It’s only recently that researchers have begun to narrow it down—and improvements in cloud research have had a lot to do with it.
Last year, a groundbreaking new study found that a doubling of CO2 likely would result in warming of anywhere from 2.6 degrees to 3.9 degrees Celsius.
It’s a substantially narrower projection, ruling out some of the higher-end projections and eliminating much of the lower range. The study pulled together all the most recent research on climate sensitivity, accounting for multiple different lines of evidence—including recent advancements in cloud research.
And over the last few months, several recent studies—focused primarily on clouds—also have supported a narrower climate sensitivity range.
A February study in Nature Climate Change suggested a likely sensitivity of around 3.5 C. A May study, also in Nature Climate Change, put it around 3 C. Both studies suggested that clouds, on a worldwide scale, probably would have a moderate amplifying effect on the rate of global warming.
These studies used real-world observations to draw their conclusions. They compiled large quantities of data on cloud behavior—how clouds react to changes in temperature, humidity and other weather variables—and then conducted statistical analyses of those observations to figure out how clouds are likely to respond to future climate change.
It’s a fairly traditional way of tackling the problem, according to Mark Zelinka, a climate scientist and cloud expert at Lawrence Livermore National Laboratory, and co-author of both the May study and the study from last year.
A newer study, on the other hand, has taken a less conventional approach. Published last week in Proceedings of the National Academy of Sciences, the study used machine learning to figure out how clouds respond to changes in their environments.
Machine learning is a branch of artificial intelligence in which computers sift through large quantities of data, identify patterns and then use those patterns to construct algorithms that predict how future data should behave under various conditions. In this case, the researchers used real-world observations of the way clouds respond to environmental change.
The machine learning approach came to a similar conclusion: a narrower climate sensitivity, which rules out most of the milder climate scenarios. The study found that there’s almost no chance of a climate sensitivity below 2 C.
“I have thought for a while the cloud problem was particularly suited for machine learning approaches,” said Ceppi, who conducted the study with fellow climate scientist and machine learning expert Peer Nowack. “If you want to understand the relationship between clouds and temperature or humidity or winds, it’s quite hard to tease out the individual effects of each of these environmental variables.”
Machine learning can be a simpler way to tackle such a complicated set of data, he said.
Machine learning is showing promise in other kinds of cloud research as well. Some research groups are experimenting with incorporating machine learning components into global climate models as a way to work around the difficulties of simulating clouds.
Clouds pose a challenge for models because they require extremely fine-scale physics—after all, clouds form from tiny water droplets in the sky. Simulating these microscopic processes on a global scale would require an unimaginable level of computing power; it just isn’t possible.
To get around it, modelers don’t typically force their models to physically simulate the formation of clouds. Instead, they manually plug in information about how clouds should form and respond to changes in their environments, a tactic known as parameterization.
Machine learning can be an alternative to parameterization. Instead of plugging in a rule about how clouds should behave within the model, a machine learning component can construct algorithms that predict the way the clouds should respond.
It’s not exactly a common strategy yet. But multiple research groups in the last few years have begun investigating how useful it might be.
These are promising advancements in the complicated field of cloud research. Still, “machine learning is a super helpful tool but no panacea,” cautioned Piers Forster, director of the Priestley International Centre for Climate at the University of Leeds, in an email to E&E News.
Machine learning is an efficient way of analyzing complicated sets of data—but it can leave some questions unanswered about the underlying physical processes behind that data. There’s still plenty of room for more traditional research on the hows and whys of cloud behavior.
“Coordinated developments on both fronts are the answer in my mind,” Forster added.
In the meantime, Zelinka added, it’s reassuring that different strategies have arrived at similar conclusions.
“If it was just one study, you might question the robustness of that result,” Zelinka said. “But if you’ve got more and more evidence from independent authors using independent techniques, and they’re all reaching a similar conclusion, that’s pretty powerful.”
Reprinted from E&E News with permission from POLITICO, LLC. Copyright 2021. E&E News provides essential news for energy and environment professionals.