Low-Level Cloud Loss Amplifies Global Warming
Low-level clouds over the earth's oceans play a prominent role in keeping our planet cool by reflecting sunlight away from the surface. But their response to climate change has been hard to model. Now, researchers at Caltech and Google have uncovered important new insights about how clouds might respond to warming sea-surface temperatures and rising CO2 levels using a large and powerful dataset of simulations developed by the group.
"One of the largest open questions in climate prediction is how low clouds will respond to global warming," says Zhaoyi Shen, lead research scientist at Caltech's Ronald and Maxine Linde Center for Global Environmental Science and a co-author on a paper outlining the team's findings published July 24 in Science Advances. "Our results show potentially large rapid adjustments of low clouds to high CO2 concentrations, which suggests the earth's climate might be more sensitive to high CO2 levels than some climate models currently project."
The team also found that the thinning of low clouds—uniform layers or large, lumpy expanses below 6,000 feet that cover massive portions of subtropical seas—amplify global warming through a feedback loop: Rising sea surface temperatures lead to fewer clouds, meaning less reflected sunlight and a warmer planet.
"This supports the growing body of evidence from the last few years," says Tapio Schneider, the Theodore Y. Wu Professor of Environmental Science and Engineering at Caltech and co-author of the paper; Schneider is also a principal scientist at Google. "We can now confidently rule out the idea that this effect is zero or that it somehow dampens global warming."
While virtually all global climate models show that the earth is getting warmer, they differ widely on predictions of the exact long-term temperature rise triggered by sustained increases in atmospheric CO2. A large part of the challenge in reaching a scientific consensus has been the inability to resolve the fine-scale atmospheric turbulence that drives low-level cloud formation and dissipation.
By combining a modeling framework for simulations developed by Shen with the power of Google computing resources, the research team used simulated large-scale weather data from a trusted global climate model developed by the National Oceanic and Atmospheric Administration to drive thousands of high-resolution large-eddy simulations.
The simulations used atmospheric and surface-level conditions from 500 randomly selected locations across the tropical Pacific Ocean, taken during four different months to represent seasonal changes. Each location–season combination was then used to drive large-eddy simulations for four climate change scenarios: a 4 degrees Celsius sea-surface temperature increase from baseline; a quadrupling of atmospheric CO2 alone; a 4 degrees Celsius warming with doubled CO2; and a 4 degrees Celsius warming with quadrupled CO2.
"We found that clouds respond directly to CO2 changes, a fact well understood in physics but perhaps a surprise to many," Schneider says. "Simply altering atmospheric CO2 shifts how infrared radiation moves through the air, directly affecting clouds even if temperatures are artificially held steady. Crucially, this effect is nonlinear; it accelerates as CO2 levels rise."
The team used more than 7,000 simulations in their investigation, which represents a massive increase in sample size compared to earlier work. For example, an earlier study by Shen combined data from 500 simulations, which was itself an increase from the dozens of simulations in previous experiments.
The massive upscaling was made computationally feasible using a large-eddy simulation code that leverages Google's tensor processing unit (TPU) clusters to build a robust collection of cloud states under different conditions. TPUs are specialized computer chips designed to accelerate artificial intelligence (AI) and machine learning workloads.
"This experiment demonstrates the scale of computation made possible by Google's hardware capabilities," says Yi-Fan Chen, a software engineering director at Google and lead of the Google team that carried out the research. "It's computing at an extraordinary scale, demonstrating how processors originally developed for AI and machine learning can accelerate scientific discovery, enabling simulations that were previously out of reach."
In addition to modeling the future, the dataset can help researchers elucidate the planet's past climates, such as during the Eocene Epoch (which began 56 million years ago and ended 33.9 million years ago), when CO2 levels were up to four times higher than today. Plus, the dataset is public, meaning anyone with a laptop can use it to build their own models.
"This new public dataset has the potential to help the broader scientific community evaluate and train turbulence, convection, and cloud models for global climate models," says Shen. Shen is also part of the Climate Modeling Alliance (CliMA), a Caltech-based coalition of scientists, engineers, and applied mathematicians from Caltech and MIT. "At the Climate Modeling Alliance, we are currently developing a new climate model designed to learn from data using AI and machine learning, and I am using this dataset to calibrate our model's new turbulence and convection schemes."
Schneider, who leads CliMA, says the group is already using the data extensively and that he is excited to see what others can glean from it.
"I hope researchers come up with new and creative ways of representing clouds in climate models and that we finally reduce these massive uncertainties in climate predictions," he says. "It's not going to happen overnight to change all sorts of climate models around the world, but over time, I think this can happen."
The Science Advances paper is titled "High-resolution simulations reveal positive global warming feedback from Pacific low clouds." Additional authors from Google are Sheide Chammas, Qing Wang, Rob Carver, Jeffrey B. Parker, Cenk Gazen, Matthias Ihme, Yi-Fan Chen, and John Anderson. The work was supported by the National Science Foundation and Schmidt Sciences.
A 3D rendering of a large-eddy simulation produced by the research team shows stratocumulus clouds over the Pacific Ocean. The bottom plane renders surface buoyancy.
Credit: Sheide Chammas / Google
A 3D rendering of a large-eddy simulation produced by the research team shows coastal stratocumulus clouds with fog over the Pacific Ocean. The bottom plane renders surface buoyancy.
Credit: Sheide Chammas / Google
Tapio Schneider
Zhaoyi Shen
