Computer processors across Amazon Web Services recently hummed into gear to create 30 simulations of what Earth could look like by the middle of this century. Normally, climate models run on supercomputers. But this effort on Amazon’s servers represents one of the first attempts to do modeling on the cloud and could revolutionize how modeling is done. Buying time on the cloud is vastly cheaper (though still not inconsequential in cost) than building and running a supercomputer. That alone makes it unique, but so, too, do some of the virtual worlds that scientists created.
If you were living in some of the 30 worlds created by researchers at the National Center for Atmospheric Research, you’d read reports of carbon dioxide climbing ever higher. The string of hottest years on record in the late 2010s would be average years by mid-century. Sea ice would fall to record lows and may even disappear completely some summers.
In a few of the worlds, though, something would be slightly off. The temperature is cooler, the sea ice around for longer. The sky might be slightly filmy, like someone smeared white into the normally crisp blue, only for the sunsets to explode in extraordinary color. This world is one where humans have decided to dim the Sun—just a bit—allowing less energy to reach Earth’s surface so that there’s less to be trapped here by rising carbon emissions.
The project, which is also being done in partnership with SilverLining, a nonprofit studying what scientists broadly refer to as solar radiation management, could offer a new way forward for how the climate is modeled. Doing computations on the cloud could open the door to making models more accessible to scientists, policymakers, and citizens alike.
Research into blocking the Sun is one of those very uncomfortable topics. The studies that have been done suggest it could pose dangers, including unleashing crop die-offs and essentially being impossible to stop once a solar dimming program begins. Yet our current, unintended experiment of loading the atmosphere with greenhouse gases is itself a risk, one that grows more dire with each passing year the world fails to wind down fossil fuel use. More research into dimming the Sun as a very, very last ditch option to turn down the heat could be vital.
There are real risks in not knowing what would happen in the model world where sunlight is dimmed. There’s real value in democratizing data access. Including a scenario on Amazon’s servers, though, also opens the door to some unsettling questions. Techniques to lower the amount of sunlight reaching Earth’s surface are comparatively cheap when you’re the world’s richest (or second richest, depending on the month) person.
Climate models are usually run on specialized supercomputers. Go through Top500, a list of the world’s 500 most powerful supercomputers, and you’ll find quite a few housed at various meteorological agencies and climate research facilities around the world. These computers link up tens of thousands of processor cores that can take on layer upon layer of complex equations to game out any number of problems, including the future of the planet.
What NCAR and SilverLining are doing with Amazon, though, is something completely different. Instead of relying on dedicated processors in one location, they’re relying on Amazon’s cloud computing architecture to handle those same calculations. That opens the door to a host of benefits. Researchers have outlined how cloud computing could allow for more datasets to be stored and integrated into models and give scientists more granular control over how much computing power they need. Setting up a supercomputer to run a model is a time-intensive process. Cloud computing could cut down on tweaks needed to run different models and study different variables—and allow scientists and organizations running supercomputers to spend more time on research and less on setup.
“Cloud computing has started to reach the point where it could contemplate supporting workloads like this,” Kelly Wanser, the executive director of SilverLining, said. “And so you have this inflection point where the underlying technology is sophisticated enough, and so the question is, can you break the adoption impasse and see if you can get this stuff running on the cloud? And then what happens if you do?”
Amazon Web Services is being used to model the period from 2035 to 2070 using NCAR’s s Community Earth System Model Version 2 and Whole Atmosphere Community Climate Model, both considered among the world’s top climate models. NCAR itself has run these models on its supercomputers, including Cheyenne, which clocks in among the top 100 fastest supercomputers on Top500's list.
Douglas MacMartin, a climate expert at Cornell who has studied solar radiation management, called the transition to the cloud a “non-trivial undertaking.”
“Climate models take a huge amount of CPU time, and they’re massively parallelizable,” MacMartin, who has worked with SilverLining, said. “They usually run on thousands of computer cores at the same time. Each one just divides up the atmosphere into all sorts of different little pieces. But the Amazon system, as it’s currently set up, doesn’t have quite as many cores available, and so that’s just been a little bit more effort. It is pretty specialized software. It does take some effort to figure out how to get it working.”
To see if there are any hiccups, the same models are being run by the UK Met Office (also home to a supercomputer in the top 100) in a closed environment. The results will also be compared to the NASA Goddard Institute for Space Studies’ model run outside the cloud.
If it works, then the project opens up the possibility for future efforts to model any number of climate variables on the cloud. In doing so, it also puts power traditionally reserved for a few research institutions in more people’s hands. Climate modeling, as with most areas of academia, has been dominated by white men in developed countries.
“The main driver [of this project] was that we were funding a group that supports Global South researchers, and they could not access the data sets,” said Wanser, who has worked in technology prior to taking up her current role. “And of course, they couldn’t use the models. And so if you’re sitting in Tonga or Bangladesh or somewhere with a weak network, that’s nowhere near what you need to work with these giant datasets or run the models. In theory, if you have them on the cloud, then you really open up the ability to work on these things to exponentially larger numbers of researchers, let alone people who aren’t researchers.”
The things that SilverLining is interested in aren’t just how hot it will get or how low Arctic sea ice will go. The group has helped fund research into solar radiation management, a form of geoengineering or, as the nonprofit and National Academy of Sciences prefer to label it, climate intervention. The idea behind blocking sunlight was verboten in climate circles for decades because of fears that it would limit the world’s appetite to end the fossil fuel use that is causing climate change in the first place. It turns out world leaders and major corporations haven’t had much of an appetite to do that regardless, even as emissions have skyrocketed. Now, the planet is at risk of warming past thresholds scientists and policymakers have dubbed relatively safe. Research into solar radiation management strategies has gained more of a toehold lately, given that risk.
We know that blocking out a bit of sunlight will cool the planet because volcanic eruptions have provided numerous natural, real-world examples. But those last only a few years before the effect dissipates. A sustained, decades-long program to mimic volcanic winter by injecting tiny, reflective particles in the stratosphere via planes or even high-altitude balloons, though, involves more unknowns.
Running climate models, on the cloud or otherwise, could help researchers better grasp what a future with climate intervention would hold and give policymakers more fodder to come up with an informed decision. (For what it’s worth, research so far hasn’t shown blocking a bit of sunlight is exactly a slam dunk.) And putting modeling on the cloud could open the door for more people to see what the data shows—and even invite new questions that modeling’s traditional barrier to entry would have left unasked. Jesse Reynolds, a geoengineering expert who runs Divided Sky Research and Consulting, said in an email that the effort was “laudable” for these reasons.
Still, it’s ironic to see this modeling done on Amazon’s cloud. The company itself neatly encapsulates the problem of the climate crisis, growing by leaps and bounds in the pursuit of profit and frying the planet in the process. The company and Jeff Bezos have moved to make amends recently by putting some of their voluminous fortunes into efforts to decarbonize. That includes the Amazon Sustainability Data Initiative, which provides grants and cloud computing time to various groups that include the current effort by SilverLining and NCAR.
“Our collaboration with SilverLining’s Safe Climate Research Initiative helps confront the urgent need for infrastructure to support and advance climate research,” an Amazon spokesperson wrote in an email. They went on to say that “hosting the resulting datasets in the AWS Open Data Sponsorship Program will open a powerful new avenue to accelerating climate research and democratizing access to tools and information that will help protect our planet.”
The grant program is a sliver of the company’s massive profits derived from essentially powering the internet through AWS and proudly leasing its cloud computing software to oil and gas companies. That’s not to say a sliver isn’t better than no sliver, but it does point to the murkier nature of Amazon’s commitment to opening up research into a radical solution to a problem it has a heavy hand in fueling.
“It’s going to be really, really hard for us to solve the climate crisis if we’re not working with the corporations that operate at scale in the system,” Wanser said. “And almost all of them are contributors, right now, to the problem.”
The grant to specifically model climate intervention on the cloud is also a bit of an eyebrow-raiser. Tech billionaires have become increasingly engaged with moonshot climate research. That includes Elon Musk’s $100 million prize for whoever can best capture carbon dioxide out of the air and Bezos’ spending $80 million from his eponymously named Earth Fund for “advanced technologies.” That money went to fellow tech billionaire Bill Gates’ Breakthrough Energy Foundation and Action. Gates himself has funded a group of Harvard researchers looking to perform very small-scale solar geoengineering research in the wild.
A full-scale program to deflect sunlight could cost under $10 billion per year. The wealth of each of these men would be more than enough to unilaterally decide to cool the planet for a few decades, if they determined it was a wise use of their time and resources. (It would also probably come with significant government pushback and sanctions.) That’s not to say that Gates or Bezos or another tech billionaire will pull a Mr. Burns and block the Sun so that they can keep part of their business model tied to fossil fuels, of course. (“Amazon’s role in this collaboration and research is not related to geoengineering,” the Amazon spokesperson said when asked about the project and the interest the ultra-wealthy have taken in the topic.)
Other, more democratic groups are also weighing similar avenues of research. The U.S. government itself is evaluating a program to study the impacts of dimming the Sun in more detail, including a massive report released earlier this year by the National Academies of Science on what such a program could look like. The House Select Committee on the Climate Crisis also included it as a small piece of Democrats’ climate plan in a roadmap it released last year.
But as Amazon’s success at cornering any number of markets and the current billionaire space race shows, there’s also an unmistakable streak among our tech overlords to find any edge. A climate intervention to cool the Earth a bit could allow the fossil fuel industry—valuable cloud computing clients of Microsoft, Google, and Amazon—to keep drilling. Or it could buy a bit of time for Bezos, Elon Musk, and Richard Branson as they push ever-larger and more polluting rockets into space.
We all need to know what could go wrong with intentional geoengineering before we get to a point where that’s even an option. Unfortunately, we’re nearly two centuries into a massive, unintentional climate experiment, which means what are currently niche research questions are becoming more and more relevant in the real world.
Correction, 12/1/21, 1:27 p.m. ET: This post mistakenly attributed a comment by Douglas MacMartin to another researcher. We regret the error.