Probabilistic is when you get a distribution

Betting money on a single coin flip is foolish. Betting it on being within the central 100 vs. the extreme 100 outcomes for 1000 coinflips is a (very much) winning strategy

Climate models are very much in the probabilistic camp (a LOT more so than 1000 coinflips)

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.