Skip to content
NatureHQ

Ecologyphenomenon

Trophic cascades

The evidence for cascades is best in ponds and shores. The claims are loudest about national parks.

An effect that travels past the animal a predator eats and reaches the plants below. It happens, it is well demonstrated in small systems you can manipulate — and in the large systems where it is claimed loudest, it is hard to establish at all.

The idea is that a predator’s influence does not stop with its prey. Fewer or warier herbivores mean less grazing, and less grazing means more plants — so an effect starting at the top reaches the bottom, through links the predator never touches. In systems small enough to manipulate, this is not in doubt. Remove a predatory fish from a pond and the algae respond; exclude urchins from a patch of reef and the seaweed returns; take sea stars off a shore and mussels take the rock. The evidence is manipulations with controls, and it is good. The difficulty begins where the systems get large. There, cascades are inferred from what happened after a change nobody arranged, in a place with no counterpart to compare against, over decades during which many other things also changed. Yellowstone is the standard case and worth walking through carefully, because the popular version has compressed a contested chain of inference into a single sentence about rivers. Wolves returned in 1995. Elk numbers fell — from wolves, and also from human hunting, drought, bears and a severe winter. Some riparian vegetation recovered in some places. Beaver colonies increased. Willow, which the story requires, turned out on experimental test to need high water tables as well as reduced browsing — and those water tables had been maintained by beaver dams, and the beavers had been gone since long before the wolves were. Meanwhile a landscape-scale test of the fear-mediated version, in which elk avoid risky places and browsing shifts accordingly, did not find the redistribution the hypothesis predicts. None of this shows that wolves changed nothing. It shows a chain of six or seven links in which some are well evidenced, some are contested, and the confident version omits every step where the evidence runs out.

Early coverage · 69% complete · reviewed 2026-09-04

What this page covers

Demonstrated by manipulation in ponds, streams, intertidal and kelp systems. Asserted most confidently for large terrestrial systems, where manipulation is impossible.

Often confused with: A guaranteed chain reaction, when cascade strength varies enormously; Evidence that predators are good for ecosystems, which is a value judgement rather than a finding; A single mechanism, when killing prey and frightening prey are different pathways

Quick facts

Where it is demonstrated
Ponds, streams, shores, kelp — systems you can manipulate with controls
Where it is asserted
Large terrestrial systems, where no control exists
Two different pathways
Eating prey, and frightening it — they predict different patterns
The Yellowstone willows
Needed high water tables as well as less browsing

Strongest where the systems are smallest

Which is the opposite of where the claims are made.

Removing a predator sometimes changes vegetation two links down the chain. It is demonstrated in small manipulable systems and much harder to establish in large ones — where it is asserted most often.

Well supported

Good evidence backs this, though some details remain open.

Trophic cascades — indirect effects of predators on primary producers via herbivores — are demonstrated by manipulation in ponds, streams, intertidal and kelp systems. In large terrestrial systems, attribution is confounded by concurrent drivers and by the absence of replication or controls.

Who this applies to
Established in aquatic and intertidal systems; contested in large terrestrial ones.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Animalia, Plantae
Why we rate it this way, and what the caveats are
Well supportedModerate confidence

That cascades occur is not in doubt. How strong they are, and in which systems, varies enormously, and the large terrestrial cases carry the weakest evidence and the loudest claims.

How far it can be extended

Cascade strength varies with food-web complexity, productivity and the number of alternative pathways; results do not transfer between system types.

Caveats

  • Cascades are real; the dispute is about magnitude and attribution in particular systems, not about existence.
  • Aquatic systems with short food chains cascade more readily than complex terrestrial ones, so the best evidence comes from the least representative cases.

Where researchers disagree

  • Landscape-scale testing in Yellowstone did not find the pattern of browsing relief that a fear-mediated cascade requires.
  • Willow recovery in the same system proved limited by water tables as well as by browsing, implicating a driver unrelated to predators.

Still unanswered

  • What predicts cascade strength across system types, which remains one of the field’s central unresolved questions.

Last reviewed 2026-09-04

The evidence (3 studies)

Aleutian islands with sea otters had small urchins and thick kelp; islands without had large urchins and bare rock. Nobody assigned the otters — history did, by hunting them out unevenly.

Well supported

Good evidence backs this, though some details remain open.

Comparative surveys of Aleutian islands with and without recovered sea otter populations found contrasting sea urchin abundance and kelp extent, consistent with otter predation limiting grazing pressure. Otter presence was determined by the history of the fur trade rather than by experimental assignment, and effect strength varies geographically.

Who this applies to
Aleutian nearshore systems; the strength of the relationship varies by region.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Enhydra lutris, Strongylocentrotus, Macrocystis
Why we rate it this way, and what the caveats are
Well supportedModerate confidence

A strong natural experiment with a large and consistent contrast. It is not a manipulation, and the islands were not assigned, so other differences cannot be excluded.

How far it can be extended

Later work found substantial regional variation, and other drivers of urchin abundance operate elsewhere.

Caveats

  • Islands with and without otters may differ in other ways; a natural experiment cannot rule that out.
  • Not every kelp forest is structured this way, and treating the Aleutian result as a general rule about kelp is the common error.

Still unanswered

  • What accounts for the geographic variation in how strongly otters affect urchins and kelp.

Last reviewed 2026-09-04

The evidence (1 study)

Diagram

Islands with otters, islands without

A natural experiment: history did the assigning.

Islands with otters, islands withoutWith ottersFew, small urchinsKelp abundantWithout ottersMany, large urchinsKelp scarceNobody assigned the otters. The fur trade did, unevenly.A natural experiment: better than a correlation, weaker than a manipulation.
The same explanation in words

Two rows compare Aleutian islands. With otters: few and small urchins, and abundant kelp. Without otters, marked out as the contrasting case: many large urchins, and scarce kelp. An arrow in each row runs from the urchin condition to the kelp condition. A note beneath records that nobody assigned the otters — the fur trade did, unevenly — and that this makes the comparison a natural experiment, better than a correlation and weaker than a manipulation.

The otter case sits between the two extremes and is worth understanding as a type. Nobody assigned otters to islands; the fur trade did, unevenly, and recovery was patchy — so there were islands with otters and islands without, in the same region, at the same time. That is a natural experiment, and it is far better than a bare correlation and weaker than a manipulation. Islands were not randomly assigned and may differ in other ways, and later work found the effect varies considerably between regions.

The Yellowstone chain, link by link

Some links are solid. The confident version omits the others.

Whether returning wolves reshaped Yellowstone’s vegetation and rivers is genuinely disputed

Contested

Researchers actively disagree, and the disagreement is substantive.

Following wolf reintroduction to Yellowstone in 1995, elk numbers fell and woody riparian vegetation recovered in some locations. Whether this constitutes a wolf-driven trophic cascade — particularly one mediated by elk fear rather than by elk numbers — is contested, with landscape-scale tests failing to find the browsing redistribution the hypothesis predicts.

Who this applies to
northern Yellowstone National Park since 1995Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Canis lupus, Cervus canadensis, Populus tremuloides

You may have heard

“Wolves changed the rivers of Yellowstone”

The phrase comes from a short film, not from a paper. Stream channels did change, and wolves may well have contributed; the claim that they caused it runs past evidence that cannot separate wolves from a drought, a bison irruption, returning grizzlies and a human hunt happening at the same time. It is the clearest case NatureHQ holds of a genuinely appealing ecological story being repeated far past what was shown.

Why we rate it this way, and what the caveats are
ContestedModerate confidence

Both the vegetation recovery and the failure of the landscape-scale risk test are well documented. What is unresolved is causal attribution in a system where drought, bison, bears and human hunting all changed over the same period, and where no wolf-free control landscape exists.

How far it can be extended

A single reintroduction into a single park with a distinctive history of elk overabundance. It is not a general statement about what wolves do to ecosystems.

Caveats

  • That wolves reduced elk numbers is not in dispute; the disputed part is how far the effects propagated and by what route.
  • The widely shared "How Wolves Change Rivers" video considerably outruns the published literature.
  • Absence of a control landscape makes strong causal inference impossible in principle here.

Where researchers disagree

  • Landscape-scale sampling of aspen along a predation-risk gradient found no preferential recovery in high-risk areas, contradicting the behaviourally mediated version of the cascade.
  • Willow and cottonwood responses differ from aspen, so results depend heavily on which species is measured.
  • The same period saw a severe drought, a large increase in bison, grizzly predation on elk calves and a substantial human elk harvest north of the park — each capable of reducing elk numbers independently of wolves.

Still unanswered

  • Can a behaviourally mediated cascade be detected at all at landscape scale?
  • How much of the elk decline is attributable to wolves rather than to drought and hunting?
  • Will slow vegetation responses become clearer over a longer horizon?

Last reviewed 2026-08-09

The evidence (3 studies)

In Yellowstone, reducing browsing alone did not bring the willows back. They also needed high water tables — which beaver dams used to supply, and the beavers had gone long before the wolves did.

Well supported

Good evidence backs this, though some details remain open.

Experimental manipulation of browsing and water availability in northern Yellowstone found that reduced browsing alone was insufficient for willow recovery; substantial recovery required raised water tables comparable to those historically maintained by beaver dams.

Who this applies to
Riparian willow in northern Yellowstone.Do not extend this beyond the taxa listed — the popular version over-reaches.
Studied in
Salix, Cervus canadensis, Castor canadensis
Why we rate it this way, and what the caveats are
Well supportedModerate confidence

A crossed experimental design with a clear result. Scaling from plots to the whole system involves assumptions the design cannot test.

How far it can be extended

A plot-scale experiment in one part of one park, in a system with a particular history of beaver loss.

Caveats

  • This does not show that browsing is unimportant; it shows browsing was not the only thing missing.
  • Simulated dams are not beavers, and reintroducing the animals would change more than water tables.

Still unanswered

  • Whether beaver recolonisation would restore the hydrology, and over what timescale.

Last reviewed 2026-09-04

The evidence (1 study)

How we know

Two things the willows were missing

Wolves returned and elk browsing fell. So why had the willows along Yellowstone’s streams not come back?

Willow plots in northern Yellowstone were assigned crossed treatments: browsing either excluded or permitted, and water tables either raised by simulated beaver dams or left as they are. All four combinations were maintained and willow growth was followed over years.

What happened

Excluding browsing alone did not restore willow to its former stature. Substantial recovery required raised water tables as well — conditions that beaver dams had maintained before the beavers were lost.

What it shows

That the missing ingredient was not only the predator. Beavers had been removed from this system long before the wolves were, and their dams had held the water table high enough for willow to thrive. Restoring wolves does not restore hydrology. A landscape can pass through changes that putting one species back does not simply undo.

What it does not show

It does not show that browsing is unimportant — browsing mattered in the design, it was simply not sufficient on its own. It is also plot-scale work in one part of one park, and simulated dams are not beavers: real ones would alter sediment, channels and much else besides water level.

The controls — what makes this evidence rather than a story
  • A crossed design, so the effect of each factor can be seen alone and in combination rather than confounded.
  • Plots left at current water levels and current browsing, giving the do-nothing baseline.
  • Multi-year monitoring, since willow response is slow and a single season would show nothing.

From Stream hydrology limits recovery of riparian ecosystems after wolf reintroduction

Diagram

Six links, told as one sentence

Each link marked by how well it is evidenced, rather than drawn alike.

Six links, told as one sentenceWolves returned, 1995RecordedElk numbers fellMeasured — several causesBrowsing pressure fellPartly; not redistributed as predictedRiparian plants recoveredIn some places; needs water tooBeavers increasedRecorded; attribution unclearRivers changedNot establishedSolid links are well evidenced; dashed are contested; the last is not established.
The same explanation in words

A chain of six links, each annotated with the strength of its evidence. Wolves returned in 1995: recorded. Elk numbers fell: measured, with several causes. Browsing pressure fell: partly, and not redistributed as the fear mechanism predicts — this and everything below it are marked as contested. Riparian plants recovered: in some places, and needing water as well as reduced browsing. Beavers increased: recorded, with attribution unclear. Rivers changed: marked out as not established. A note records that solid links are well evidenced, dashed ones contested, and the last is not established.

The willow experiment is the most useful single piece of this, because it does not argue about attribution — it manipulates. Browsing excluded, water tables raised, both, neither. Reduced browsing alone was not enough. And the water tables that were missing had been held up by beaver dams, in a system where beavers were trapped out long before wolves were removed. The landscape had changed in a way that returning one predator does not undo.

The rest of this

Eating them, and frightening them

Two mechanisms with different predictions.

A predator can reduce grazing by killing herbivores, or by making them avoid dangerous places — the second usually called a landscape of fear. The distinction matters because the two predict different patterns. Reduced numbers should relieve browsing roughly evenly; redistributed fear should relieve it specifically where the risk is highest, leaving safe places browsed as hard as before.

Diagram

Two arrows that should not be drawn the same

A measured direct effect, and an inferred indirect one.

Two arrows that should not be drawn the samePredatoreats — measuredHerbivoremay affect — inferredVegetationAlso acting here: rainfall, fire,soil, other herbivores, land useSolid: a direct effect, measured. Dashed: indirect, and competing with all of those.
The same explanation in words

A predator connects to a herbivore by a solid arrow labelled "eats — measured". The herbivore connects to vegetation by a dashed arrow labelled "may affect — inferred". A further dashed line into the vegetation box notes the other things acting on it: rainfall, fire, soil, other herbivores and land use. A closing line distinguishes the two: solid marks a direct effect somebody measured, dashed marks an indirect one competing with all of those other influences.

That is a testable difference, and in Yellowstone it was tested. A landscape-scale study comparing aspen recovery against measures of predation risk did not find the pattern the fear mechanism predicts. The result does not rule out the killing pathway, and it does considerable damage to the version of the story in which wolves change the landscape by frightening elk — which is, in most retellings, the version being told.

Claims about this, checked

Things people have heard, and what the evidence actually supports.

The research behind this page

4 studies, newest first. Each one has a page explaining what it found and what it could not show.

2013Proceedings of the Royal Society B: Biological Sciences

Stream hydrology limits recovery of riparian ecosystems after wolf reintroduction

Reduced browsing alone did not restore willow to its former stature.

2010Ecology

Are wolves saving Yellowstone’s aspen? A landscape-level test of a behaviorally mediated trophic cascade

Aspen were not recovering preferentially in high-risk areas.

1974Science

Sea otters: their role in structuring nearshore communities

Islands with otters had few and small urchins and extensive kelp.

1966The American Naturalist

Food web complexity and species diversity

On the removal plot, mussels expanded and progressively monopolised the rock.

This page is a stop on a longer route

A guided journey reads several subjects in a deliberate order, with an argument for why one follows another. You can join in the middle.

Where to go from here

Each of these follows from something on this page — a relationship in the evidence, a claim people ask about, or the next mechanism along.

How complete this page is, and what it is still missing

NatureHQ publishes its own gaps. This page is at 69% completeness against what we would call a finished subject, and was last reviewed on 2026-09-04. It carries 4 claims and answers 8 mapped search questions.

  • no research from the last few years is attached — check for newer work
  • Aquatic cascade experiments, which are the strongest evidence base, are described rather than covered individually.
  • Bottom-up control is treated on its own page and only referenced here.
  • Mesopredator release, a related indirect effect, is mentioned but not covered.