Pinniped Culling, Salmon Recovery, and the Case for Causal Discovery
A recent article1 describes legislation (HR 9621) that the U.S. Congress has introduced to allow culling (killing) pinnipeds (harbor seals, California sea lions, and Steller sea lions) in Washington state to recover native salmon populations. Estimates are that in certain parts of the Sound two-thirds of the predation of salmon smolts (juvenile salmon) is by pinnipeds. The number of seals and sea lions has increased dramatically since the introduction of the Marine Mammal Protection Act in 1972, arguably a conservation success that has partly led to the current situation. Proponents of the bill suggest that targeted killing of seals and sea lions would reduce their numbers and allow salmon smolts to survive better as they migrate out to sea, where they mature before returning to Puget Sound to breed. Already there is limited sanctioned killing of pinnipeds in the Columbia River system, numbering 240 as of the end of last year.
Proponents of the bill have an intuitive causal story: fewer pinnipeds leads to more surviving smolts, which leads to healthier salmon runs. But intuitive causal stories are exactly what get species killed for nothing. Below, I assert that before we commit to killing pinnipeds, whether through the currently proposed bill or through experiments some scientists have proposed, there's a lower-cost step we should take first: testing the causal story itself against the data we already have.
Opponents of the legislation argue that the food web in Puget Sound is quite complex, with many species (including some other prey of seals and sea lions, like hake) also preying on salmonids. Scientists, including the Washington State Academy of Sciences, have concluded that the true causal structure leading to salmonid mortality is complex. A report from the State Academy of Sciences in 20222 stated: “Given the large number of trophic links between pinnipeds and salmonids and the potential for direct and indirect ecological interactions, it is impossible to predict with certainty the outcomes for salmon and the rest of the food web under scenarios where the pinniped population size is changed.” Further, ecological niche theory suggests that if habitat conditions sufficiently support a certain population size of seals and sea lions in competition with other predators, killing some will simply open more availability in this niche, which will then be filled once again by increased breeding and in-migration of pinnipeds from other areas. (The diagram below actually shows two specific mechanisms for this: more on that shortly.) In other words, killing pinnipeds may only be a short-term solution, if it works at all. Opponents argue that killing animals is much too high a price to pay for such an uncertain outcome. On top of this, marine animal rescue organizations are reporting a dramatic increase in the number of pinnipeds they are treating with bullet wounds from attempted poaching or other unsanctioned killing3, causing additional pinniped suffering without doing anything to address the problem at hand.
“Given the large number of trophic links between pinnipeds and salmonids and the potential for direct and indirect ecological interactions, it is impossible to predict with certainty the outcomes for salmon and the rest of the food web under scenarios where the pinniped population size is changed.”
What to do? All involved seem to recognize and acknowledge that the small and non-recovering salmon runs in Puget Sound are undesirable (including Tribes, the Washington State Department of Fish and Wildlife, commercial fishing interests, marine biologists, and the public at large). The disagreement isn't over the goal. It's over how confident anyone can be that a given action will actually reach it.
It helps to be precise about what “confident” means here. Judea Pearl's “ladder of causation”4 distinguishes three levels of claim: association (what we can observe, e.g. areas with more seals tend to have fewer smolts), intervention (what happens if we act, e.g. if we kill X seals, smolt survival changes by Y), and counterfactual reasoning (what would have happened otherwise, e.g. would this year's run have been larger if we hadn't culled). Most of the data anyone currently has, from smolt counts at detection points to pinniped population estimates to predator and prey surveys, sits on the bottom rung: association. Getting up the ladder is where the disagreement actually lives.
Estimating the counterfactual effect of pinniped killing on salmon run size, true Rung 3 causal inference, is possible in principle from observational data alone, but only if you already have a complete and correct causal diagram of the system's confounders. Running that kind of inference on an incomplete or wrong diagram yields a confident, precise, and wrong answer, arguably worse than admitting uncertainty. Right now, nobody has a diagram anyone would call complete. So Rung 3 is currently off the table.
As such, the approach suggested by the Academy of Sciences (controlled experiments, a Rung 2 causal approach) is supported from a mathematical viewpoint. Such experiments could help uncover the true causal structure of the small salmon runs, although they may not do anything to help resolve the problem itself. Instead, well-designed experiments would help define the direction of arrows in the causal diagram and the functional form of interactions between certain nodes. For example: does killing pinnipeds in one area have different effects than killing them elsewhere? Does it change smolt predation rates in measurable ways? Do estimated numbers of Bigg's killer whales shift when pinnipeds are selectively killed?
The issue, as noted above, is that such experimentation would involve killing large numbers of seals and sea lions with no guarantee of successfully recovering salmon populations. From an ethical perspective, there is a valid claim about whether humans controlling populations by killing specific species while leaving others unharmed is even justified. This claim strengthens when the outcome has large uncertainty, as in this case. Rung 2, in other words, buys real causal knowledge, but at a cost we should not pay before we've tried something without such tremendous cost.
What the Diagram Actually Says
A less costly approach starts with the causal diagram itself. A partial causal diagram taken from the cited article as well as other literature is shown below, with known gaps indicated (note that this says nothing about unknown gaps: the ones we haven't thought to draw yet). This is a wicked ecology problem, and the diagram shows that plainly. Six of its nodes are flagged with a bare “?” because we don't know the functional form of the relationship, or don't have the data to estimate it at all.
A few of those gaps are worth naming, because they're the difference between an intervention that works and one that doesn't:
- Whether a subset of the pinniped population functions as salmon-smolt specialists, and what fraction of animals that describes: this changes everything about whether broad culling would even touch the relevant predation.
- Whether hatchery releases “lure” predators toward hatchery fish and away from wild Chinook, or “swamp” predators with more prey than they can eat either way. The diagram marks this relationship as genuinely unknown in direction, not just unmeasured.
- Chinook smolt are physically too small for the tagging approaches already used on steelhead, so there's no equivalent smolt-level predation-rate estimate for the species the debate cares most about. (This is a task for a clever marine engineering team!)
- And, arguably the single most consequential unknown on the whole diagram, what share of pinniped predation deaths are additive versus compensatory. If a smolt eaten by a seal or sea lion would otherwise have died anyway from disease, starvation, or another predator, killing pinnipeds does nothing for the salmon populations the bill is trying to protect. The diagram marks this node as unresolved. Everything the bill assumes depends on it not being small.
The Case for Causal Discovery, Before Rung 2
Understanding the functional forms of the intermediates would do much to allow testing the validity of this CDD (or a more complete version, built with formal elicitation methods), and here's the key point: it can be done without killing pinnipeds at all. Even a CDD as rudimentary as the one shown may accomplish one goal: causal discovery, a genuinely different tool from causal inference, could be attempted to determine whether the relationships shown in the CDD are even supported by existing data. If two nodes in the diagram are shown as independent but statistical analysis shows they are conditionally dependent, that's a signal the diagram is wrong, and it lets us revise it. This in turn adjusts both the additional data required and the case for killing pinnipeds at all. Unlike Rung 2 experiments, this works on Rung 1 data, the observational data we already have; and unlike Rung 3 inference, it doesn't require the diagram to be correct going in. Testing the diagram is the point. I recommend this approach first: causal discovery algorithms are well understood, automatable in code, and could bring all stakeholders closer to a common understanding of the likely causal structure of salmon survival in Puget Sound, before anyone commits to a policy that requires killing animals to learn whether it worked.The diagram also does something the “niche theory” argument above only gestured at: it shows specific, named mechanisms for why a cull might self-correct. Two balancing loops sit right on top of the harbor seal node. In one, reduced pinniped abundance opens room for pup survival and in-migration to refill the population; in the other, pinniped density itself draws in more Bigg's killer whale predation, which pushes numbers back down. A third loop closes around the policy itself: HR 9621's removal cap is written as a percentage of estimated abundance, so the quota mechanically shrinks as seals become scarcer. A fourth loop closes around low salmon runs generating the political pressure for more removals in the first place; the bill isn't just an input to this system, it's a node inside it. None of this means killing won't work. It does mean the system is built, by ecology and by policy design alike, to resist large or lasting changes from removal alone, which is exactly the kind of claim causal discovery could confirm or overturn using data we already have.
A Harder Question Underneath
In parallel, a robust public debate should be continued as to the point of all of this work. One of the objectives of salmon recovery is to provide food for the dwindling Southern Resident Killer Whale population, down to 74 individuals at last count5. Recent research shows that the "resident" orca populations are in fact less resident than the heretofore-called "transient" populations6. What is the value of trying to save resident orca populations if they aren't even here much anymore? That doesn't help the local ecosystem, it may not help the orcas themselves, and it doesn't help tourism. Have we crossed a threshold into a new basin of attraction? If the point is simply to have enough salmon for people to catch and eat, the feedback loop inherent in that goal is unlikely to produce stable salmon populations if the human population continues to grow. Indeed, the stated target for salmon recovery in Puget Sound reads7:- By 2050, 2 to 4 populations of natural-origin Chinook salmon in each biogeographic area meet their abundance recovery goals to achieve self-sustaining, harvestable salmon runs, with sustained, measurable increases in natural-origin Chinook abundance.
- By 2050, all Chinook salmon populations increase, and at least 50% of populations reach their recovery goals.
Are we fundamentally working at cross purposes?
References and Further Reading
- Dunagan, C. Congress debates lethal removal of seals and sea lions in Puget Sound | Encyclopedia of Puget Sound. https://www.eopugetsound.org/article/congress-debates-lethal-removal-seals-and-sea-lions-puget-sound.
- Pinniped Predation on Salmonids in the Washington Portions of the Salish Sea and Outer Coast | Washington Department of Fish & Wildlife. https://wdfw.wa.gov/publications/02579.
- Henderson, B. 700-pound sea lion found dead in West Seattle with numerous projectiles in its body. king5.com (2026). https://www.king5.com/article/life/animals/700-pound-sea-lion-found-dead-west-seattle-numerous-projectiles-its-body/281-85078bd7-3065-4d78-bd84-8d87a18582d5.
- Pearl, J. & Mackenzie, D. The Book of Why: The New Science of Cause and Effect. (Basic Books).
- Meet the Southern Residents. Orca Conservancy. https://www.orcaconservancy.org/meet-the-southern-residents.
- Rand, Z. R., Koehn, L. E., Morrigan, A. & Hanson, M. B. Increasing presence of Bigg's killer whales and changing seasonality of Southern Resident killer whales in Washington waters. PLOS ONE 21, e0350181 (2026).
- Puget Sound Salmon Recovery Hub. https://pssalmonhub.wa.gov/pages/recovery-targets.