Plenty of Fish in the Sea for Big Data, AI to Tackle

By Adam Allington

British fisheries scientist John Shepherd was famously quoted as saying that managing fisheries is “like managing a forest, in which the trees are invisible and keep moving around.”

This fundamental problem has bedeviled fisheries managers for decades—how to prevent overfishing, without knowing exactly how many fish are in the sea.

That task could soon become much easier: The same image-recognition technology used by police and the military may soon be able tell the difference between a flounder and a red snapper as they swim by.

As companies rush to develop increasingly high-resolution image applications, scientists and government agencies are poised to unlock a treasure trove of data on some of world’s most valuable fisheries.

Scientists and fishermen want that data, because the numbers provide the foundation for a bevy of federal and state rules governing how many fish anglers are allowed to catch.

Need for Better Data

The National Oceanic and Atmospheric Administration tracks 474 fish stocks—subpopulations of a particular fish species—managed by 46 separate management plans.

“But a large fraction of those stocks are what we’d call ‘data poor,’” Steve Murawski, a fisheries biologist with the University of South Florida, told Bloomberg Environment.

Much of the data fisheries managers rely on is still collected using methods that are 150 years old, Murawski said. They include the use of trawl nets or hooks to catch fish, count them, and measure them by hand. That data is then used to create a biomass index—not a full census.

“It’s sort of like the Dow Jones,” Murawski said, referring to the stock market index showing how large, publicly owned companies have traded. “The index can be going up or down, but it’s not a detailed picture of the total economy.”

In an effort to get better data, Murawski’s lab is teaming with Silicon Valley developers to adapt facial recognition technology to what he calls “fish-o-recognition.”

“It’s basically a military knockoff, the computer is able to distinguish moving bodies, using underwater video. Every time a fish swims by the computer creates a new file.”

The major criticism from the recreational fishing industry is that regulations often don’t match up with the conditions that fisherman observe out on the water.

More Timely Data

In its annual “Status of Stocks” report delivered to Congress on May 17, NOAA found that of the 235 stocks with prior overfished status, just 35 are still considered overfished. On the list are such popular species as the Atlantic cod and Pacific bluefin tuna.

Environmentalists and many commercial fishermen hailed the news as proof that the quotas and catch limits in the Magnuson-Stevens Fishery Conservation and Management Act are working.

Others, like the boating industry and many recreational fishermen, took it as evidence that the time has come to relax what they consider overly restrictive federal regulations.

“Clearly we have perception issues, that somehow things are out of sync,” said Roy Crabtree, regional administrator for NOAA’s Southeast field office, which includes the Gulf of Mexico and Caribbean.

A big problem with many stock assessments, Crabtree told Bloomberg Environment, is that they can cost millions of dollars and require research vessels to collect data across a wide swath of ocean.

“We’re always struggling to maintain funding levels to pull that off,” he said. “Here in the Southeast, we have stocks that have never been assessed, and others are only counted every three, four, five years.”

After data is collected, Crabtree said, it then goes through a lengthy analysis and peer review before the council’s scientific and statistical committee makes a final catch-limit recommendation.

“The trouble is, we don’t always know where the fish are coming from until they start catching them,” he said. “And by the time we get a new assessment and catch up, they may already be shut down for a year or more.”

One Fish, Two Fish

Atlantic sea scallops are one of most valuable commercial fisheries in the U.S, which, until the mid-1990s, were also severely overfished.

“The traditional method for doing scallop assessments was by dredging,” said Scott Gallager, an associate scientist at the Woods Hole Oceanographic Institution on Cape Cod.

Not only was dredging invasive, Gallager said, but it was time-consuming to capture the scallops and measure them by hand. Eventually, researchers learned to make assessments using two-dimensional photos of the sea floor.

“We got pretty far, but realized quickly that 3D would provide a huge leap,” he said.

To make that jump, Gallager’s team designed the Habitat Mapping Camera System, or HabCam. Roughly the size of a Mini Cooper, HabCam is towed behind a ship on a fiber-optic cable. The device uses stereo cameras, which function like human eyes, to create overlapping images that a computer uses to reconstruct fish measurements, even when they are swimming.

“That’s really important for assessment of scallops, which are based on a total biomass, so we need as accurate a measurement as possible,” said Gallager.

A previous version of HabCam is already being used by NOAA to conduct scallop assessments.

The goal now, says Gallager, is to design a newer version with the capacity for machine vision and learning built into the vehicles itself, “so it can move along the sea floor, learning as it goes.”

“Smart surveying could have a tremendous impact on the efficiency of stock assessments by taking the human out of the loop,” said Gallager.

“There is a lot of work to get there, but it is a matter of years, not decades.”

This Fish, That Fish

In addition to scallops, Gallagher told Bloomberg Environment that technology will soon be able to quantify the entire environment on the sea floor, allowing scientists to study the relationships between noncommercial species as well.

“The holy grail of fishing management is species identification,” said Mike Hillers, technical sales manager in Lynnwood, Wash., for Simrad Inc., a Norwegian company that makes electronics for the commercial fishing industry.

“We realized 6-7 years ago that we could put live video on a drone, send it down 200 fathoms [1,200 feet] and identify fish with your eyes,” said Hiller.

A big advantage of using drones, he says, is that they can spot fish that might normally be scared away by loud boats and dredgers. The problem is that it still takes an actual human to review the footage and count the fish.

But that may also soon be changing.

“Computer vision and machine learning is huge in oceanography right now,” said Tom Dempsey, an ocean program director for The Nature Conservancy. Dempsey directs a pilot program called “Poseidon” that uses a mobile app to capture length data of red abalone from geo-tagged images uploaded by harvesters.

Dempsey told Bloomberg Environment that length data correlates strong with fertility, not just in abalone, but other fish as well. It offers the potential for a real-time data stream that can be scaled quickly to other fisheries as well.

“We have demonstrated that these length models can create high-resolution data sets to support much more adaptive management decisions, all at a tiny fraction of the cost of a traditional stock assessment.”

Dempsey says TNC is currently adapting Poseidon for use on other species such as spiny lobster, as well as some finfish, by the end of the year.