Article body copy
Determining the borders of the ocean’s ecological regions is challenging. On land, different ecoregions such as rainforest or tundra can be classified by the species of animals and plants and their abundances, but in the ocean, most species are microscopic, and their movements mean boundaries are ever-changing.
Typically, scientists studying the distribution of life in the sea use satellite images to measure a region’s chlorophyll levels—a chemical compound made by photosynthesizing phytoplankton—to get an idea of how much life is in an area. But these measurements don’t differentiate between species of phytoplankton, some of which support specific combinations of animal and plant life.
New research led by Maike Sonnewald, a physical oceanographer at Princeton University in New Jersey, outlines a new way to classify marine ecosystems. She says that the ocean can be broken down into 100 different ecoprovinces, which together make up 12 main megaprovinces with similar balances of animal and plant species.
These megaprovinces are all distinctive, with most being shaped by landmasses such as trenches or continents, or by oceanographic processes, such as where upwelling brings cold water to the surface. One megaprovince, which the scientists simply called H, spans most of the equatorial area of the Indian Ocean and has a mix of phytoplankton that allows it to support a rich assemblage of life. The K megaprovince is found only in the high Arctic Ocean and supports fewer species, but the physically larger phytoplankton that reside there make the total biomass about the same as in H. A chlorophyll-based detection method would make these two areas look more similar than they actually are.
Sonnewald says that having a deeper understanding of the different ecoprovinces, which were calculated through a machine learning approach that parsed huge sets of ocean data, including information on 51 phytoplankton species, could allow oceanographers to better measure marine health. Recognizing the different zones might also make it easier for scientists to track changes in species abundance or diversity, which would help with understanding the effects of climate change or be valuable to commercial interests like fisheries.
“The ocean and its biomass are changing together with the rest of the climate,” Sonnewald says. “Even though some are of greater socioeconomic interest than others, keeping track of overall changes can be very valuable.”
To make the work possible, the researchers developed a machine learning algorithm, the Systematic AGgregated Eco-province (SAGE). They trained SAGE using data from the Massachusetts Institute of Technology’s Darwin Project, which compiles data on wind, current, temperature, and phytoplankton populations around the world. The algorithm took the model’s dense and interconnected data and found that some ocean regions had common characteristics. These clusters of similarity became the ecoprovinces.
Orhun Aydin, a researcher at Esri, a geographic information system software company, says that even with large data collection efforts like the Darwin Project there is still plenty of data researchers would like to have about the ocean that is currently unavailable. Machine learning offers a way.
“We need models that can extrapolate beyond things we can observe right now,” Aydin says. “[Machine learning] could be invaluable going forward.”
Sonnewald says she hopes to find more uses for SAGE and is collaborating on a project about ocean acidification to see if the tool can localize and compare similar provinces of acidification.
“The topic is quite different, but the SAGE method is letting us discover similarities and differences that could otherwise be obscured,” Sonnewald says.