Producing reliable climate models is a difficult job and with the malicious scrutiny of those who want to derail the idea of human-driven climate change, minimizing uncertainties is crucial. Researchers might now have a new weapon in their arsenal: machine learning.
Machine learning is a special type of algorithm that can perform tasks without being programmed to do so explicitly. The artificial intelligence (AI) can learn, and this ability allows the algorithm to do things normal software can’t do. In this case, it can improve the resolution of climate models. The approach is described in Geophysical Research Letters.
Current models have a resolution of roughly 100 kilometers (62 miles), good enough for the atmosphere as a whole but too coarse for the more subtle features of clouds. But machine learning can help, improving such models without having to use more complex software (which requires longer computation and/or more powerful computers). This approach could give narrower ranges of predictions when it comes to the consequences of climate change.
"This could be a real game-changer for climate prediction," lead author Professor Pierre Gentine, from Columbia University's School of Engineering and Applied Science, said in a statement. "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."
They called their algorithm Cloud Brain or CBRAIN. As a proof of concept, they began by testing it on planets where all the parameters were controlled by the researchers. Each planet was either an aquaplanet or a planet with continents. They trained the deep neural network to understand how the clouds would warm up, moisten, and radiate heat away. CBRAIN was then able to predict many other features.
"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," added Gentine.
All the components of the atmosphere influence each other in subtle or less subtle ways. This approach could finally help us break through the current limits of climate models.