New AI approach sharpens picture of carbon export in the Southern Ocean

Applying machine learning algorithms to a trove of data collected by robotic floats has improved estimates of ocean productivity and carbon export in the Southern Ocean.

Why It Matters

The Southern Ocean plays an important role in global climate and carbon cycling. Understanding carbon export in this region is critical for modeling Earth’s changing climate and evaluating potential ocean-based climate interventions.

In the vast expanse of the Southern Ocean, invisible highways of carbon flow from the surface to the ocean floor. This “carbon superhighway” is a key part of Earth’s climate system, moving carbon dioxide from the atmosphere into the ocean’s depths, where it can be stored for decades or centuries. Measuring the speed and capacity of that highway—and how it is changing—has long challenged scientists, especially in the planet’s most remote waters.

A new study led by MBARI Postdoctoral Fellow Guillaume Liniger, in collaboration with the University of Washington Cooperative Institute for Climate, Ocean, and Ecosystem Studies (CICOES) and the Southern Ocean Carbon and Climate Observations and Modeling (SOCCOM) project, sheds new light on ocean productivity and carbon export in the Southern Ocean. Combining a massive trove of data from robotic floats with cutting-edge machine learning, the team found the ocean’s ability to store atmospheric carbon is even greater—and growing faster—than previously thought.

Published this month in Global Biogeochemical Cycles, the research shows that annual net community production (ANCP)—a measure of how much carbon is converted into organic matter and exported to depth—increased by nearly one percent per year between 2004 and 2022. On average, the Southern Ocean exports 3.91 billion metric tons of carbon each year. The research also suggests that traditional methods for estimating ANCP from seasonal nitrate changes may underestimate true carbon export by roughly one-third.

“The ocean plays an integral role in Earth’s climate. This work is an important step forward in understanding productivity and carbon cycling in the Southern Ocean that can help improve models of our changing climate,” said Liniger.

Unlocking the secrets of a remote ocean

Encircling Antarctica, the Southern Ocean plays an outsized role in global climate. It covers only about a third of the global ocean’s surface area but accounts for a much larger share of oceanic carbon uptake. Powerful winds and currents mix deep, nutrient-rich waters to the surface, fueling blooms of microscopic phytoplankton. These tiny plants use sunlight and nitrate to grow, pulling in carbon dioxide from the atmosphere in the process.

As phytoplankton die or are consumed, a portion of their carbon-rich material sinks, carrying atmospheric carbon to the deep ocean. This biological carbon pump is a critical climate regulator. But harsh weather, sea ice, and sheer remoteness make it one of the most difficult systems to observe directly.

A network beneath the waves

An illustration shows the cycle of a yellow robotic float collecting profiling data in the water column. Text on the left reads Drift at 1,000 meters for 9 days under a white dashed horizontal line. The next text caption to the right reads Descend to 2,000 meters under a white dashed diagonal line connected to a yellow robotic float. The next text caption reads Ascent to surface while collecting data under a dashed white diagonal line connected to yellow robotic floats. The next text caption reads Relay data to shore via satellite with a yellow robotic float and red waves connecting to a gray satellite. The final text caption on the right reads Descend to 1,000 meters under a white dashed diagonal line connected to a yellow robotic float. The background is a blue gradient from blue sky and ocean surface to turquoise to dark blue water. — Over the course of 10 days, a robotic BGC-Argo float moves up and down the water column, collecting ocean conditions across vast regions and then transmitting that data to scientists. Image: Kim Fulton-Bennett © 2020 MBARI

SOCCOM, an international collaboration hosted by Scripps Institution of Oceanography to study the Southern Ocean, is part of the international Biogeochemical-Argo (BGC-Argo) program that deploys autonomous profiling floats equipped with sensors that measure temperature, salinity, oxygen, nitrate, pH, and other key variables. Since 2014, SOCCOM has released more than 300 floats into the Southern Ocean. These floats surface every 10 days to transmit their data via satellite.

MBARI has been instrumental in developing and refining the chemical sensors that make these measurements possible. “These floats are our eyes and ears in a part of the ocean where shipboard measurements are sparse,” said MBARI Senior Scientist Ken Johnson, a co-author on the new study. “They give us year-round, basin-wide coverage that was unimaginable a decade ago.”

Teaching machines to see patterns

A map shows the data collected from robotic floats in the Southern Ocean from 2012 to 2022. The circular map has a solid black illustration of Antarctica at the center, surrounded by several squiggly lines of varying shades of blue. Around the edge of the map are South America, Africa, and Australia, as well as longitude markers. On the right is a legend with a color gradient of pale yellow to pale green to turquoise to blue to dark blue and markers for the years 2012, 2014, 2016, 2018, 2020, and 2022 with a label that reads Time (years). — Machine learning can help researchers analyze patterns in oceanographic data. Training a neural network on data from BGC-Argo floats allowed the team to more accurately predict nitrate levels and carbon export year-round. Image courtesy of Guillaume Liniger

The new research hinged on an ambitious step: teaching an artificial neural network to recognize patterns in nitrate data collected by the floats. Nitrate is a vital nutrient for phytoplankton growth and a reliable way to measure ANCP. By constraining the neural network with the dense spatial and temporal coverage from BGC-Argo, and correcting for physical and sampling biases using the Biogeochemical Southern Ocean State Estimate model, the research team generated consistent nitrate estimates across the entire Southern Ocean and throughout the year.

“We wanted a model that could fill in the gaps of both space and time and approximate the physical and biogeochemical conditions of the Southern Ocean as realistically as possible,” explained Liniger. “Machine learning gave us a way to do that.”

A clearer, and bigger, number

With these improved nitrate fields, the researchers recalculated ANCP and found that the Southern Ocean’s carbon export is not only substantial, but also increasing. The results aligned with satellite observations showing rising surface chlorophyll concentrations and with model outputs predicting higher export fluxes.

Equally important, the analysis revealed a blind spot in common approaches. Estimates based solely on the seasonal drawdown of nitrate during spring and summer missed significant export happening outside that window, leading to undercounts of about 38 percent. This finding has implications for how scientists calculate global carbon budgets and how they validate climate models.

Global data, shared impact

Like MBARI, SOCCOM makes all of its data freely available, supporting research on ocean chemistry and climate worldwide. This openness is by design: the Southern Ocean is a shared global resource, and understanding its role in climate regulation requires international collaboration.

“This work shows the power of combining open, high-quality data with innovative analysis,” said Johnson. “It’s not just about producing a better number, it’s about providing the tools policymakers and resource managers need to make informed decisions about the future of the ocean—and the entire planet.”

Looking ahead

A satellite photo shows a bloom of plankton in the Southern Ocean. The spiraling eddies of green plankton swirl in the blue ocean with white clouds at the top and bottom. — Blooms of microscopic phytoplankton kickstart the biological carbon pump, which exports carbon to the deep sea. Measuring carbon transport in the Southern Ocean is crucial for modeling Earth’s changing climate. Image courtesy of NASA

As climate change accelerates, keeping a close eye on the Southern Ocean’s carbon dynamics is crucial. This machine learning approach offers an improved way to monitor long-term trends and detect emerging changes in this critical system.

As part of MBARI’s ongoing work to understand the ocean-climate connection, we envision applying similar methods to other regions and expanding the range of variables predicted by neural networks. Using innovative technologies from state-of-the-art research vessels to advanced AI, our researchers will continue to explore the integral role of the ocean in Earth’s carbon cycle and climate. As we make the invisible flows of the carbon superhighway more visible, we hope to increase understanding of the climate system we all depend on and gain new insights to help guide decision-making about climate change and evaluate ocean-based climate interventions.

This research also demonstrates MBARI’s commitment to leveraging our advanced technology and engineering expertise to answer fundamental questions about polar environments and other components of the cryosphere. This information can help resource managers and policymakers make decisions about the future of this important ecosystem.

This work was funded by the U.S. National Science Foundation’s SOCCOM project, with additional support from NOAA, NASA, and the David and Lucile Packard Foundation.

Research Publication:

Liniger, G., J.D. Sharp, Y. Takeshita, and K.S. Johnson. 2025. Two decades of increase in Southern Ocean net community production revealed by BGC-Argo floats. Global Biogeochemical Cycles, 39(8), e2024GB008371. https://doi.org/10.1029/2024GB008371

For additional information or images relating to this article, please email pressroom@mbari.org.

You might be interested in: