Sea lions don't hunt the whole river. They hunt the places it narrows.
Every returning salmon in the Columbia is funneled, at least once, through water it cannot avoid — a tailrace shoulder, a ladder entrance, a tributary mouth. Predators found those places long before we did. A pinch-point exclusion gate takes the ambush lane back.
The argument in five sentences
Adult salmon transit engineered passage structures at extremely high rates — 96.6% mean passage efficiency across eight Columbia and Snake River dams, measured on 26,886 radio-tagged fish over fifteen years. Whether salmon will use a man-made structure is not an open question.
The limiting variable is where you put it, not how big it is. Attraction and siting, not scale, govern whether fish commit to a passage lane.
Physical exclusion has already been proven on this river: the sea lion exclusion devices installed at the Bonneville fish ladder entrances beginning in 2005 have kept pinnipeds out of the fishways for two decades with no documented breach and no habituation — a record no acoustic, visual, or hazing method comes close to.
But those devices protect eight ladder entrances. They do nothing for the open tailrace or the tributary mouths, which is where nearly all measured predation actually happens.
The proposal is narrow: apply the demonstrated exclusion threshold at the downstream chokepoints the existing devices were never built to cover.
What the evidence supports — and what it doesn't
This site is written for agency staff, tribal fisheries managers, and funders who will check the citations. Where the evidence is strong, it says so. Where a claim is a proposal rather than a finding, it says that too.
Physical exclusion at enclosable apertures
SLEDs at Bonneville's eight fishway entrances, 2005–present. USACE reports no pinnipeds observed inside the fishways across the monitoring record. When the devices were absent in fall 2015, animals got in — a natural before/after.
The current toolkit →Hazing, acoustics, and deterrence
A peer-reviewed study found a 31% temporary reduction in Steller numbers during active hazing, full habituation within a single season, and vigilance falling roughly 25% per exposure. In 2024 alone, 9,867 pyrotechnic rounds were fired at Bonneville.
Why deterrence habituates →Relocation
ODFW moved ten California sea lions 210 miles from Willamette Falls to the Oregon coast in 2018. All ten came back, most within four to six days. One returned twice.
The relocation record →Lethal removal
517 animals removed since 2008. Steller presence at Bonneville fell 91% in sea-lion-days between 2020 and 2024. But reviewers at the Northwest Power and Conservation Council state the program has "not been particularly effective" at its overall goal.
Law, numbers, limits →Pinch-point exclusion gates
A seasonally deployed, open-topped bar gate on the dominant flow line at a tributary mouth or tailrace shoulder. Extends a proven mechanism to unprotected geometry. Not yet built or tested.
Modeled benefit →Haul-out access barriers
Fencing between high-use beaches and the water, to make access inconvenient. No published evaluation exists, and every comparable case produced local displacement rather than departure. Presented here honestly, with the contrary evidence.
The honest assessment →The scale of the problem
Two bodies of evidence on this topic are routinely mixed together and should not be. Direct tailrace observation at Bonneville yields roughly 1–6% of the run. Telemetry across the whole 234-kilometre reach from the estuary yields 20–45%. They measure different things over different distances, with error running in opposite directions, and they must never be added together. The projections page explains the reconciliation in full.
Pilot site
The Cowlitz River mouth is the lead candidate: a constricted confluence with documented sea lion presence, inside the geographic scope of existing federal management authority, and far enough from the dam that it addresses the coverage gap agencies have themselves identified.
A twenty-four year record below Bonneville Dam
Sea lions were not always here. The Columbia's predation problem has a start date, a measured trajectory, and a species composition that changed halfway through — and each of those facts matters for what you build in response.
Systematic observation of pinniped predation in the Bonneville tailrace began in 2002. In that first season, thirty California sea lions were identified and no Steller sea lions were recorded at all. What followed is one of the better-documented predator range expansions in North American fisheries management.
The count, year by year
The U.S. Army Corps of Engineers has published seasonal estimates continuously since 2002. The figures below cover the January-to-May window and are the most-cited series in the record.
| Season | Run | CSL | SSL | Fish consumed | % of run |
|---|---|---|---|---|---|
| 2002 | 284,732 | 30 | 0 | 1,010 | 0.4% |
| 2005 | 81,252 | 81 | 4 | 2,920 | 3.4% |
| 2008 | 147,558 | 82 | 39 | 4,927 | 3.2% |
| 2010 | 267,167 | 89 | 75 | 6,321 | 2.4% |
| 2012 | 171,665 | 39 | 73 | 2,360 | 1.4% |
| 2014 | 219,929 | 71 | 65 | 4,621 | 2.1% |
| 2015 | 239,326 | 195 | 69 | 10,859 | 4.3% |
| 2016 | 154,074 | 149 | 54 | 9,525 | 5.8% |
| 2017 | 109,040 | 92 | 63 | 5,384 | 4.7% |
| 2018 | 100,887 | 67 | 66 | 3,112 | 3.0% |
The widely repeated figure of "264 sea lions at Bonneville in 2015" is the sum of individually identified animals across an entire season — 195 California plus 69 Steller. It is not the number present at once. Peak daily counts are far lower: in 2024 the peaks were 40 California sea lions in March and 38 Stellers in May, against season averages of roughly 3.5 and 6.9 animals per day. Anyone presenting the cumulative figure as a simultaneous count will be corrected by the first agency reader who sees it.
Halfway through, the species changed
California sea lions drove the early expansion. Stellers arrived later and increased steadily, and in 2017 they crossed over: Steller consumption reached 3,242 fish against 2,142 for California sea lions. For a period around 2019 and 2020 the Corps reported Stellers accounting for more than 90% of spring Chinook predation at the dam.
That dominance was a window, not a permanent state. By spring 2025 the peak counts were close to level again — 50 California sea lions against 48 Stellers. If you cite the 90% figure, date it.
The Steller arrival also moved earlier in the year. Corps analysis found arrival advancing by roughly 11.6 days per year from 2011, and by 2019 Stellers were absent from the tailrace only about six weeks annually — approaching year-round residence. That is a fitted trend over a limited period rather than a constant that can be extrapolated forward, but the direction is unambiguous.
White sturgeon: a Steller-only problem
Only Steller sea lions have been observed taking white sturgeon at Bonneville. Between the falls of 2017 and 2021 the mean was roughly 613 sturgeon annually, peaking at 1,119 in 2021 and with 762 observed killed in fall 2019 alone. The Corps has stated that this additive mortality over time "may be contributing to the questionable status of the stock." Recent seasons are dramatically lower — estimates in the teens for 2023 and 2024 — coinciding with Steller removals.
Which stocks are at stake
Washington's Section 120(f) application identifies thirteen ESU and DPS designations subject to predation within the affected reach. Named stocks include Lower Columbia River Chinook, coho, and steelhead; Columbia River chum; Southern DPS eulachon; Snake River spring/summer Chinook; and Upper Columbia River spring Chinook, which is listed as endangered rather than threatened.
The figure "89%" circulates widely, including in congressional testimony, as though it meant an 89% chance that Willamette winter steelhead go extinct. It does not. ODFW's population viability analysis found an 89% probability that at least one of three modeled winter steelhead populations reaches quasi-extinction within 100 years under 2017 predation levels. The highest single-population figure in the study is 64.4%, for the North Santiam. The precise version is still a striking number, and using it correctly is a credibility asset rather than a cost.
Willamette Falls: the closest thing to a controlled experiment
In 2017 only 512 adult winter steelhead crossed Willamette Falls, down from roughly 25,000 in 1974. Sea lions were taking somewhere between a quarter and a third of that return — agencies report 20–25% and 33% from different analyses, so the honest phrasing is "roughly a quarter to a third."
Thirty animals were removed. By 2023 and 2024 predation had fallen to under 2% of returning adults, modeled quasi-extinction probability dropped to under 10%, and the 2024 run was the strongest since 2004. Ocean conditions also improved over that period, so this is a strong association rather than a clean causal isolate — but it is the best before-and-after the system has.
A break in the record after 2018
Monitoring protocol changed. From 2024, sampling only occurred when at least twenty pinnipeds were present, producing 157 one-hour spring observation periods against roughly 3,000 hours in 2010 and 2015. In fall 2024 the trigger was never met and no predation monitoring was conducted at all.
That is itself meaningful — it reflects how far counts have fallen — but it means recent point estimates rest on a small fraction of the effort behind the mid-2010s figures. Plotting 2002–2018 and 2024–2025 as a single continuous trend without noting the methodological break is not defensible.
What is being used right now, and what the record says about each
Four categories of intervention have been tried on this river. One works and has worked for twenty years. Two have measurably failed. The fourth works but is expensive, legally constrained, and its own reviewers describe it as falling short.
1. Sea Lion Exclusion Devices — the one unambiguous success
"SLEDs continue to be effective at keeping pinnipeds out of the fishways, as none were observed in fishways during this reporting period." — USACE, 2024 Bonneville pinniped report
Metal grating installed in front of the fish ladder entrances, first deployed at Powerhouse Two in 2005 and Powerhouse One in 2006, and now present at all eight fishway entrances. Sixteen floating orifice gates at Powerhouse Two carry smaller exclusion grating. In 2016 the Corps added sea lion incursion barriers after Stellers were observed climbing over the floating gates; there were no further sightings.
The Corps' 2016 field report describes grates with rounded edges — for fish safety and hydraulic efficiency — creating slots roughly 15.4 inches wide. Effectiveness is documented across the entire monitoring record, with no reported breach, no bypass attempt of note, and no habituation. There is even a natural experiment: in fall 2015, when devices were not installed, pinnipeds did access the fishway.
The bar spacing figure of 38–40 cm that circulates in secondary sources is not published in any USACE engineering document we could locate. The Corps' own reports give no bar spacing, bar diameter, materials specification, or panel dimensions. Any design work should obtain drawings directly from USACE Portland District or the Fish Passage Operations and Maintenance coordination team rather than relying on a repeated number.
What SLEDs do not do
They close eight apertures. They do not enclose the tailrace, where essentially all measured predation occurs. Animals also bypass the dam entirely through the navigation lock — the Corps documents pinnipeds transiting to the forebay, with some California sea lions resident in Bonneville Reservoir between Bonneville and The Dalles for multiple years. And predation continues at Tanner Creek, a salmonid spawning tributary immediately downstream, which receives point-count monitoring and no exclusion at all.
This is the gap the pinch-point proposal addresses. It is an inference from the Corps' own documented facts rather than an agency statement in those words, and it is presented here as such.
2. Hazing and acoustic deterrence — measured, and measured to fail
Since 2002 managers have deployed cracker shells, underwater seal bombs, acoustic harassment devices, boat chase, rubber buckshot, and paintballs. The scale is substantial: 4,956 deterrence devices in the 2017 season; 5,602 cracker shells and 2,708 seal bombs across 642 hazing events in the 2017–2020 authorization period; 9,867 pyrotechnic rounds in 2024 alone.
The definitive study is Tidwell and colleagues, published in Frontiers in Conservation Science in 2021 — 3,289 behavioral observations across 28 sampling days, using interval scan sampling in three tailrace zones. Its findings are unusually clean:
- Active hazing produced a predicted 31% decrease in the number of Stellers present.
- The effect did not persist. Behavior returned to baseline once hazing stopped.
- Vigilance fell by nearly 25% with each subsequent exposure.
- With each application, animals became less likely to travel away and more likely to simply keep foraging.
- After three months of hazing, there was no measurable difference in responsiveness from the first day.
The authors' conclusion: hazing "failed to repel individuals from the dam and does not reduce SSL foraging behavior." The Corps concurs, describing functionality as minimal and noting the animals maintained vigilance levels comparable to baseline. Acoustic devices installed at fishway entrances in 2010 failed outright — pinnipeds were observed swimming and eating within twenty feet of them — and researchers recommended discontinuing their use that year.
Two findings from the wider literature explain the pattern. Götz and Janik (2010) found that in captive trials with food present, seals habituated quickly to all sound types — food motivation destroys deterrence, which is precisely the condition below a dam. Schakner and Blumstein (2013) add that motivation to depredate increases exposure rate, which accelerates habituation, and that habituation is fastest when exposures are closely spaced. A permanently installed deterrent is the highest-frequency, lowest-variability stimulus available: the fastest possible habituation schedule.
A physical barrier is not subject to this. It does not ask the animal to change its behavior; it removes the option.
3. Trapping and relocation — the cleanest failure in the record
ODFW ran a controlled test at Willamette Falls in 2018. Ten California sea lions were captured and released at a beach south of Newport, roughly 210 miles away.
One animal was relocated twice and returned both times. ODFW's Dr. Shaun Clements: "Clearly our experience on the Willamette River this year demonstrated the futility of relocation." Site fidelity is the governing fact — roughly 90% of individually identifiable Stellers remain at the dam eight to ten months annually, and a 2003–2007 tracking study found all fourteen tagged California sea lions that had used Bonneville or Willamette Falls returned to those areas.
4. Lethal removal — effective, constrained, and contested
The Endangered Salmon Predation Prevention Act, signed 18 December 2018, replaced Section 120(f) of the Marine Mammal Protection Act. It shifted authority from individually identifiable animals to an area basis, extended coverage to Steller sea lions, extended the geography upriver of River Mile 112 and into tributaries, and added tribal co-applicants. NOAA authorized removals in August 2020; the permit was renewed in August 2025 and runs to August 2030.
Authorization is held by Washington, Oregon, and Idaho; the Nez Perce Tribe, Umatilla, Warm Springs, Yakama Nation, Grand Ronde, and Siletz; and the Columbia River Inter-Tribal Fish Commission. Five-year ceilings are 540 California and 176 Steller sea lions, capped at 10% of Potential Biological Removal.
| Season | Steller | California | Total |
|---|---|---|---|
| Fall 2020 – Spring 2021 | 20 | 23 | 43 |
| Fall 2021 – Spring 2022 | 33 | 15 | 48 |
| Fall 2022 – Spring 2023 | 6 | 22 | 28 |
| Fall 2023 – Spring 2024 | 36 | 30 | 66 |
| Fall 2024 – Spring 2025 | 19 | 26 | 45 |
| Total | 114 | 116 | 230 |
Including the earlier authority and Willamette Falls, 517 animals have been removed since 2008. The measured response at Bonneville is real: Steller sea-lion-days fell 91% between fall 2020 and fall 2024, mean counts fell 90%, and peak counts fell 84%.
Modeled benefit varies by method and window, and should always be presented as a range: roughly 42,000 to 52,000 salmonids under agent-based and sea-lion-days models for the animals removed under 120(f), against agency and tribal figures of 50,000 to more than 110,000 cumulative since 2008. These are projections of counterfactual consumption, not counted fish.
Removal utilization sits at roughly a third of the permitted ceiling — 230 animals against 716 authorized. The Northwest Power and Conservation Council's March 2026 briefing states the program has "not been particularly effective at achieving the overall goal of balancing the requirements of the ESA and MMPA," and NOAA's own staff describe it as having "done good at a small scale in certain areas" while "still kind of falling short."
Managers identify a specific operational bind: most trappable haul-out sites are in the mainstem Columbia, but the take conditions attached to those locations do not permit removal there. This is a permit-condition and location mismatch, not a coverage gap — the 120(f) authorization plainly includes mainstem River Mile 112 to 292. Managers have asked Congress for a legislative fix.
The gap this leaves
Agencies have stated the shortfalls on the record. Sea lions forage across the other 140 miles of the lower river and in tributaries, where an animal at Astoria has roughly a 7% chance of ever reaching the dam — meaning the great majority of lower-river animals are never encountered by a Bonneville-focused program. Nighttime predation is largely unmeasured and acknowledged as underestimated. Bioenergetic models estimate food requirements, not consumption, and do not account for predation away from the dam. And after removal, no mechanism prevents recruitment of new individuals into the same learned behavior.
A structure that occupies the ambush lane does not require an observer, a permit for each animal, a boat crew, or an annual appropriation. That is the case for exclusion as a complement to what already exists.
How salmon actually behave below a dam
The passage literature answers the question most often raised against this proposal — will salmon use the structure — and it answers it decisively. It also identifies the real failure mode, which is not size.
Keefer, Jepson, Clabough and Caudill tracked 26,886 radio-tagged adult Chinook, sockeye and steelhead across eight Columbia and Snake River dams between 1996 and 2014. It is among the largest passage datasets ever assembled, and it is the foundation of this page.
Salmon will use engineered passage
Run-specific efficiencies are consistently high: 0.975 for spring–summer Chinook, 0.933 for fall Chinook, 0.986 for sockeye, 0.979 for steelhead. Whether adult salmonids will physically enter and transit a man-made structure is settled. They do, reliably, at some of the highest rates recorded for any migratory species.
Note that attraction and entrance efficiency are both higher than overall passage efficiency at these dams. The bottleneck is inside or after the ladder, not in finding it — an important corrective to the assumption that fish struggle to locate passage.
The exposure window is the tailrace, not the ladder
This is the single most load-bearing finding for siting a predator exclusion structure. Snake River Chinook spend three to four times longer in tailraces than inside fishways. Median tailrace passage time is 2.2 to 2.7 hours; a fish making a typical run at all eight dams accumulates roughly 136 hours — 5.7 days — from first fishway entry to final passage.
During that time, fish are not sitting still. The median fish makes three to eight fishway approaches and one to three entries per dam, and some make dozens. Searching is normal behavior, not pathology. But every hour of searching is an hour spent in open water alongside a predator that has learned exactly where to wait.
Keefer et al. report that roughly 30–62% of adults exited into the tailrace at least once at Bonneville and The Dalles, and 56–85% at John Day. This is fishway fallout — entering a ladder, turning around, and dropping back into the tailrace. It is not fallback over the dam, which is a separate and separately measured metric. The paper does not report dam fallback at all. Material that labels this figure "fallback" is wrong, and it is the sort of error that costs a technical document its standing.
Fish orient along banks — and so do predators
At each dam, the fishway openings with the highest discharge volume were located adjacent to shorelines and at the ends of powerhouses, taking advantage of the tendency of migrating adult salmonids to orient along river banks. Keefer, Jepson, Clabough & Caudill, PLOS ONE, 2021
This explains the geometry of the whole problem. Passage engineers deliberately sited entrances where fish already concentrate. Sea lions independently discovered the same locations. Predator and prey are concentrated at the same points by the same hydraulic logic — which is exactly why a structure placed there intercepts the fish the predators target without needing to span open water.
Attraction and siting, not scale
Bunt, Castro-Santos and Haro (2012) analyzed 101 records across 26 species and found attraction efficiency varying enormously by fishway type — mean 77% for pool-and-weir, 63% vertical-slot, 61% Denil, 48% nature-like — with ranges reaching 0 to 100% for three of four types. Critically, they found mean passage efficiency was inversely related to mean attraction efficiency by type, establishing attraction as an independent and highly variable failure mode.
Williams and colleagues, in the same journal issue, put it plainly: "Even well designed fish ladders or nature-like bypass channels for upstream migrants, even those with good attraction flows, will fail if incorrectly sited."
Noonan, Grant and Jackson (2012) found mean upstream passage efficiency of 61.7% for salmonids across 65 studies — against just 21.1% for non-salmonids. Their identified predictors were fish order, fishway type and fishway length. Attraction flow is often attributed to this paper as the dominant failure cause; it is not in it, and Bunt and Williams are the correct citations for that claim. Hershey's 2021 meta-analysis of 252 estimates has since superseded Noonan, finding no evidence that any particular design is favorable.
Design constraints this literature imposes
Two findings shape what a pinch-point structure can look like:
No overhead cover, no abrupt flow acceleration
Kemp, Gessel and Williams ran flume experiments at McNary Dam in which fish passing through encountered a choice of route. Direct observation showed individuals "elicited strong avoidance behaviour when they encountered areas where hydraulic conditions changed rapidly or were covered." Their companion work found both initial channel selection and subsequent rejection was higher for a constricted channel, with rejection probably due to fish detecting rapidly accelerating flow.
The Kemp work is on juvenile downstream migrants, not returning adults. Using it to predict adult behavior at a fishway entrance is an extrapolation across life stage and migration direction, and a NOAA reviewer will flag it. What it legitimately establishes is that salmonids detect and actively avoid overhead cover and sharp velocity gradients — a real design constraint on any in-water structure, and the reason a pinch-point gate must be an open-topped gate rather than a tunnel or funnel.
Structure geometry is not a reliable steering tool
Swanson, Tullos and Goodwin (2020) tested juvenile Chinook against a floating guidance structure at two deployment angles. Fish did respond to the wall with behavior changes — but did not distinguish between the two angles, with differences "potentially too small for the fish to produce a behavioural response." The authors suggest any flow contraction or visual cue may produce similar responses.
This is a partial-to-negative result and is presented as one. It supports the claim that fish detect floating structures. It does not support the claim that geometry can be tuned to steer them. A design that depends on precise hydraulic steering is a design betting on something the literature has not demonstrated.
What causes fish to abandon a passage lane
Keefer et al. list the drivers of fallout explicitly: confusing hydraulic cues, water temperature gradients, time of day, olfactory cues, misdirected attraction to spillway or turbine discharge, and high spillway discharge with the resulting tailrace turbulence. Their management conclusion is that "modifications that reduce fish exits from fishways into tailraces should be considered by CRB dam managers."
A structure sited on the dominant upstream flow line — where the strongest migratory cue already is — competes with none of these. It occupies the lane the fish are already following rather than asking them to choose a new one.
"Thalweg following" is not language used in this literature and does not appear in the Keefer work. The sourced and defensible phrasing is bank and shoreline orientation, quoted above.
Exclusion devices in practice, worldwide
Pinniped exclusion is not a speculative technology. It is deployed on three continents, in fisheries and at fish passage facilities, with a documented record of successes and instructive failures. The failures are the more useful half.
The central question any reviewer will ask is whether a new exclusion structure is a speculative idea or an application of something already working. The answer is the latter, and the evidence spans four continents and three regulatory regimes.
The principle at national scale: excluder grids in U.S. fisheries
The clearest proof that a rigid excluder grid can be engineered, mandated, and deployed across an entire American fleet is the Turtle Excluder Device. TEDs are federally required in U.S. Atlantic and Gulf of Mexico shrimp trawls, enforced by NOAA Fisheries with published construction specifications, approved designs, and dockside compliance inspection. The mechanism is identical to what is proposed here: a rigid bar grid sized to pass the target catch while physically blocking an animal that must not be caught, with a designed escape route for the excluded animal.
What makes TEDs the strongest available precedent is not the biology — it is the institutional record. A grid-based excluder went from concept to contested trials to federal rulemaking to routine fleet-wide compliance. That path exists. It has been walked in U.S. waters, under U.S. agencies, against significant early industry resistance and against the same objections raised here: that it would not work, that it would cost target catch, that fishermen would not adopt it.
NOAA publishes an effectiveness figure for properly installed TEDs and a specification history for approved designs. Those figures are not reproduced here because they have not yet been verified against the primary rule text and current NOAA guidance. Do not quote a TED effectiveness percentage from this page until that check is complete.
There are no sea lion exclusion devices on the U.S. East Coast, for the simple reason that there are no sea lions in the Atlantic. California and Steller sea lions are Pacific species — California sea lions range from southeast Alaska to central Mexico, Stellers around the North Pacific rim. NOAA's guidance for New England and the Mid-Atlantic lists two common pinnipeds: gray and harbor seals.
The absence runs deeper than species range. The Federal Interagency Nature-like Fishway Passage Design Guidelines for Atlantic Coast Diadromous Fishes does not mention seals, pinnipeds, or marine mammals anywhere, and the 2019 Gulf of Maine Atlantic Salmon Recovery Plan mentions seals once in a historical aside and recommends no seal management action.
This is worth stating rather than hiding, because the pinniped exclusion precedent set is genuinely strong without it — Pacific, Baltic, Australasian and Scottish — and because a document that gets an easy geographic fact wrong invites doubt about the harder claims it makes.
Bonneville Dam — the anchor case
Covered in detail on the current methods page, and it remains the strongest precedent in existence: metal grating at eight fishway entrances, installed from 2005, with no documented breach across twenty years and no habituation. It is the only physical pinniped exclusion structure at a fish passage facility anywhere in the world with sustained, agency-verified success.
It is also the source of the mechanism this project extends. Nothing in the pinch-point concept depends on an unproven principle — only on applying a demonstrated one at geometry it has not yet been applied to.
Ballard Locks — the failure that defines the boundary
Barrier nets at the Ballard Locks in Seattle "were unsuccessful at decreasing California sea lion predation on steelhead. The sea lions moved to other areas, including Lake Washington, above the locks, and continued to feed on steelhead." A 2024 technical workshop reached the same conclusion: the net did not reduce predation; predation shifted farther downstream.
The Ballard record is the most important cautionary case in this field, and any serious proposal has to engage with it. The Lake Washington winter steelhead run fell from roughly 2,500 fish in 1983 to 184 by 1992 and under 50 by 2000, with sea lions estimated to take 42–65% of the run between 1986 and 1992. WDFW's assessment is that the run "was effectively destroyed."
Everything was tried. Seal bombs worked for one year before animals learned to dive and swim unpredictably. Acoustic devices habituated within eight weeks to two years, and functioned as dinner bells. Cracker shells produced a median time away of four minutes. Killer whale playback produced no reaction; a fiberglass orca produced no reaction. Lithium chloride taste aversion taught the animals to avoid humans in yellow suits rather than the fish. Relocation failed on site fidelity. Only removal of specific named individuals changed anything.
A barrier net across open water is a fundamentally different object from a gate at an aperture. The Ballard net failed because there was somewhere else to go. That distinction is the load-bearing one for this entire project, and it is stated here rather than buried.
Baltic Sea — pontoon trap exclusion devices
Calamnius, Lundin, Fjälling and Königson (2018), in PLOS ONE, tested a seal exclusion device on Baltic pontoon traps: an 800 × 800 mm aluminium frame strung with 3 mm Dyneema. Traps fitted with the device caught heavier salmon with no loss of catch — a clean result on the fish side.
But seals destroyed three of the four devices during the experiment. Structural survivability under sustained pinniped attention is a real engineering constraint, not a theoretical one, and it should be designed for explicitly.
Trawl fisheries — the best efficacy dataset anywhere
Sea lion and seal exclusion devices in trawl nets are mature technology with published, quantitative outcomes. The design lessons are counter-intuitive and directly transferable.
New Zealand squid SLED
Mandatory in the SQU6T fishery since January 2022. Maximum 23 cm bar spacing, grid at roughly 45°, minimum 20 mm stainless bar, triangular top escape opening. A 2007 audit found 38% of 60 devices failed compliance — most commonly because the escape opening had gone slack. Maintenance discipline is part of the design.
Australia — the configuration lesson
Early devices doubled bycatch. Top-hatch forward-facing worked: 3.1% of shots against 20.7%. Bottom-hatch made things worse — 12.3% against 3.9% — because the opening functioned as an entrance. Midwater nets with an open device: 48% seal survival against 0% without.
Falkland Islands
Iriarte and Winter (2025) report fresh mortalities falling from 0.019 to 0.002 per trawl — roughly a 90% reduction.
New Zealand hoki — the warning
48% of target fish struck the grid and 14% became lodged, because hoki have poor directional swimming capability. Target species behavior determines feasibility, and it is the usual thing that kills these designs.
The Australian work also produced the constraint that most directly shapes bar spacing: spacing was set at 20 cm after juveniles passed through a 30 cm grid. Juveniles set the binding constraint, not adults. Any Columbia design must be specified against the smallest fish it must pass, including jacks and Pacific lamprey, not the largest.
The hoki failure connects directly to the passage literature. Kemp's finding that salmonids detect and avoid overhead cover and sharp velocity gradients is exactly the class of constraint that sank the hoki device. A design that ignores target-species swimming behavior fails regardless of how well it excludes predators.
Meyer and colleagues argued in PNAS (2017) that exclusion devices "have paradoxically contributed to ongoing decline rather than recovery" by concealing cryptic mortality — animals excluded but injured or drowned outside the observed catch. Roberts and colleagues rebutted in 2018, with a counter-reply following. Reported exit probability was 0.88 in midwater but only 0.57 in bottom trawl. Anyone building around this acronym should be able to discuss this exchange.
Aquaculture exclusion — regulatory default, thin evidence
Physical exclusion is the standard approach at Scottish and Norwegian farm sites, largely because acoustic alternatives failed: Coram and colleagues, reporting to the Scottish Government, concluded "no authors have shown convincing evidence for the long-term efficacy of acoustic deterrent devices." In British Columbia, device power escalated from 135–140 dB to 195–220 dB and still failed, with some of the largest harbour seal kills at salmon farms occurring while the devices were in use.
But the exclusion side is not well evidenced either. Northridge, Coram and Gordon (2013) state flatly that "there is no published research to support" the claim that net tensioning prevents damage. The operative mechanism is worth knowing: most depredation occurs through slack mesh from underneath without breaching the net, meaning tension rather than material strength is the governing variable. Predator nets also drown wildlife — one historical installation drowned thirty seals — which is why Scotland largely abandoned them.
A NOAA report documents a barrier net at a Rich Passage aquaculture site causing the entanglement and death of two California sea lions. Under the Marine Mammal Protection Act that is a take. Any in-water structure proposed for the Columbia must be designed as a rigid bar array rather than mesh or netting, specifically to eliminate entanglement risk, and that design choice should be foregrounded in any permit application rather than defended after the fact.
What the precedent set actually establishes
Read together, these cases draw a clear line. Physical exclusion works where the geometry is constricted and the barrier can fully close the aperture — a fishway entrance, a lock, a trap mouth, a net opening. It fails where it spans open water and the animal has somewhere else to go.
Sea lions were never removed from Bonneville by the SLEDs. They stayed in the tailrace and kept eating salmon; only the location of predation changed. That is the honest frame for this whole project: exclusion redistributes predation rather than eliminating it, and the case for it rests on redistributing predation away from the points where the fish are most concentrated and most vulnerable.
How many salmon could pinch-point exclusion actually save?
The honest answer is a scenario, not a number. This page shows the inputs, the arithmetic, and the places where the evidence runs out — because a transparent estimate survives review and a confident one does not.
First, the two numbers that must never be added together
Almost every credibility failure on this topic comes from mixing two incompatible bodies of evidence. Here is the reconciliation.
Tailrace observation: 1–6% of run
Trained observers record surface predation events at Bonneville using a stratified random sampling design with bootstrap expansion. Yields 1,010 to 10,859 fish per year across 2002–2018.
Biased low. The Corps states plainly that all consumption estimates "should be considered minimum estimates." Consumption below the surface is never seen — Stellers have surfaced with fish already partially swallowed. Pacific lamprey are nearly invisible to the method: four Steller stomachs contained 46 lamprey in a year when surface observers recorded zero lamprey events. Observation hours fell from roughly 3,000 in 2010 and 2015 to 157 in 2024.
Reach telemetry: 20–45% of run
Wargo Rub and colleagues (2019) tracked tagged adult spring Chinook from the estuary to Bonneville, 234 river kilometres, across 2010–2015. They estimated 51,751 to 224,705 salmon dying annually in that reach from causes other than harvest.
Biased high as a predation figure. It attributes all non-harvest, non-arrival loss to predation, so it necessarily absorbs tagging and handling mortality, disease, straying into tributaries, unreported harvest, and thermal stress. The authors identify pinniped predation as the most likely dominant source — a modeled causal inference, not a measurement.
They differ because they measure different things over different distances. The tailrace figure covers a few hundred metres of water at one dam. The reach figure covers 234 kilometres including the estuary, the Astoria haul-outs, and every tributary confluence. One bridging study — Stansell et al. (2012) — compared both methods at the dam and found surface observation captured roughly half of the telemetry-based estimate.
Between 20% and 45% of returning spring Chinook disappear between the ocean and Bonneville Dam without being harvested. Peer-reviewed analysis attributes most of that loss to pinniped predation, though the estimate also absorbs other unmeasured losses. Separately, direct observation at the dam itself accounts for 1–6% of the run, and the Corps describes that as a floor.
Two findings from that study worth more attention than they get
- Odds of salmon survival decreased 32% (95% CI 6–51%) for every additional 467 sea lions present. Predator abundance is doing measurable work.
- Odds of survival increased 32% (95% CI 8–61%) per 1.5-unit increase in log American shad abundance. Alternative prey buffers salmon — which means the value of an exclusion structure is partly a function of what else is available to eat at that location and in that week. This is underused in the policy conversation and is directly relevant to pinch-point siting.
Where the evidence runs out: tributary mouths
There is no published fish-per-year or percent-of-run predation estimate for any Columbia tributary mouth other than Willamette Falls. None. Washington's 120(f) application documents confirmed California sea lion presence in the Sandy, Clackamas, Cowlitz, Kalama, Lewis and Washougal rivers, characterized as typically one to two animals making daily foraging migrations, with an exception of six animals at river mile 19 of the Clackamas in 2017. Agencies clearly regard these as predation locations — they were written into the removal authorization — but they have not quantified the loss.
Any tributary-mouth number on this site is therefore an explicitly labeled scenario with visible assumptions, not a finding. That is what the calculator below is for.
Scenario calculator
Move the inputs. The arithmetic is shown so anyone can check it or substitute their own assumptions. Defaults are set conservatively, using the observation-based evidence rather than the higher reach estimates.
Single pinch-point exclusion gate, annual fish protected
It is not a prediction. It multiplies four numbers, three of which are sourced and one of which — the share of predation falling inside the protected lane — has no published basis at any Columbia tributary mouth. It also assumes protected fish are not simply taken a hundred metres downstream, which is precisely what happened with the Ballard barrier nets. Displacement is the central unresolved question, and no honest model can assume it away.
Run size baselines
| Year | Spring Chinook | Steelhead |
|---|---|---|
| 2020 | 77,458 | 111,192 |
| 2021 | 87,232 | 70,502 |
| 2022 | 173,733 | 123,345 |
| 2023 | 136,784 | 114,104 |
| 2024 | 109,314 | 181,896 |
| 2025 | 146,061 | 156,832 |
| 10-yr average | 119,407 | 136,764 |
USACE passage figures do not match DART. The Corps lists 79,241 spring Chinook for 2024; DART lists 109,314. The date windows and count definitions differ. If you compute a percentage, take the numerator and denominator from the same document and say which one.
Cost framing
A per-fish dollar value is the weakest-sourced element of any economic case here, and it should be presented as a range rather than a figure. The most-cited per-fish numbers — roughly $1,176 for in-river recreational, $281 ocean recreational, $151 commercial — come from 2013 California data, not the Columbia. The canonical Columbia Basin economic study, from the Independent Economic Analysis Board in 2005, reports approximately $142 million annually in personal income to West Coast communities from basin anadromous production but offers no per-fish valuation and is now twenty-one years old.
On the management-cost side, a figure of roughly $38,000 per animal removed and $203 per salmon saved has circulated since December 2025. It originates in a letter from Representative Marie Gluesenkamp Perez to the Commerce Secretary, not in any agency publication, and no calculation basis has been disclosed. It should be attributed that way or not used.
What can be said with confidence is structural rather than numerical: a fixed exclusion structure has a capital cost and a seasonal deployment cost, and does not require an annual permit allocation, a boat crew, an observer program, or a per-animal authorization. That is a different cost curve from the one the current program is on, and it is the honest form of the economic argument until better per-fish figures exist.
Haul-out access barriers: the idea, and everything working against it
The concept is that fencing between a high-use beach and the water makes access inconvenient enough to relocate animals. This page states the idea, then gives the contrary evidence in full — because the contrary evidence is substantial and agency readers already know it.
No published study evaluates fencing between a haul-out and the water as a means of relocating pinnipeds. Not a weak study — none. Every comparable intervention on record produced displacement of metres rather than departure, and the one agency report commissioned specifically to assess shoreline fencing predicted relocation rather than abandonment. This page is included because the idea is worth testing, not because the evidence supports it.
The idea
Sea lions haul out where access is easy. A beach with a gentle gradient and an unobstructed waterline is cheap to use; a beach requiring a long detour around a barrier is expensive. If the animal is optimizing for effort — and pinnipeds demonstrably do optimize haul-out selection — then raising the access cost at the highest-use sites nearest to critical predation chokepoints should shift the distribution toward sites farther away.
The appeal is that it requires no permit per animal, no crew, no ongoing operation, and no lethal action. It is a one-time capital intervention that works while nobody is watching, which is exactly the property that makes SLEDs succeed where hazing fails.
The one relevant Columbia precedent, and what it showed
In 2010 concrete blocks were placed along the Powerhouse Two tailrace shoreline concrete apron at Cascades Island. The result, as recorded by CRITFC: the blocks "seem to have encouraged the seals and sea lions to use Tower Island instead of Cascade Island."
This cuts both ways, and both halves matter.
A passive physical obstruction changed where sea lions hauled out. No habituation, no ongoing labour, no crew. The mechanism is real: make a site inconvenient and animals use a different one.
They moved to the next island. Denied one haul-out, animals took the nearest alternative — still in the tailrace, still in position to forage. The predation did not go anywhere.
The displacement record is consistent and it is not encouraging
Astoria East End Mooring Basin
Peak occupancy around 3,000 animals and more than $2 million in cumulative dock damage. The full list of what was tried: chicken-wire fencing, plastic and vinyl fences, and streamers — all ignored. Beach balls — brief scatter, then ignored. An electrified mat in April 2015 — failed, with a sea lion resting on it. A fiberglass orca in June 2015 — capsized and swamped, though counts dropped from 400–500 to about 200 for one night. Air dancers in 2016 — reported as working, with no durability follow-up. An electrified mat on one finger of P Float in June 2016 — worked on that finger.
When animals returned in late summer 2016, they occupied the main dock of P Float "and all the fingers except for one with the dock deterrence system installed." Textbook within-structure displacement — recorded against the vendor's own commercial interest.
Port Executive Director Jim Knight's summary: "Nothing works for long or it costs too much." NOAA's Robert Anderson: "Once you apply whatever deterrence method you do, if you walk away they're going to come right back."
Ballard and Shilshole, Seattle — a displacement chain roughly one kilometre long
In 2021, "fencing and other physical deterrents encouraged the group to relocate" — to the rock jetty at Shilshole Bay Marina. From there they moved onto the marina's guest docks, reaching roughly 120 animals by May 2025. Three steps, no departure.
Newport, Oregon — total platform removal as a natural experiment
Winter storms in 2022–23 destroyed all six sea lion dock sections. The animals joined others on the port's breakwater rocks nearby. Removing the structure entirely produced displacement of tens of metres.
La Jolla Cove — the report commissioned to evaluate exactly this idea
Hanan (2016), for the City of San Diego, assessed four-to-six-foot fencing with PVC rollers and low-voltage livestock fencing. Its conclusions are the most on-point sentences that exist on this question:
Sea lions excluded from one area are likely to relocate to other areas which may have unintended consequences to create new problems in the new areas… They are not likely to abandon the LJC area entirely. Hanan, D.A. (2016), California Sea Lion Observations at La Jolla Cove, for the City of San Diego
The California Coastal Commission approved two roughly three-foot barred gates at La Jolla in December 2017. The staff report was candid that "there is the chance that the sea lions may still gain access." No published monitoring result appears to be available.
The behavioral evidence points the same direction
- Kucey and Trites (2006) found disturbance recovery to 50% of pre-disturbance counts within under six hours, and 75–100% within one to six days, with no documented case of permanent site abandonment.
- Suryan and Harvey (1999) found seals surviving repeated harassment become more tolerant, allowing closer approaches. Sites harassed on 71% or more of survey days across two pupping seasons remained occupied.
- London and colleagues (2012) found harbor seals increased haul-out probability by 40–50% when mammal-eating killer whales were present. Even a real predator does not clear a haul-out.
- The only study ever designed to shift a haul-out — Calambokidis and colleagues at Dosewallips State Park, 1990 — hedged hard: "Seals may not readily adapt to new structures," cost over $100,000 in 1990 dollars, barriers recommended only after seals had begun using an alternative, and the authors said it was "impossible to provide assurance of success."
That 36-year-old, unreplicated report is close to the entire evidence base for the proposition that fencing relocates pinnipeds. It is not enough to build a programme on.
So what would make this worth testing anyway
Two things keep the idea alive despite the record above.
The first is that the displacement finding is not automatically fatal — it is fatal only if the animal lands somewhere equally good. Every documented case involves a marina, a dock, or a cove where an equivalent haul-out sat a few metres away. Nobody has tested a barrier at a site where the nearest alternative is genuinely distant from the foraging chokepoint. That is a different experiment from the ones on record, and its result is unknown rather than known-negative.
The second is that a small number of individuals do a disproportionate share of the predation. NMFS reported to Congress in 1999 that impact on salmonids "is likely due to opportunistic behavior by certain individual pinnipeds that have learned to exploit situations where salmonids are concentrated," and later work frames these animals as super-spreaders whose behavior transmits through California sea lion social networks. If the relevant population at a given site is a handful of animals rather than a herd, the bar for a useful intervention is lower.
A defensible test would need: a site where the nearest alternative haul-out is a measurable distance from the predation chokepoint, not adjacent to it; before-and-after counts at both the treated site and every candidate alternative within a defined radius, so displacement is measured rather than assumed; a predation-rate measurement at the chokepoint rather than a haul-out count, because moving animals is not the objective — moving predation is; and a pre-registered duration long enough to catch the return that every prior case eventually recorded.
Absent that design, a barrier project would be repeating an experiment that has already been run several times with a consistent answer.
Where this sits in the proposal
Haul-out barriers are the weakest element of this project and are ranked accordingly. The in-water pinch-point gate rests on a mechanism with twenty years of documented success on this river; the fencing concept rests on one displacement observation from 2010 and a set of analogues that mostly went the other way. Presenting them as equally supported would undermine both.
If the fencing concept advances, it advances as a monitored pilot with a pre-registered design and a published result — including a negative one.
Note on sourcing quality. Several dock-exclusion cases trace to a single vendor's marketing material. There is essentially no independent or peer-reviewed evaluation of dock-level pinniped exclusion, no published duration-of-effect data, and no controls. A widely repeated account of a guard dog deployed at Astoria could not be sourced and is deliberately omitted here.
About
This project started on the water, not in a library. The citations came later, because the argument had to survive people who would check them.
Why a fisherman is proposing engineering
Anyone who has spent enough seasons on the lower Columbia has watched the same thing happen. The runs get thinner. The water below the dam gets busier with animals that were not there twenty years ago. And the conversation about what to do about it splits into two camps that mostly talk past each other.
This project exists because there is a third option that neither camp spends much time on. The most effective predation intervention on this river is not lethal removal and it is not hazing — it is a set of metal grates bolted in front of eight fish ladder entrances in 2005, which have worked continuously ever since and which nobody argues about. They generate no litigation, no annual appropriation fight, and no permit ceiling. They just sit there and work.
The obvious question is why that approach stops at the ladder entrance.
The array is not meant to protect 145 miles of river. It is meant to remove the ambush advantage at a handful of high-loss chokepoints. At that scope — and only at that scope — the passage data supports it. Pinch-Point Salmon Exclusion Array, design summary
What this project is
A proposal for a seasonally deployed, open-topped bar gate that physically excludes California and Steller sea lions from a tributary-mouth or tailrace pinch point, while presenting returning salmon an unobstructed high-attraction passage lane on the river's natural flow line.
Every design choice answers a documented failure mode. The exclusion principle comes from the only device proven on this river. The open top exists because the passage literature shows salmonids avoid overhead cover and abrupt hydraulic transitions — this is a gate, not a pipe. The siting and flow alignment exist because attraction and placement, not size, govern whether fish commit to a structure. The seasonal, anchor-only build exists to keep the footprint reversible and the permitting path short: helical screw anchors with a tensioned cable spine to both banks, no fill, no streambed disturbance, out of the water more than six months a year.
The lead pilot candidate is the Cowlitz River mouth.
What this project is not
It is not a claim that sea lions are the primary cause of salmon decline. The hydrosystem, ocean conditions, and habitat loss are larger factors, and any document that pretends otherwise gets dismissed by the people who most need to read it.
It is not a replacement for existing management. Removals under Section 120(f) have measurably reduced pinniped presence at Bonneville. This is a complement aimed at the coverage gap that agencies have themselves identified — the tributary mouths and the open tailrace that fixed exclusion has never reached.
And it is not finished. The largest assumption in the model — how much of the local predation actually falls inside a gate's protected lane — has no published basis anywhere in this basin. That is what a pilot is for.
On the fishing
[Will: this is where your own story goes. A few things worth including, if they're true for you — the water you know best, how long you've been fishing it, what the runs looked like when you started against what they look like now, and the moment this stopped being something you complained about and started being something you worked on. People fund people. This page is the one that does that work.]
Contact
Will Robertson · will@billrobertsonnissan.com
Technical review, agency contact, and collaboration inquiries are all welcome — particularly from anyone who can correct something on this site. Supporting figures are drawn from NOAA Fisheries, USACE Bonneville pinniped reports, WDFW and ODFW, current through the 2025 season.