DURHAM, N.C. – Using machine learning – a branch of artificial intelligence – to analyze and integrate thousands of datapoints collected from field observations, a Duke University-led research team has produced the most accurate estimates to date of gross primary production (GPP) in Earth’s oceans.
Marine GPP is a measure of how much carbon is fixed and oxygen is produced by phytoplankton as they use sunlight to convert carbon dioxide and water into food through photosynthesis.
The Duke-led work helps fill a gap in scientists’ understanding of how much photosynthesis and GPP is occurring at a planetary scale. It also sheds light on the metabolic demands of phytoplankton and other single-celled marine organisms and their role in the global carbon cycle – and how this all might be changing in response to Earth’s changing climate.
“Photosynthesis is one of the most important biological processes on Earth. It’s the main source of organic matter in ecosystems and modulates cycles of carbon, nutrients and water, thereby influencing the global climate. Being able to quantify how much photosynthesis is occurring, and if it is changing over time, is essential for understanding ecosystem functions and predicting feedbacks between the biosphere and climate,” said Nicolas Cassar, professor of biogeochemistry at Duke’s Nicholas School of the Environment.
Previous studies have shed light on the large-scale variability in marine net primary production (NPP) – which is the amount of carbon fixed through photosynthesis minus any carbon that gets released back into the environment through respiration. But marine GPP has thus far received less attention.
“Until now, we’ve known quite a lot about one of the two main fluxes relevant to marine photosynthesis, but fairly little about the other,” Cassar said. “Our study helps fill in the blanks and give us a more complete understanding.”
Having this knowledge is critical, he said, “because any change that happens at the bottom of the marine food web reverberates up and can have far-reaching impacts on species at higher trophic levels, including fisheries.”
The scientists published their peer-reviewed paper March 24 in the journal Global Biogeochemical Cycles.
To conduct their study, they compiled thousands of individual GPP datapoints from previous studies into two extensive datasets – one measured by light and dark bottle incubation, the other measured by in situ estimates based on the triple isotopes of dissolved oxygen in seawater samples. Both methods are well established techniques for estimating localized GPP.
Using machine-learning pattern recognition algorithms, the scientists trained satellites’ optical sensors to recognize and integrate the thousands of individual observations from each dataset, creating the first maps of marine GPP on a global scale.
Ideally, the two independent marine GPP estimates calculated from these maps should have been identical, Cassar said, but because both approaches have inherent uncertainties, including biases, the estimates initially differed by a factor of about 1.6. After accounting for some of the potential biases using first order approximations, however, the two models were found to converge and produce similar estimates. Both show that marine GPP is 1.5 to 2.2 times greater than marine NPP and comparable to the GPP occurring on land.
Armed with the new models, scientists will now be better able to track local or planetary shifts in primary production and more accurately predict future changes. The new estimates can also be used to validate other process-based Earth system models that integrate the interactions of the atmosphere, oceans, land, ice and biosphere to study and predict climate change.
In addition to his faculty post at Duke, Cassar holds a research appointment at the Laboratoire des Sciences de l'Environnement Marin (LEMAR) of the Institut Universitaire Européen de la Mer in Brest, France.
He conducted the new study with colleagues from Duke, Woods Hole Oceanographic Institution, and the State Key Laboratory of Marine Environmental Science and Fujian Provincial Key Laboratory of Coastal Ecology and Environmental Studies at Xiamen University in China.
Funding came from the “Laboratoire d’Excellence” LabexMER research program; the French government’s Investissements d’Avenir program; the National Key and Development Program of China; the Chinese National Science Foundation; the Chinese State Scholarship Fund; NASA; and the Woods Hole Oceanographic Institution.
CITATION: “Global Estimates of Marine Gross Primary Production Based on Machine Learning Upscaling of Field Observations,” Yibin Huang, David Nicholson, Bangqin Huang and Nicolas Cassar; March 24, 2021. Global Biogeochemical Cycles. DOI: https://doi.org/10.1029/2020GB006718