Editors’ Highlights are summaries of recent papers by AGU’s journal editors.
Source: Journal of Geophysical Research: Biogeosciences
Identifying and cataloging the myriad species of microorganisms found in modern sediments (and the fossil record), such as the phytoplankton group of diatoms, can be a daunting and laborious task, fraught with uncertainties and errors when attempted manually. Consequently, it is difficult to document the changes in these communities as they respond to environmental alterations associated with climate change.
Diatoms are delicate and beautiful photosynthetic marine microorganisms (as seen in the image above), which together account for a significant portion of the global Net Primary Productivity that sequesters carbon and provides the atmosphere with oxygen. However, when they die, their delicate glass shells (literally silica glass) tend to break up into pieces. Which of the tens of thousands to millions of species they belong to can typically only be identified by certain parts called frustules.
When looking at sediment through a microscope (see image below), it is difficult to distinguish these frustules from the surrounding sediment, and this is where machine learning can help. Enter the emerging tools of Artificial Intelligence (AI) in the form of Convolutional Neural Networks (CNNs) and our ability to characterize such microscopic communities is greatly enhanced. CNNs involve deep learning that enables classic neural networks to optimize learning by greatly reducing the number of nodes needed to connect to each other. These nodes are organized in layers that carry information from one to the other for analysis and, in so doing, can recognize patterns that are difficult to observe through simple observation. Whereas “traditional” feed-forward networks can only learn imagery patterns tied to specific locations, CNNs can transfer what is learned in one region of an image to other regions, thus greatly enhancing the ability to recognize patterns, and in this case, diatoms.
Godbillot et al. [2024] have developed a method to use such deep learning to more reliably identify the community makeup of diatoms and thus account for changes in species abundance and distribution in response to climate change. As a result, they help to usher in a new generation of tools for the biogeosciences that can be extended to other large data sets that may be difficult to observe and catalog manually. We can expect to see growing use of convolutional neural networks in in the coming years.
Citation: Godbillot, C., Marchant, R., Beaufort, L., Leblanc, K., Gally, Y., Le, T. D. Q., et al. (2024). A new method for the detection of siliceous microfossils on sediment microscope slides using convolutional neural networks. Journal of Geophysical Research: Biogeosciences, 129, e2024JG008047. https://doi.org/10.1029/2024JG008047
—Dork Sahagian, Associate Editor, JGR: Biogeosciences
