For the people of the Maldives, a string of islands off the southern tip of India, the realities of climate change lie right outside their front door. A 2007 report from the United Nation’s Intergovernmental Panel on Climate Change predicted that unfettered carbon emissions could push sea level rise to 23 inches by 2100. With an average elevation of less than five feet, even a slight increase in sea level could make these islands inhabitable. The teal blue sea is swallowing them up.
But not everyone experiences the ravages of climate change so overtly. Climate change denialism persists despite the overwhelming scientific consensus that it’s real, and even those who recognize the reality often seem caught in the mental trap of thinking that the most serious impacts will be felt some place far away.
To remedy this, a team of researchers at the Mila Quebec Artificial Intelligence Institute wants to bring the not-so-distant climate crisis reality closer to home. They’re creating an A.I.-powered platform that shows users how climate change-related natural disasters may impact their homes. The goal is to develop more intimate understandings of how climate change will upend communities, while empowering people to make more informed decisions about whether they fuel or fight the impending crisis.
“It’s hard for people to relate to climate change when we only mention remote areas and polar bears,” said Victor Schmidt, a PhD Candidate at Mila and lead author of a paper from May that outlined the team’s approach. “But there are so many consequences of climate change. It’s going to impact everyone. We want to help people better understand that and help them engage in actually taking action.”
Mila’s Visualizing Climate Change platform is designed to show people what the future has in store if we don’t significantly cut emissions soon. Similar to the National Oceanic and Atmospheric Administration’s Sea Level Rise Viewer, which offers an aerial perspective on the how far the ocean will creep up or shores, Visualizing Climate Change will let users input street addresses and see what things will look like in the aftermath of a natural disaster. The platform will focus on flooding to begin with, before tackling more difficult to depict, climate-change related events, such as wildfires.
“We feel that showing people the potential consequences of climate change in their neighborhoods is a good way of making climate change more personal and less distant,” Schmidt said.
The Mila team uses an image-to-image translation algorithm to transform photos captured from Google Street View into ones depicting the aftermath of flooding. They use a generative adversarial network (GAN) to train the system. GANs work by pitting two algorithms against each other—one algorithm generates an image and the other tries to guess whether that image is real or fake. In this way, the first algorithm creates more realistic images as the second challenges it to perform better.
“Showing people the potential consequences of climate change in their neighborhoods is a good way of making climate change more personal and less distant.”
One of the biggest challenges holding the Mila team back is its lack of images from the aftermath of extreme weather events, which it uses to train its algorithm. The group launched the ClimatePix application in August to gather pictures from the public.
“We need images of houses in populated areas that have gone through floods,” he said. “It’s easy enough to get images houses without floods.”
Schmidt and his colleagues don’t claim to be climate scientists and their platform is not meant to be scientifically precise. Rather, they see their role as communicators, helping people interpret the predictions of the latest climate science. And the Mila team want to provide more than just a wake-up call. They hope to integrate resources to guide users on ways to address the climate issue.
Efforts are underway elsewhere to use A.I. to tackle climate change issues more directly. Earlier this month, a consortium of science institutions including the European Space Agency put out a call for proposals for a €500,000 (about $550,000) A.I. Moonshot Challenge. The challenge hopes to fund projects that use machine learning and space technologies to combat climate change.
Meanwhile, David Rolnick, a postdoctoral research fellow at the University of Pennsylvania, is leading Climate Change A.I., a group he founded to support the use of machine learning in addressing the climate crisis.
“The tools that work on in machine learning can have a huge impact when applied to the problem of climate change,” he said.
In June, Rolnick and his colleagues published a paper called Tackling Climate Change with Machine Learning, which presented various ways in which A.I. can be used to aid in climate change mitigation, resiliency, and adaptation. Algorithms can help provide data about flooding to city planners, for example, aid in the development of more efficient batteries, or help optimize transportation networks.
“Machine learning is not a silver bullet. It is not suddenly going to come in and solve any of these problems.”
However, Rolnick stressed, “machine learning is not a silver bullet. It is not suddenly going to come in and solve any of these problems. There are many aspects of action on climate change to which machine learning is holy inapplicable.” The people behind Climate Change A.I. insist that the technology be considered one piece to the complicated puzzle of addressing climate impacts.
These initiatives come at a critical time. As California burns, Venice floods. The canaled city is no stranger to ocean encroachment but tides high enough to flood Venetian squares now happen more often than ever, as sea level rises and warmer oceans amplify storms. Five of the ten highest Venetian tides have occurred in the last twenty years. If trends like this continue, the public may not need an algorithm to show them what climate change looks like—they can simply look out their front door.
