And the carbon problem in research is hardly limited to computer science.
Large astronomical observatories and space-based telescopes are big emitters. One study, published earlier this year in the journal Nature Astronomy, found that over the course of their lifetimes the world’s leading astronomical observatories will produce about 20 million metric tons of carbon dioxide equivalent (CO2e). In a news conference announcing their results, the authors said that if the world is to meet the challenge of net-zero greenhouse gas emissions by 2050, astronomers will have to reduce the carbon footprint of their research facilities by up to a factor of 20. That might mean building fewer big observatories. When these researchers analyzed their own facility, the Institute for Research in Astrophysics and Planetology (IRAP) in Toulouse, France, they found the average greenhouse gas emissions per person were 28 metric tons of CO2e a year, compared with 4.24 metric tons per person for the average French citizen.
Other scientists have focused on the carbon footprint of research conferences. One of climate science’s most important gatherings is the annual meeting of the American Geophysical Union (AGU), usually held in San Francisco. Climate modeler Milan Klöwer and his colleagues calculated the travel-related carbon footprint of the 2019 AGU meeting at 80,000 metric tons of carbon dioxide—about three metric tons per attending scientist. That per person output was almost as much as the annual output of an average person living in Mexico. Klöwer offered footprint-reducing ideas: moving the meeting to a central U.S. city to shorten travel, holding the conference biennially and encouraging virtual participation. Taken together, these changes could reduce the travel footprint by more than 90 percent. The AGU has said it plans to rotate locations in the future and use a hybrid meeting format.
But as the analyses of astronomy and computer science show, it’s the research, not just travel, that enlarges the scientific carbon footprint. Emma Strubell, a computer scientist at Carnegie Mellon University, and her colleagues concluded—in a study that has not yet been peer-reviewed—that from a carbon budget standpoint, the extreme amount of energy spent training a neural network “might better be allocated to heating a family’s home.” Similar complaints have been raised about bioinformatics, language modeling and physics.
This is a hard reality to face. But as time runs out to prevent a climate calamity, scientists will have to find a way to do more of their work with much less of our energy.