Forgot your password?
typodupeerror
Earth Science

Climate Physicists Face the Ghosts in Their Machines: Clouds (quantamagazine.org) 25

Climate scientists trying to predict how much hotter the planet will get have long grappled with a surprisingly stubborn problem -- clouds, which both reflect sunlight and trap heat, account for more than half the variation between climate predictions and are the main reason warming projections for the next 50 years range from 2 to 6 degrees Celsius.

Two research groups are now racing to close that gap using AI, though they disagree sharply on method. Tapio Schneider at Caltech built CLIMA, a model that uses machine learning to optimize cloud parameters within traditional physics equations; it will be unveiled at a conference in Japan in March. Chris Bretherton at the Allen Institute for AI took a different path -- his ACE2 neural network, released in 2024, learns from 50 years of atmospheric data and largely bypasses physics equations altogether.
This discussion has been archived. No new comments can be posted.

Climate Physicists Face the Ghosts in Their Machines: Clouds

Comments Filter:
  • "Two research groups are now racing to close that gap using AI, though they disagree sharply on method."

    Isn't AI supposed to be THE METHOD? Don't you use AI to improve the model? And are they "racing"?

    "...his ACE2 neural network, released in 2024, learns from 50 years of atmospheric data and largely bypasses physics equations altogether."

    Seems unlikely, but by doing precisely what? Producing an alternative model? Why does that matter? Isn't the result what matters?

    There are climate models. You judge t

    • by PPH ( 736903 )

      There are climate models. You judge them on the reliability of their predictions.

      Right. But if they are admitting to some major uncertainties in the models, they aren't ready to use them as the basis of trillion dollar investments. So keep working. The science isn't done yet.

      • by dfghjk ( 711126 )

        "But if they are admitting to some major uncertainties in the models, they aren't ready to use them as the basis of trillion dollar investments."
        Conveniently conflating two different groups as "they", then lying about one of them.

        "The science isn't done yet."
        Science is never done, that's not a statement of whether science IN THIS CASE is good enough.

        • by PPH ( 736903 )

          whether science IN THIS CASE is good enough.

          account for more than half the variation between climate predictions and are the main reason warming projections for the next 50 years range from 2 to 6 degrees Celsius.

          It appears not to be. One tail of the probability curve is over in ice age territory.

    • Another way of phrasing this:

      Two projects aim to come up with new models. One is using vast amounts of data and back propagation to learn the values of various coefficients in an existing model. Another is using vast amounts of data and back propogation to come up with a new model that is a vast array of linear equations combined with sigmoid functions.

      Phrased that way, it becomes clear that the former is likely to be far more useful for extracting understanding of whatâ(TM)s going on, while the latt

    • by Ed_1024 ( 744566 )
      I think they are saying that if they use AI, we will get 6C warming from all the extra data centres and thus close the gap in the predictions. Yay!
  • Alchemy? (Score:4, Insightful)

    by Quakeulf ( 2650167 ) on Monday February 23, 2026 @12:59PM (#66005692)

    At what point does this cease to be science and starts becoming alchemy? What is the cutoff point?

    • Re: (Score:3, Insightful)

      by groobly ( 6155920 )

      It was never science. Modeling is only science if the model is actually tested. These models claim to predict what will happen in 50-100 years. They will become science only after someone checks the predictions in 50-100 years.

      • Yes and no - they claim to predict things *over* the next 50 years. We can check them each and every second until 50 years passes with increasing confidence (or lack there of) in them. The predictions made in the 80s and 90s have so far turned out to be largely accurate, so seems like weâ(TM)re more in the science box than the alchemy one.

      • by smbell ( 974184 )

        This ignores both how models are tested and verified against existing data, and how reliable these models have been over the past decades.

      • by taustin ( 171655 )

        They've been making predictions for 50 years.

        With very close to a 100% failure rate.

        • by smbell ( 974184 )

          With very close to a 100% success rate.

          FTFY

        • by Tablizer ( 95088 )

          > With very close to a 100% failure rate.

          Bullshit! While specific forecasts about specific places have been wrong, in general warming continues and the sea is gradually rising. You are cherry-picking after the fact.

      • We call it Climate Ouroboros.

        The models are tuned quite literally to create today's weather from yesterday's inputs.

        This is how they calibrate them.

        And then the 'climate scientists' turn around, run today's weather through them and - voila! - warming!
        "Proved" by models.

        If these models didn't show warming, you would throw them out as broken models and look for another.

        The early IPCC reports didn't even regard water vapor and clouds as - essentially - too complicated to model.

        This paper from 2024: https://www [mdpi.com]

  • We really don't know clouds at all.

  • Clouds at night keep the heat in. (Keeping it warm)
    Clouds in the daytime reflect the heat of the sun, making it cooler.

  • We know clouds are a problem in computer modeling. But the headline says "their computers".

Some people manage by the book, even though they don't know who wrote the book or even what book.

Working...