By Nancy Steinberg
The ocean teems with microscopic plants and animals — phyto- and zooplankton — in a mindboggling array of forms. These organisms are at the base of complex food webs, fueling fisheries and impacting ecosystems in countless ways. For these reasons, scientists often want to find, identify and count them.
But determining which plankton species are present at particular place and time is labor intensive. It used to be that the only way to examine plankton communities was to take samples in the field, preserve them and bring them back to the lab. Thousands of scientists have spent tens of thousands of hours examining such samples under microscopes all over the world.
Now, new technologies are making studies of plankton much easier. Two amazing instruments are being used by CEOAS scientists to take pictures of plankton species in situ (in the natural environment), and even identify them to species using artificial intelligence and machine learning.
Photogenic phytoplankton
Flow cytometers, now standard equipment in biomedical labs, are instruments that count and photograph individual cells in a fluid flowing past a camera. They have been used for decades to look for cancer cells, examine immune deficiencies and determine cell function. Scientists at Woods Hole Oceanographic Institution had the brilliant idea to adapt this instrument for monitoring phytoplankton in the field, adding image recognition to the device’s functionality so it can record not only how many phytoplankton cells are in a sample, but also the species present. The plankton-measuring cytometer is called an Imaging FlowCytobot.
Maria Kavanaugh, an oceanographer in the College of Earth, Ocean, and Atmospheric Sciences, has been using an IFCB for years, using large data sets and artificial intelligence to “teach” it to identify local and regional phytoplankton species.
“It’s so beautiful,” Kavanaugh says of the IFCB’s capabilities. “Before, we were relegated to understanding phytoplankton community structure through parameters like light scattering properties. Now we can get counts, size structure of the population, actual species abundances … it’s a great tool.”
The IFCB can either be deployed directly in the water, sampling for multiple months, or it can sit safely on a benchtop inside a laboratory. In either case, the meter-long device sips a small sample of water, which flows through a tunnel so narrow that only one cell at a time passes in front of a laser. The laser then triggers a camera to take an image of the single cell. The IFCB files the photo away, and goes on to the next cell in the stream. In one ten-minute sampling period, the IFCB can collect and categorize about 5,000 images. The old-school way of examining that same sample under a microscope would have taken six to eight neck-stiffening hours.
Machine learning and artificial intelligence play an important role here. The IFCB crops the image, then a series of algorithms identify the plankton species based on measurement of more than 250 attributes of each cell. The images are compared to a “field guide” prepared and fed to the algorithm by scientists. The scientists check the machine’s work and correct it, and over time, the classifications get increasingly accurate.
Kavanaugh’s lab uses the IFCB – they now operate two, nicknamed Luci and Pandora – for a range of projects. In the research realm, it is being used to characterize the phytoplankton communities on Oregon’s continental shelf and in the northern California Current, leading to understanding of regional food webs. Kavanaugh also uses one as a teaching tool with field biological oceanography classes.
Fantastic beasts and how to find them
In addition to microscopic plants, the ocean is home to spectacularly diverse wee beasties, otherwise known as zooplankton, including the larval stages of fish and crustaceans. CEOAS oceanographer Bob Cowen has been studying larval fish ecology for decades. Cowen and his team have dragged lots of nets through the water, filtering out larval fish and examining them under a microscope to identify them.
For the past fifteen years, though, he’s been using a more high-tech approach to examine larval fish (called ichthyoplankton) and other zooplankton in situ. Along with colleagues at the University of Miami and elsewhere, he developed the in situ ichthyoplankton imaging system (ISIIS), a camera system towed by a research vessel that takes pictures of the critters on the fly.
While in situ imaging systems existed pre-ISIIS, they had significant shortcomings when it came to finding larval fish. “Existing systems could only image very small volumes of water,” explains Christian Briseño-Avena, a former post-doc in Cowen’s lab. “There was no way to capture the least abundant zooplankton groups like fish larvae.”
ISIIS solves the problem by using a line scan camera, which emits a sheet of light a few microns thick. The sheet is projected from one port through the water to another port, and the instrument is towed through the water perpendicular to the sheet of light. Because the organisms in the frame are backlit, the camera is capturing the shadows of whatever comes through that plane. “It scans the water line by line, stacking the lines to make a picture,” Briseño-Avena says. “If we tow it for a kilometer, it’s creating a kilometer-long picture.”
The catamaran-shaped instrument, a bit larger than a mini-fridge, can be towed at a specific depth or in an undulating pattern, taking images from several liters of water at a time. Environmental data are collected simultaneously, so that plankton data can be compared to critical variables like temperature and salinity.
All of the critters in the ISIIS portraits need to be identified, and just like with the IFCB, Cowen’s research group and others are working with an AI-enabled image-recognition system to automate that process. “We create learning libraries for ISIIS by identifying the species we know,” Briseño-Avena explains.
ISIIS data have been used to examine larval distribution and transport in the Florida Keys, to determine the distribution of larvae in areas of the Gulf of Mexico affected by the Deepwater Horizon oil spill, and to look at food web dynamics in the Mississippi River plume. Locally, the lab has examined food webs and larval dynamics the California Current system.
There is no end to the questions that the lab can answer using ISIIS. “We keep pushing the boundaries of the instrument all the time,” Briseño-Avena says.
