FAPESP and the Sustainable Development Goals

Scientists use artificial intelligence to simulate the effects of warming on the Southern Ocean

Scientists use artificial intelligence to simulate the effects of warming on the Southern Ocean

A study published in Frontiers in Marine Science is the first to predict the impact of long-term changes in sea surface temperature on local microbial diversity. The methodology is also innovative (aerial view of the Polar Research Vessel Almirante Maximiano; photo: Luciano Candisani)

Published on 10/25/2021

By Karina Ninni  |  Agência FAPESP – Attempting to “simulate life” by means of an artificial intelligence technique known as machine learning to predict the impact of global warming on the surface of the Southern Ocean, and particularly on the microorganisms that live there, was the aim of a study conducted by a multidisciplinary group of researchers including oceanographer Marcos Tonelli and biologist Amanda Gonçalves Bendia, postdoctoral fellows and collaborating professors at the University of São Paulo’s Oceanographic Institute (IO-USP) in Brazil, alongside five other scientists affiliated with the institution: Juliana Neiva, Bruno Ferrero, Ilana Wainer, Camila Signori, and Vivian Pellizari

In the study, the researchers considered four emission scenarios to estimate the surface sensitivity of the Southern Ocean to a warming climate. Focusing on microorganisms at the bottom of the food chain, they discovered a decrease in some beings that are involved in crucial biogeochemical processes, and that produce nutrients needed by themselves and many other life forms. They also detected an increase in groups that depend on consumption of these nutrients because they are heterotrophic (they do not produce their own food).

Two forms of prediction were used in the study. The first was a diversity index method focusing on differences between the scenarios in terms of the predicted decrease in diversity.

“The high-emission scenarios predicted a significant loss of diversity,” said Tonelli, first author of the article reporting the study, published in Frontiers in Marine Science.

The second was prediction for specific taxonomic groups (at the order level). Here the researchers found a decrease in abundance for groups with key roles in the environment generally, not just in Antarctica. The study was supported by FAPESP via two projects (12/23241-0 and 18/14789-9).

Innovative methodology

The four socioeconomic scenarios considered by the scientists were established by the World Climate Research Program (WCRP), which coordinates the development of climate and earth systems models by major modeling centers as part of the Coupled Model Intercomparison Project, now in its sixth phase (CMIP6). The CMIP6 models simulate climate change under different scenarios of future anthropogenic impacts on the environment, known as shared socioeconomic pathways (SSPs).

The researchers selected four SSPs that illustrate the possible anthropogenic drivers of global warming in the current century: SSP1-2.6 (the “green road” to sustainability, with low challenges to mitigation and adaptation); SSP2-4.5 (a “middle of the road” scenario, with medium challenges to mitigation and adaptation); SSP3-7.0 (a “rocky road” with regional rivalry and major challenges to mitigation and adaptation); and SSP5-8.5 (“taking the highway” with development powered by fossil fuels and major challenges to mitigation but low challenges to adaptation). The Southern Ocean surface temperature rises corresponding to these four SSPs for the period 2015-2100 were 0.3 °C, 0.7 °C, 1.25 °C and 1.6 °C respectively.

“The high-emission scenarios predicted a much earlier surge in temperature driven by human activity throughout the Southern Ocean,” Tonelli said.

Microbial community datasets were obtained by the group from previously published studies conducted under the aegis of the Brazilian Antarctic Program, comprising a total of 105 surface water samples collected in the Northwest Antarctic Peninsula and northwest Weddell Sea. Bendia took part in several of these studies, conducted by Signori and Pellizari.

“Water samples were collected at several different points and at a depth of about 5 meters. Large amounts of water were filtered to concentrate the microorganisms. The aim was to analyze all the organisms found in the environment and to focus on its diversity. Cell DNA was extracted and sequenced. The microbiological data were obtained in previous projects funded by the Brazilian Antarctic Program [Interbiota, EcoPelagos, Microsfera, and Criosfera], with Signori and Pellizari participating. Tonelli gave us the idea of assembling all the samples and datasets available for inclusion in the model. We had data for several years,” Bendia explained.

Finally, to “simulate life” the researchers used machine learning, a branch of artificial intelligence in which algorithms process sample inputs to develop a model capable of making predictions or decisions in accordance with the data supplied.

“I work with climate projections and global models,” Tonelli said. “The problem with these climate models is that they can’t simulate life. They’re based on physical equations. Modeling the physical part numerically is relatively straightforward, but we have yet to find equations that accurately represent life and biological processes. So we wondered what we could do if the models available failed to reproduce the impact on life [in this case, microbial communities].”

According to the article, random forest algorithms and neural networks (machine learning models inspired by the human brain and used to recognize patterns in vast amounts of data) are among the most effective tools for analyzing microbiomes.

“We decided to use a random forest model to investigate the microbial response to long-term change in sea surface temperatures, in terms of diversity and composition,” he continued. “It was a machine learning model that combined several decision trees, each one of which was trained on a slightly different set of observations. The final prediction was produced in accordance with the outcomes of each decision tree.”

The main challenge was calibrating the equipment. “A great many samples were needed to ‘train the machine’ to reproduce reality,” he explained. “We had 105 samples, so we used about 80 for training, and the rest to test the calibration. We knew what ‘reality’ was because we had samples collected in the Antarctic. We adjusted the machine until it was able to reproduce the real situation. Then we fed in the climate data, and the model gave us an answer.”

According to Tonelli, this is the first time machine learning has been used in a study of this kind. The methodology can be used to study other oceans.


The simulations indicated a decrease in microbial community richness and diversity under all scenarios. The high-emission scenarios, especially the most critical (SSP5-8.5), led to the most substantial decreases.

While only minor changes in the relative abundance of microorganisms were predicted under the low-emission scenario (SSP1-2.5), the three scenarios that assumed a higher temperature rise, including the “middle of the road” scenario (SSP2-4.5), predicted changes in microbial community structures that included diversity loss, and a decrease in microorganisms with key roles in biogeochemical processes and the functioning of ecosystems in the Antarctic Peninsula and Weddell Sea.

Bendia highlighted the Nitrosopumilales order of ammonia-oxidizing archaea, explaining that all life on Earth is divided into three domains – Bacteria, Archaea and Eukarya – and Archaea are the least studied. 

“Our simulations predicted a drastic reduction in archaea of the Nitrosopumilales order, which oxidize ammonia and fix carbon dioxide [CO2]. They’re relatively well-known to science. They like cold water and are abundant in Antarctica. They remineralize organic matter and recycle nutrients for use by other microorganisms. If this process is interrupted, nutrients will go missing for other microorganisms. Our simulations also predicted a reduction in a group of bacteria that oxidize sulfur compounds, some of which can be toxic to certain organisms.”

Decreases were also predicted in little-known groups such as Marine Group II, the most abundant planktonic archaeal group in surface ocean waters. On the other hand, the model predicted an increase in the relative abundance of Flavobacteriales, an order that comprises several families of heterotrophic microorganisms.

“We expected the predictions based on the various scenarios and machine learning algorithms to point to several changes, but didn’t think they would occur in these key groups of microorganisms, which are most important to the region’s ecosystems and biogeochemical cycles,” Bendia said, recalling that the orders in question include several species with key roles in the functioning of marine ecosystems, such as the sulfur, nitrogen and carbon cycles, and species currently considered abundant in surface waters.

“We’re talking about the base of the food chain, the primary producers. The entire trophic system starts there and rises until it reaches the large mammals. If the consumers are increasing but the producers aren’t, how will this affect the higher levels? We don’t yet know. We need a good team of specialists to understand this. I believe some groups will benefit from the changes, while others will suffer,” Tonelli said.

The implications of a reduction in ammonia oxidation for the ecosystems they analyze are not clear, but some modeling studies have shown that it could affect nutrients, denitrification (microbial reduction of nitrite and nitrate to gaseous forms of nitrogen, which may return to the atmosphere), marine productivity, and biological carbon sequestration by the oceans.

Temperature changes modulate the dynamics of the microbial community in the Southern Ocean, the authors conclude in the article “Climate projections for the Southern Ocean reveal impacts in marine microbial communities following increases in sea surface temperature”, which is at: www.frontiersin.org/articles/10.3389/fmars.2021.636226/full.

Source: https://agencia.fapesp.br/37159