Using historical ship measurements and data from Argo floats, researchers have developed a machine learning method that enhances the evaluation and comprehension of the declining oxygen levels in the ocean.
Oxygen is vital for all living organisms, especially multicellular beings, as it assists in metabolizing organic matter and powering life processes. Approximately half of the oxygen we inhale originates from land-based plants, like forests and grasslands, while the other half is generated by marine algae through photosynthesis in the ocean’s surface waters.
Oxygen levels are diminishing in various regions of the world’s oceans. Experts suspect that this decrease is related to the warming of ocean surfaces and its effects on seawater’s physics and chemistry, although the issue remains not entirely understood. Temperature is a key factor influencing how oxygen dissolves in seawater; as the water heats up, its capacity to retain gas decreases.
“Determining the amount of oxygen lost from the oceans is complex due to a scarcity of historical data and irregular timing,” stated Taka Ito, an oceanographer and professor at Georgia Tech’s School of Earth and Atmospheric Sciences. “To grasp global oxygen levels and their fluctuations, we must address numerous data gaps.”
A group of student researchers aimed to tackle this problem. Under Ito’s leadership, they created a new machine learning method to better analyze and illustrate the decrease in global ocean oxygen levels. The team utilized datasets to produce a monthly map displaying the oxygen content trends over several decades. Their findings were published in the Journal of Geophysical Research: Machine Learning and Computation.
“Marine scientists need to comprehend the distribution of oxygen in the ocean, the extent of its changes, where these changes occur, and the reasons behind them,” explained Ahron Cervania, a Ph.D. student in Ito’s lab. “Traditionally, statistical approaches have been used for these estimates; however, machine learning techniques can enhance the precision and detail of our oxygen assessments.”
The project initiated three years ago with backing from the National Science Foundation, where the team initially concentrated on data from the Atlantic Ocean to refine their method. They employed a computational model to create hypothetical observations, allowing them to evaluate the effectiveness of reconstructing missing oxygen level data based solely on a subset of information combined with machine learning. After refining this method, they expanded their focus to include global ocean observations, engaging undergraduates and assigning tasks across different ocean regions.
Guided by Ito, Cervania and fellow student researchers crafted algorithms to examine the connections between oxygen levels and factors such as temperature, salinity, and pressure. They utilized a dataset of ship-based oxygen observations dating back to the 1960s alongside recent data from Argo floats—autonomous drifting devices that gather temperature and salinity measurements. Although oxygen data exists from before the 1960s, earlier records suffer from accuracy issues, so the team concentrated on records post-1960. They then generated a global monthly map of ocean oxygen content spanning from 1965 to the present.
“By using a machine learning approach, we were able to evaluate the rate of oxygen loss more accurately across different periods and locations,” Cervania remarked. “Our results reveal that incorporating float data significantly improves the estimate of oxygen loss while minimizing uncertainty.”
The researchers discovered that the world’s oceans have been losing oxygen at a rate of approximately 0.7% per decade from 1970 to 2010. This figure indicates a relatively swift response of the oceans to recent climate changes, which could have long-lasting effects on the health and sustainability of marine ecosystems. Their estimates also align with other studies, affirming the validity and effectiveness of their method.
“We analyzed trends in global oxygen levels and the ocean’s overall inventory, focusing on the rate of change over the last fifty years,” Cervania said. “It’s reassuring that our calculated rate is consistent with earlier estimates from other methodologies, which boosts our confidence. We are strengthening our findings alongside other research.”
According to Ito, the team’s innovative method addresses a persistent challenge within the oceanographic field: effectively combining diverse data sources that vary in accuracy and certainty to gain a better understanding of ocean changes.
“Utilizing advanced technologies like machine learning will be crucial in bridging data gaps and offering a clearer insight into how our oceans are adapting to climate change.”
