A pair of projects that are leveraging Yale research to address the global climate threat — a game-changing forecasting system for marine carbon dioxide removal strategies and a new approach to reducing methane emissions from livestock systems — are among the winners of a worldwide competition for climate change solutions that harness the power of AI.
The two research teams will each receive $2 million over the next two years from the Bezos Earth Fund’s AI Grand Challenge for Climate and Nature, in addition to a $50,000 development grant each team won earlier this year. The challenge selected 15 winners on Oct. 23.
More than 1,000 organizations entered the competition, which began last year. Of those entrants, 24 reached Phase I status in May. The projects ranged from new ways to track and protect endangered species to strategies that optimize the power grid for renewable energy integration.
The Yale-affiliated winners are a research group in the Yale Center for Natural Carbon Capture (YCNCC) and the Department of Earth & Planetary Sciences in Yale’s Faculty of Arts and Sciences (FAS) that is building an AI-enabled model to predict the carbon removal potential of ocean interventions, and a partnership between the Alliance of Biodiversity International and the International Center for Tropical Agriculture (CIAT) and Purushottam Dixit of the Yale School of Engineering & Applied Science that is developing a “Rumen Digital Twin” that uses AI to reduce livestock methane emissions.
Purushottam Dixit
Each project “pushes the boundaries of knowledge and drives impact far beyond campus” and exemplifies the kind of “bold, interdisciplinary” research that Yale Planetary Solutions (YPS) — a university-wide initiative that works to advance Yale’s commitment to generating innovative ideas in support of a thriving planet — seeks to catalyze, said Julie Zimmerman, Yale’s provost for planetary solutions.
“This support from the Bezos Earth Fund is critical to advancing our understanding of both marine carbon removal and livestock methane emissions and their potential for climate mitigation,” Zimmerman said.
More information about the two projects follows:
The mCDR forecasting ‘Stack’
The marine carbon doxide removal (mCDR) forecasting “Stack” — which is led by Elizabeth Yankovsky, assistant professor of Earth and planetary sciences, and YCNCC associate research scientist Luke Gloege — adds a machine learning component to simulations underpinning monitoring, reporting, and verification (MRV), a process used in tracking climate change mitigation.
Specifically, the researchers say their innovation will support new technologies that remove carbon dioxide from the atmosphere and store it stably in the ocean as dissolved bicarbonate. Such work is critical to climate mitigation, but tough to quantify in terms of efficiency and durability. Noah Planavsky, a professor of Earth & planetary sciences in FAS, helped develop the Stack proposal and will be part of its implementation.
“The Stack is designed to be fast, intuitive, and scientifically robust to bring advanced forecasting within reach for a broad range of potential users,” Yankovsky said. “It unifies and builds onto existing AI and physics models from various disciplines to tackle the challenges of MRV for carbon dioxide removal.”
The world’s oceans have an effectively infinite capacity to store bicarbonate, the researchers say. But inefficiencies and turbulence in the process make it difficult to scale up geochemical interventions for climate change mitigation.
For example, small-scale turbulence can move alkalinity — added to the ocean surface through an intervention — away from the ocean surface before it can remove atmospheric carbon dioxide. Existing models for tracking these carbon fluxes are either not precise enough or prohibitively slow to run.
The Stack will combine a new, graphics processing unit (GPU)-optimized ocean model and a biogeochemistry model, tailoring them specifically for simulating carbon dioxide removal interventions. It will also utilize AI-derived atmospheric forecasts using tools developed by NVIDIA, a technology company that designs and manufacturers GPUs for numerous uses, including at data centers.
Stack users will be able to have a day-by-day forecast of net carbon dioxide removed, as well as the durability of that removal over a user-specified amount of time.
The Stack will be able to provide decadal-scale forecasts in hours and millennia-scale forecasts in days — something that has never been achieved previously.
“We’ll be able to do granular forecasting of atmospheric carbon dioxide uptake and storage to enable high-fidelity MRV, better planning of specific interventions, site selection, and a basis for durable carbon crediting,” Gloege said.
The Rumen Digital Twin
This project is a collaboration between the Alliance of Biodiversity International and the International Center for Tropical Agriculture (CIAT), BiomEdit (an Indiana-based company), and Dixit, an assistant professor of biomedical engineering at Yale Engineering.
Karthik Srinivasan, left, and Purushottam Dixit are part of a research team designing a new approach to reducing methane emissions from livestock systems.
Methane emissions from livestock are a significant component of climate change, and new approaches to the mitigation of enteric methane — a byproduct of rumen microorganisms in the gut biome of livestock animals — are being tested around the world. Yet much remains unknown about the complex interactions between microbial metabolism, diet, and ruminant genetics.
The “Rumen Digital Twin” is being developed to capture this interplay in unprecedented fashion. Using publicly available data, it will allow users to create virtual livestock groups with precise detail in order to identify characteristics and conditions in which specific feed additives or ingredients will be most effective at reducing methane.
“The microbiome of livestock is incredibly complicated,” Dixit said. “What may work to reduce methane emissions in a small-scale experiment often does not generalize to a different farm on a different continent.”
The Digital Twin’s AI-driven model will allow users to test their sustainable livestock plans quickly and less expensively, according to Dixit and his collaborators. As the project is being developed, Dixit said, the AI component will be produced at Yale.
The project’s model and fine-tuning protocols will be shared with communities and organizations seeking to implement similar strategies or build their own approaches using the Digital Twin as a foundation model. The researchers also plan to build a web-based dashboard for creating and analyzing virtual cohorts of livestock.
“This award will allow us to focus on building our foundational framework,” he added. “My colleagues and I are quite happy and excited to continue our work.”
