The Verification Venue · a beloved story, measured
The Wolves That Were Supposed to Change the River
In 1995 and 1996, 31 gray wolves were carried from Canada into Yellowstone. Two decades later a four-minute video, How Wolves Change Rivers, told 43 million people that those wolves narrowed the rivers and changed the land's physical geography. The spine of that story is real. Its magnitude is overstated, and its mechanism is contested — and that gap is the whole subject here.
This is not a debunking. Wolves came back; elk fell and moved; willow and aspen released in patches. The honest verdict is narrower and stranger than either the fan version or the sceptic version: real effect, overstated magnitude, contested mechanism. Below are two things you can operate. First, flip the famous cascade from the story to the evidence and watch each arrow keep its shape while it fills with the confounds the video left out. Then push a live model of the deepest result — why removing the wolves' effect need not bring the river back.
1 · Flip the cascade
The viral argument is a clean chain of arrows: wolves → fewer, warier elk → willow and aspen soar → beavers and birds return → the rivers themselves change. Press The evidence. No arrow disappears. Each one gets thinner and heavier — the certainty drains out, the caveats pour in, and the last arrow, the river itself, greys to contested.
The shape survives the scrutiny. That is the honest half the sceptics can miss: the elk did decline and redistribute, browsing on young aspen did fall sharply after 2000, willow and aspen are taller at many sites than they were in the 1990s. What does not survive is the magnitude (how much of it the wolves caused) and the last arrow (that wolves re-plumbed the rivers). Hold that distinction; the second instrument lives inside it.
2 · Why bringing back the predator need not bring back the world
Here is the sophisticated dismissal: "fine, the cascade is oversold — strip out the wolf hype and it's just fewer elk browsing, so less browsing must mean the willows come back." The deepest result says even that is too simple. In a controlled experiment, Marshall, Hobbs & Cooper (2013) fenced willows to remove browsing entirely — and the willows still did not recover to tall stature, unless the water table was also raised with simulated beaver dams. When beavers and their dams vanished, the water table dropped past a threshold, and the riparian system settled into a second stable state. It does not rewind.
The instrument below is the minimal maths of that idea — the canonical alternative-stable-state model (Scheffer's shallow-lake form). Willow-and-beaver vigor v is pushed up by a water-table support a (beaver dams) and pulled down by browsing b (elk; wolves lower it), with a positive feedback — established willow and dams hold the water that lets more willow establish. Drag the two levers and watch the state settle. Removing browsing at a low water table barely moves it. Only raising the water table jumps it — and lowering the water table again does not put it back.
Willow & beaver vigor v
0.13
state: suppressed
The lever that just moved
—
on the lower (degraded) branch
The lever Marshall, Hobbs & Cooper found is the real threshold. Push it past the fold and the valley jumps to tall willow.
Wolves lower this. On its own, at a low water table, it slides the willow up only a little — never across the barrier.
Start degraded: dams gone, elk browsing hard. Willow vigor sits near 0.13 — the low, suppressed branch. Walk the three steps and watch which lever actually crosses the threshold.
The equilibrium set is the S-shaped curve on the canvas: two solid stable branches with a dashed unstable one between them. Where the curve folds, the state has nowhere to sit but the other branch — so it jumps. The gap between the up-jump and the down-jump is hysteresis: the path back is not the path out. That is the reframing the whole controversy hides inside. Ecosystems can carry thresholds and memory, so a fix is not a rewind.
The check — every number recomputed in front of you
The green column is the cross-check, not a claim. The reintroduction counts and the model's fold points are recomputed from scratch; the field figures are quoted from the cited papers with their exact values and their uncertainty kept.
| quantity | from | value | recomputed |
|---|
The model is the canonical alternative-stable-state equation — not a fit to Yellowstone data, but the minimal maths of the water-table feedback Marshall, Hobbs & Cooper describe. Its bistability, its two fold points, and the fact that the browsing lever alone cannot cross them are all re-derived here: node research/the-wolves-that-changed-the-river/verify.mjs.
The honest boundary. Confirmed: 31 wolves reintroduced 1995–96 from Canada; the northern elk herd fell from about 17,000 to about 8,000; aspen browsing dropped sharply after 2000; willow and aspen are recovering at many (not all) sites. Overstated: how much of the elk decline and vegetation change the wolves caused. Contested-to-unsupported: that wolves changed the rivers' physical channels. Not settled as of 2026 — an active dispute, not a verdict.
What's confirmed, what's idealised, and the verification limits in full
Confirmed from the record. Fourteen wolves were released in 1995 (from Alberta) and seventeen in 1996 (from British Columbia) — 31 in all, from Canada — to a park whose last wolves were killed by 1926. The northern Yellowstone elk herd declined from roughly 17,000 in the mid-1990s to roughly 8,000 by the mid-2000s — a fall of about 53% (52.9% exactly). Browsing on young aspen fell from nearly every leader shoot in the late 1990s to a small minority a decade later, and willow, aspen and cottonwood are taller at many sites (Ripple & Beschta 2012). The video How Wolves Change Rivers was posted in 2014 (audio from George Monbiot's 2013 TED talk) and has surpassed 43 million views.
Overstated — the magnitude. The elk decline is multi-causal. Vucetich, Smith & Stahler (2005) found human harvest and climate (low rainfall, severe winters) were major drivers, and that much wolf predation was compensatory — many wolf-killed elk would likely have died anyway — so a model of harvest and climate alone reproduces most of the decline. Grizzly and cougar recovery and bison competition also press on elk. We deliberately do not print the widely-repeated figure that puts nearly all of the elk decline down to hunting and climate: we could not verify that exact percentage against the Vucetich 2005 primary text, so it is omitted rather than asserted. For vegetation, Brice, Larsen & MacNulty (2022) show that non-random sampling — preferentially measuring tall, released plants — exaggerated the apparent aspen cascade; the randomly-sampled picture is weaker. MacNulty and colleagues, and Ripple & Beschta, continue to dispute the size of the effect. This is a live debate; we crown no winner.
Contested to unsupported — the river. The last arrow, that wolves narrowed channels and changed the physical geography, is the weakest link. No study cleanly attributes Yellowstone channel change to wolves, and where river-channel change has been studied closely elsewhere, physical drivers (sediment supply, flood history) dominate. We flag this contested-to-unsupported — neither proven nor debunked, simply not established.
The beaver confound. Beaver colonies did rise on the northern range after wolves returned, but that rise is confounded: 129 beavers were reintroduced from the Gallatin National Forest between 1986 and 1999 — before, and independent of, the wolves — and vegetation was also recovering from the 1988 fires. The willows-bring-back-beavers arrow is real in direction and confounded in cause.
Idealised — the second instrument. The bistability model is the canonical Scheffer shallow-lake equation, dv/dt = a − b·v + r·vᵖ/(vᵖ+hᵖ) with r=1, h=1, p=4. These shape parameters are a free choice made to illustrate the mechanism; they are not fitted to willow data, and the numbers on the axes are model units, not metres or elk. What is faithful — and what Marshall, Hobbs & Cooper's controlled experiment actually shows — is the structure: removing browsing at a low water table does not restore tall willow, a water-table threshold does, and the system exhibits hysteresis (it does not retrace). The model is the idea made operable, not a claim about a specific river's numbers.
What would change this page. A clean, controlled attribution of Yellowstone channel narrowing to wolves would move the river arrow off "contested." A resolution of the Brice–MacNulty–Ripple dispute would let us state a magnitude. Neither exists as of 2026-07-09; until then the honest reading stays: real effect, overstated magnitude, contested mechanism.