Anyone who has tried their hand at fishing knows how elusive their prey is. Obscured by the water, it’s nearly impossible to catch sight of fish in most circumstances.
Now imagine being an ecologist trying to count them. Scientists have relied on all manner of methods to track fish populations, many of them relatively crude: applying an electric shock to a stream and counting the stunned fish; hauling a large net through the ocean, then calculating entire populations based on the catch; snorkeling down a river; standing at a fish ladder on a dam and ticking off the fish as they swim past.
The results can be vital, helping fisheries managers determine whether to open a fishing season, conservationists gauge the health of a species, and scientists figure out if a restoration project made a difference. But the work is time consuming and plagued with uncertainty.
But some of the latest technological innovations are starting to trickle down to such humble pursuits as counting fish. In the last two months, researchers have independently announced the potential for tracking salmon populations with artificial intelligence (in one case), and DNA floating in the water (in the other).
The results are encouraging, and they could help scientists track their quarry far more easily.
In one case, researchers in British Columbia teamed with First Nations fisheries managers to see if they could find a better way to count salmon moving up rivers to spawn. The First Nations have fish counting weirs—basically small dams—on the Kitwanga and Bear rivers on the central coast of the Canadian province. There, the fish swim past video cameras that record their passage. But people need to laboriously scan the footage to figure out how many of different species pass by.
Fisheries managers, computer scientists and other researchers joined forces to see if computers could do the fish counting for them. If so, it promises to deliver results quickly enough that managers can promptly adjust how many fish people are allowed to catch.
“Without real-time data on salmon returns, it’s extremely difficult to build climate-smart, responsive fisheries,” says Will Atlas, a scientist with the Portland-based Wild Salmon Center who helped lead the work.
To speed things up, the scientists turned to computer programs designed to “learn” how to distinguish between different objects in an image. They fed the computer more than 500,000 pictures of fish swimming past the weirs. Twelve different fish species in the images had already been identified by human observers. That enabled the computer to know when it correctly identified a fish, and modify how it distinguished between them to become more accurate.
At its best, the system eventually could correctly identify two key salmon species—coho and sockeye—more than 90 and 80% of the time, respectively, the researchers reported in September in Frontiers in Marine Science.
The team that developed what they dub “Salmon Vision” is now working to provide regular automatic counts on several B.C. coastal rivers next year.
Across the border in Washington state, scientists wanted to devise a better way to track how salmon benefit from a massive $3.8 billion effort to replace tunnels through which streams flow beneath infrastructure such as roads.
Known as culverts, poorly designed or aging ones can block migrating fish by creating insurmountable waterfalls or long, dark tunnels. The construction spree is the result of a federal court ruling that the state was violating the treaty rights of area tribes by failing to maintain passable culverts.
University of Washington scientists wanted to know if they could harness new DNA tools to tell whether culvert replacement would help fish get upstream. They focused on replacements of two culverts along a stream that, coincidentally, flows near my house in the small city of Bellingham.
They scooped up samples of water above and below each culvert multiple times over a year and a half. They then used a technique called “metabarcoding” that scans the DNA filtered from the water to identify which species were nearby. If it found DNA for fish of interest, including sockeye and coho salmon, then it meant those fish had reached at least that section of water.
The results revealed that for one of the culverts, the replacement didn’t make much difference for which fish made it upstream. That’s because the eDNA showed the different fish species were already making it through. However, the larger culvert further up the stream did appear to be a barrier. The mix of species in the DNA before the renovation differed upstream and below, the scientists reported in late August in Ecological Applications.
“It is clear that not all things that are marked as a blockage to salmon are, in fact, blockages to salmon,” said Eily Andruskiewicz Allan, an environmental engineer who led the research.
While this is just two small culverts on a single stream, the results suggest that eDNA could be used more broadly to decide which western Washington culverts are a higher priority to replace. And beyond that, it signals that a small glass of water (and some fancy lab work) could be used to gauge the results of aquatic ecological restoration efforts anywhere.
“It was really an ‘Aha!” moment for us,” said Allan.
Atlas, et. al. “Wild salmon enumeration and monitoring using deep learning empowered detection and tracking.” Frontiers in Marine Science. Sept. 20, 2023.
Allan, et. al. “Quantifying impacts of an environmental intervention using environmental DNA.” Ecological Applications. Aug. 28, 2023.
Underwater video footage.courtesy of Gitanyow Fishery Authority