For years, UC Berkeley snow hydrologist Manuela Girotto has combined disparate remote sensing datasets from satellites into models to understand snow as a water resource. In an era of extreme drought and climate change, her work is increasingly urgent.
A recent rapid expansion of available observations from space could unlock important insights. But integrating that amount of data into researchers’ existing models is difficult. So when Berkeley computer science doctoral candidate Colorado Reed reached out asking how artificial intelligence could help, she saw an opportunity.
“It changes the value of what work I might do,” said Girotto, an assistant professor in Berkeley’s Department of Environmental Science, Policy, and Management, of working with artificial intelligence experts. “Having someone that can filter and understand all of the data streams and the various other solutions from these products, it benefits us. It has the potential to translate some of the nitty-gritty science that we do as researchers into more of a societally-applicable context.”
Girotto is now one of several scientists working with artificial intelligence researchers like Reed as part of the new Berkeley AI Research Climate Initiative. The hub, which Reed helps lead, aims to build partnerships and conduct groundbreaking artificial intelligence research in service of solving one of society’s most intractable problems: climate change.
“Climate researchers have very hard jobs. There's only so many of them, and there's a lot of climate out there to look at,” said Ritwik Gupta, an organizer of the initiative and computer science doctoral student. “We can take all of this really complicated data the climate researchers work with where they're limited by manpower and scale it up using AI. Our real goal is to have a lasting impact on the world.”
The climate initiative includes three organizers, eight faculty members and scientists and 13 researchers and students. The student-led hub is part of the Berkeley AI Research Lab’s BAIR Open Research Commons.
BAIR Climate Initiative
The initiative stemmed from the desire of the organizers – Reed, Gupta and computer science doctoral candidate Medhini Narasimhan – to help address climate change, an issue that President Joe Biden has called “one of the most pressing threats of our time.”
While artificial intelligence tools are used in areas such as climate research, often initial AI research focuses on meeting “benchmarks,” or goals, that aren’t intentionally designed to tackle real-world issues. That means AI progress can sometimes be a solution in search of a problem.
Climate change illustrates that issue. Less than 1% of publications presented at the three most prestigious AI conferences over the last 12 years mention ‘climate' at all, a review by the students shows.
“If climate change is the most pressing problem of our time and AI has a central role to play in it, then as we go and we look at the top AI institutes and industrial labs around the world, why aren’t we working on it?” Reed asked in a May presentation at Berkeley on the initiative.
“If climate change is the most pressing problem of our time and AI has a central role to play in it, then as we go and we look at the top AI institutes and industrial labs around the world, why aren’t we working on it?”
To ensure that their AI research is solving relevant problems, the initiative looks for partners among climate experts, government agencies and industry. Already, it has a following that is growing rapidly: researchers at several UC campuses and computing industry partners are collaborating with the initiative. Together, they identify a project and relevant goals that are meaningful for climate and AI and work to meet them.
Any technology or research done by the climate initiative will be openly published and won’t be exclusively licensed or proprietary, said Trevor Darrell, BAIR’s co-founding director and a Berkeley computer science professor. It is one of several climate-related computing and data science initiatives starting on campus.
“It's a very significant AI problem at a scale that wasn't addressed in previous challenges that drove the internet-AI era like building search engines and perception systems that were looking at images on the web,” Darrell, an advisor for the initiative, said of climate change. “Here, we actually need to have systems that function in the real world.”
‘The Fate of Snow’
One of the first projects the initiative is tackling is one with Girotto, now an advisor for the group. “The Fate of Snow” – which is a partnership between the climate initiative, Girotto and others at Berkeley, Lawrence Berkeley National Laboratory (Berkeley Lab), Meta AI and the Center for Western Weather and Water Extremes – illustrates the climate initiative’s approach.
Researchers are using AI techniques to combine measurements from aircraft observations of snow and a multitude of openly available weather and satellite data sources. These data track snow from each falling snowflake to how they aggregate in the snowpack to how they melt out in streams, but since each data source is different new methods are needed to gather all the information in the data together to provide a complete view of the lifecycle of snow. They’ll use AI methods to estimate and predict how much water is in the snow in the Sierra Nevada mountains and forecast what that will mean for streamflow in the region.
The state of snow impacts public health and the economy. Approximately 1.2 billion people rely on snow melt globally for water consumption or other purposes, Reed said in his presentation. The Sierra mountains alone underpin $50 billion in agricultural businesses and provide the water that over half of California’s population relies on, he said.
“Addressing climate change is one of the most important challenges facing humanity today,” said Joe Spisak, the product lead for Meta AI Research, a sponsor for the project. “The Fate of Snow project is exactly the type of collaboration that our researchers believe has the potential to bring real impact to the fight against climate change, both in the short and long term.”
“The Fate of Snow project is exactly the type of collaboration that our researchers believe has the potential to bring real impact to the fight against climate change, both in the short and long term.”
Not knowing the amount of snow in this area is “a persistent problem,” due to a lack of measurement capabilities in parts of the mountains, said Daniel Feldman, a staff scientist at Berkeley Lab. AI can help, he said.
While it focuses on the Sierra Nevada mountains, similar methods and approaches could be used to understand the snowpack in other areas across the globe, too, Feldman said.
“Between 60 and 90 percent of the world's population relies in whole or in part on water that's derived from mountains. So, we really need to have a good sense of how much water is available for ecosystems, for society – especially for vulnerable populations,” said Feldman. Referring to the overall climate initiative, he added, “This is a real significant effort and one that I'm just delighted to be a part of.”