The Federal Government Stopped Counting the Salmon. So a Community of 1,200 People Built a Better System.

The system’s name is SalmonVision. It was built by the Heiltsuk Nation in partnership with the Pacific Salmon Foundation, the Wild Salmon Center, and computer scientists at Simon Fraser University. By the end of that season, it had given the Heiltsuk something they hadn’t had in years: enough reliable data on returning sockeye to open a harvest. The community took roughly 500 fish. Elders received three per household. It was, by any commercial measure, a tiny catch. For a community of roughly 1,200 people on British Columbia’s central coast—part of a nation whose total registered membership is approximately 2,500, most living elsewhere—who had gone years without enough data to harvest their own fish from their own river, it was the first tangible proof that a different way of watching was working.

That matters now because the institution that was supposed to be watching—Canada’s Department of Fisheries and Oceans—has been stepping back for decades, and the retreat is accelerating. What the Heiltsuk built in its absence is more accurate, more timely, and more locally accountable than what it replaced. It is also spreading. As of the 2024 season, SalmonVision is running across a dozen First Nations on BC’s north and central coast. The question is no longer whether the technology works. The question is what it means when the people who need the data most are the ones who built the system to collect it.

The collapse no one measured

The scale of what has been lost is hard to see precisely because the loss is in the measurement itself. A study published in August 2025 in the Canadian Journal of Fisheries and Aquatic Sciences (Atkinson et al., doi:10.1139/cjfas-2024-0387), led by University of Alberta researcher Emma Atkinson, found that annual monitoring of Pacific salmon stocks has declined by roughly 50% since the 1980s. The worst decade on record is the most recent one: between 2014 and 2023, nearly two-thirds of historically monitored salmon populations in British Columbia and the Yukon had no reported spawning estimates at all. That is the poorest scientific coverage since broad-scale surveys began in the 1950s.

The decline tracks a pattern that Atkinson’s research connects to commercial priorities. For three of five Pacific salmon species—coho, chum, and pink—changes in monitoring closely mirrored changes in commercial value. When a fishing moratorium on coho was introduced in the early 2000s, monitoring of coho populations dropped off shortly after. The fish were still there. The commercial incentive to count them was not. Pull back far enough, and the shape is visible: nearly 7,000 distinct salmon populations across British Columbia and the Yukon, and the country’s capacity to track them is contracting toward whatever the market still values. Zoom in, and each lost data point is a specific river where no one walks the banks anymore.

A January 2026 study by Michael Price and Jonathan Moore at Simon Fraser University, published in the Canadian Journal of Fisheries and Aquatic Sciences, puts the consequence in blunt terms: as of 2023, there is now insufficient data to assess the health of 44% of Canada’s Pacific salmon conservation units. Price, who is also director of science at SkeenaWild Conservation Trust, described federal monitoring as increasingly reliant on “indicator streams,” where data from one stream is used to estimate fish health across many others nearby. The approach has a fatal flaw. Salmon populations fluctuate enormously, even between adjacent streams. Extrapolation from a handful of sites does not produce accuracy. It produces a comforting illusion of coverage.

On BC’s north and central coast specifically, the number of streams that are routinely monitored has declined by 50% since 2006, according to Pacific Salmon Foundation data reported by The Narwhal. In the summer of 2025, Canada’s National Observer reported that seasonal “creek walkers”—contractors who trek along streams to record salmon returns—had not been hired for the spawning season from Bella Bella to the Alaskan border, including major watersheds like the Skeena, Nass, and Kitimat systems. Misty MacDuffee, wild salmon program director at Raincoast Conservation Foundation, called it unprecedented.

The federal budget is making it worse. According to DFO’s Comprehensive Expenditure Review, released as part of the November 2025 federal budget, the department projects spending cuts of $54 million in 2026–27, $101.9 million in 2027–28, and $193.8 million in each of the two years after that—roughly $550 million over four years, as calculated by SkeenaWild Conservation Trust. On top of that, all federal departments except Defence and a handful of others have been directed to cut operational spending by 7.5% for the 2026–27 fiscal year, 10% the following year, and 15% in 2028–29. DFO says it will achieve these targets by winding down research and monitoring activities “that have either achieved their objectives or for which alternative data sources exist.”

Then, on March 20, 2026, Canada’s National Observer reported that DFO had renegotiated the British Columbia Salmon Restoration and Innovation Fund—the primary federal–provincial program that has funded salmon conservation work, including SalmonVision’s development—with conservation quietly removed from the fund’s mandate. (The report is available at nationalobserver.com.) The original program, which supported 170 projects and invested $128.56 million since 2019, was explicitly designed to conserve and restore Pacific salmon populations. The renegotiated version replaces that language with references to economic development and market competitiveness. The fund expires on March 31, 2026—days from this writing.

Gordon Sterritt, executive director of the Upper Fraser Fisheries Conservation Alliance, described DFO’s silence on whether conservation would remain in the mandate as damning. The institutional floor is being removed.

A weir on the Koeye River

Bella Bella sits on Campbell Island, at the heart of the Great Bear Rainforest on BC’s central coast. It is the largest community of the Heiltsuk Nation—approximately 1,200 people on-reserve as of the 2021 census, with a total registered nation of roughly 2,500. The town is accessible by float plane or ferry. There are no roads connecting it to anywhere else. From a sufficient distance—a satellite’s distance, or a national budget’s—a community this size is easy to miss entirely. But resolution is a choice, and some things only become visible when you decide to look at the scale where they exist. The Koeye River, about 60 kilometres south, flows through some of the most productive salmon habitat on the coast.

The Heiltsuk have managed fisheries on these rivers for thousands of years. The tool they used for much of that time was the fish weir—a fence built across a river that guides salmon through narrow openings where they can be selectively harvested or counted. Archaeological and oral history evidence suggests that Heiltsuk weirs and fish traps date back millennia; nearby stone traps on Hunter Island have been dated to at least 270 to 370 years ago, and the settlements associated with them are roughly 6,000 years old. The technology was not primitive. It was precise, selective, and sustainable across a timespan that dwarfs the existence of the Canadian state.

In the late 1800s, the colonial government banned fish weirs and traps. The stated rationale was that weirs damaged fish stocks—a claim made despite generations of evidence to the contrary. The actual effect, as the historical record makes clear, was to consolidate control of fisheries resources under the federal government and the commercial sector. Traditional fishing technologies, including weirs, spears, and certain net configurations, were criminalized. In the early 1900s, DFO actively destroyed fish traps in Heiltsuk territory. The Fisheries Act of 1894 confined Indigenous fishing to a “food fishery” under strict permission, and by the early twentieth century, Indigenous people across British Columbia had been systematically excluded from the fisheries that had sustained their communities since before European contact.

In 2013, the Heiltsuk began rebuilding. Working with the Qqs Projects Society, academics, and conservation organizations, they constructed a modern weir on the Koeye River. The project combined traditional design principles with contemporary monitoring techniques—tagging fish, conducting stream walks, and building the first mark-recapture estimates of sockeye abundance in the watershed. The results were published in a peer-reviewed study in 2017, documenting sockeye escapement estimates ranging from 4,600 to 15,000 across four years.

The weir worked. But counting fish by hand, in a remote watershed accessible only by boat, remained labour-intensive and expensive. The Heiltsuk needed more data than their field capacity could produce. That is where SalmonVision entered.

Five million frames

SalmonVision is, at its core, a computer vision system. Solar-powered cameras are installed inside the fish passage boxes of weirs—the narrow openings through which salmon swim as they move upriver. The cameras record continuously. The footage is transmitted by satellite to computing infrastructure where AI models, developed by scientists at Simon Fraser University’s NetMedia Lab, perform object detection and multi-object tracking. The models identify salmon species, count individual fish, and in some cases assess sex and whether the fish is wild or hatchery-origin. The data streams back in near-real time.

The training dataset is substantial. As of the 2024–25 season, the models have been trained on more than five million annotated video frames encompassing twelve salmon species. The system achieves over 90% accuracy for species detection. At the Haida Nation’s Yakoun River sonar program, AI tools were used to produce an annual estimate of Chinook returns—a task that would otherwise have been impossible because high numbers of pink salmon made it impractical for human reviewers to distinguish the larger Chinook visually.

The upfront cost is approximately $50,000 per site for equipment and installation. That is not trivial for small communities. But the system gets cheaper over time, and the alternative—flying staff into remote watersheds for weeks of manual counting—is more expensive per data point and less comprehensive. Dr. William Atlas, the Wild Salmon Center scientist who serves as SalmonVision’s principal investigator, has worked with the Heiltsuk since 2012. He frames the system’s value in terms of what it replaces: pre-season estimates based on historical models that assume this year will resemble the past. In an era of accelerating climate change, that assumption has become dangerous.

Christina Service, a biologist with the Kitasoo Xai’xais Stewardship Authority—a neighbouring nation of 350 to 400 people—put the practical impact simply. The Kitasoo Xai’xais operate a weir on the Kwakwa River that DFO does not monitor because the salmon there do not support major commercial fisheries. Without SalmonVision, the community would not have the capacity to generate this scale of monitoring data on its own. The AI tool gives them high-resolution, high-quality data without exhausting their local workforce.

As of the 2024 season, the system is running at weir sites operated by at least eight First Nations: the Heiltsuk, Kitasoo Xai’xais, Gitga’at, Gitanyow, Nuxalk, Wuikinuxv, Haida, and Taku River Tlingit. The Skeena Fisheries Commission, an umbrella organization representing multiple First Nations on the Skeena River, is also a partner. SalmonVision has become a platform.

What 'sovereignty' means here

It would be easy to tell this as a technology story and stop there. A community needed data. AI provided it. The system works. But that version strips out the dimension that makes the Heiltsuk’s achievement structurally different from a typical tech deployment—and it is the dimension that the brief requires this piece to hold.

The Koeye weir is built from traditional designs. The technology that the Canadian government banned in the late 1800s to consolidate control of fisheries resources is the same technology the Heiltsuk rebuilt and then equipped with artificial intelligence. That is not a symbolic gesture layered onto a tech project. It is the load-bearing structure. The weir is both the monitoring platform and the act of reclamation. Without the physical weir, the cameras have nowhere to mount. Without the cultural and political decision to rebuild what was banned, there is no infrastructure at all.

Housty has described the Koeye weir as “a modern-day expression of not only Heiltsuk title and rights, but also an avenue for us to be a part of the latest science.” Atlas frames it in terms of self-determination: by combining the ancient technology of weirs with computer vision, First Nations now hold the data to inform decision-making in-season and out. The community has moved from being dependent on DFO’s estimates—which arrived late, relied on extrapolation, and were often inaccurate—to holding their own data, collected from their own infrastructure, on their own terms.

The word “sovereignty” is used deliberately here, and not as rhetoric. Data sovereignty means controlling the collection, ownership, and use of information about your own community and resources. Before SalmonVision, the Heiltsuk relied on DFO for information about fish in their own rivers—information that was often unavailable, outdated, or generated using methods the community had no role in designing. Now they produce the data themselves. They decide when to harvest and when to wait. The 500-fish harvest in 2023 was not an act of hope or tradition. It was a management decision made with confidence, based on data the community owned.

The floor beneath their feet

The timing of this story carries its own argument. The BC Salmon Restoration and Innovation Fund—which provided grant funding for SalmonVision’s development through the Pacific Salmon Foundation—expires on March 31, 2026. DFO’s renegotiated version of the fund drops conservation from the mandate. The federal budget is cutting DFO’s research and monitoring capacity by hundreds of millions of dollars over the next four years. Creek walkers are not being hired. Stock assessment budgets are being slashed. A coalition of 26 organizations—including academic scientists, Indigenous groups, commercial and recreational fishers, and conservation nonprofits—sent an open letter to the fisheries minister calling the situation untenable.

The institutional floor is being removed at exactly the moment the community-built alternative has proven itself. SalmonVision is expanding. The data it produces is more accurate and more timely than DFO’s pre-season estimates. The technology is designed to be adopted by other communities—the SalmonVision Collaborative is building a web application to make its AI tools accessible to users beyond the original partners, with plans to extend across the Pacific Rim. A letter published in Science in January 2026 warned that Canada’s dismantled safeguards are creating a widening gap between the country’s pace of industrial development and its capacity to monitor the consequences.

This is the tension the piece holds without editorializing: the federal system is retreating from its most basic monitoring responsibilities at exactly the moment a community-led system demonstrates that the work is possible and scalable. Whether that retreat will strangle the alternative’s funding or force its independence is a question the next year will answer.

What gets counted

This story appeared in one regional outlet. It was discussed in academic fisheries journals and on the project pages of conservation organizations. No trending metric flagged it. No algorithmic feed surfaced it for a general audience. A community of 1,200 people counting 500 fish does not clear the filters that determine what most people see. It surfaced here because this is what happens when you look at everything without a commercial reason to look away—you find what the filters were never designed to find.

There is a parallel worth noticing. The Heiltsuk built their own monitoring system because the institution that should have been paying attention stopped paying attention. The reasons were structural: optimize for what’s commercially valuable, fund what’s scalable, count what counts for the largest stakeholders. The same structural logic applies to how stories reach audiences. A story about data sovereignty on a remote coast, involving a community smaller than a city block, does not optimize for reach. It optimizes for understanding.

What the Heiltsuk proved on the Koeye River is that the data gap is not inevitable. It is a choice—made by institutions that decided what was worth measuring and what was not. The communities filling that gap are not waiting for the institution to come back. They are building the infrastructure to watch for themselves, with tools precise enough to count individual fish by species in real time, mounted on structures their ancestors designed and a colonial government once burned.

Three fish per Elder household. Five hundred sockeye from one river in one season. A dozen nations across BC’s coast, each running their own weirs, generating their own data, making their own management decisions. It is small. It is measurable. And it was invisible until someone looked. The question worth carrying out of this story is not whether the Heiltsuk’s system will survive the next round of federal cuts. It is how many other Koeye Rivers are out there right now—problems being solved, data being gathered, structures being rebuilt—in the gaps between what the institutions that shape your attention have decided is worth showing you.