We have talked previously about the extreme difficulties of modelling clouds, and why it is a problem for the science-is-settled crowd since clouds strongly influence the climate. A change of only 4 percent in the amount of global low cloud cover, which is a mere rounding error in the models, would change the global climate by 2 degrees C, the amount the Paris Accord is supposedly trying to prevent. A new review of the state of cloud modeling (h/t NoTricksZone) goes even deeper into the problems for climate forecasting. Models currently use loose statistical approximations of the physics to predict how cloud patterns will change at the global level. Why not use the actual physics? Awkwardly, it’s because if scientists were to try to do all the calculations of the known processes that operate within clouds, using the fastest computers available, they would be only able to predict the behaviour of a cloud no larger than... one cubic meter.

Now scale the problem up to the size of the global atmosphere and try to predict global cloud cover to within 4 percent accuracy a century from now. It can’t be done, end of story.

So climate modelers make do with their approximation schemes, called parameterizations. But the authors of the new study (Morrison et al.) explain that there are fundamental problems with the way climate (and weather) modelers go about this. First, they’re trying to approximate physical processes that no one has been able to measure accurately. And if you don’t know what you’re trying to approximate, you can never tell if you’re getting the wrong answer. Morrison et al. even quote the famous philosopher of science Karl Popper who in the 1950s warned against scientists peddling abstract theories and models that couldn’t be tested against the data: “In so far as a scientific statement speaks about reality, it must be falsifiable: and in so far as it is not falsifiable, it does not speak about reality.” Climate model cloud schemes might be right. Or they might not be. And until we know one way or the other, they aren’t reality.

Can the measurements and computer approximation schemes be improved? Well yes, say Morrison et al., but only if scientists test them in laboratory settings and start discarding the ones that fail. But for some reason, scientists have all but stopped doing this. Morrison et al. cite data showing that in the 1970s, about 40% of the papers presented at cloud physics conferences were reporting on experimental studies, whereas today the number is down to about 8%, and some of the leading cloud experimental laboratories have shut down. The cost is that many approximation schemes used by the modeling community never get subjected to experimental testing, making the outputs of said models little more than clouds of guesswork.