How a single fish's startle becomes a school-wide wave
When a predator strikes, the fish closest to it typically react first with a C-startle response—a rapid, C-shaped body bend that propels the fish away. In a school of hardyheads escaping diving terns, this maneuver appeared both underwater and in mid-air as fish leapt from the water [1]. The key is that a startle is not a private event: a turning fish is a visual and hydrodynamic cue to its neighbors, and if enough of them turn, the change in heading can cascade outward.
High-speed video of schooling fish under attack shows that the order in which individuals startle is not random; some fish consistently lead and others follow [3]. This matches observations from undisturbed schools, where avalanches of consecutive large turns are generally triggered by a small number of fish acting as effective leaders, and these leaders can induce large rearrangements of the group's trajectory [2]. In other words, a cascade often begins with a few responsive individuals rather than with the whole school reacting at once.
What the cascade looks like around a predator
The most visible outcome is a flash expansion: the school suddenly spreads outward, leaving a gap—sometimes called a vacuole—around the predator. In the hardyhead study, fish formed a vacuole around terns as the birds slowly ascended to the surface, and this escape maneuver worked about 80% of the time [1]. The fish that were caught, notably, did not respond to the escape movements of fish above them and became isolated, suggesting that being out of sync with the cascade can be fatal.
These cascades are not necessarily all-or-nothing. In empirical schools, avalanches of large heading changes show scale-free distributions—meaning small cascades are common and large ones are rare but possible, with no typical size [2]. A model based on the classical Vicsek model of collective motion, in which one individual acts as a leader subject to random reorientations, reproduces this avalanche behavior qualitatively and hints at a connection between effective leadership, long-range interactions, and cascades [2].
Why the wave usually stops short—and when it doesn't
A cascade that reached every fish every time would be maximally sensitive, but also costly: false alarms would constantly disrupt the school. Measurements of startle cascades in fish schools show they are subcritical—not maximally responsive to environmental cues—and that the distance to criticality decreases when perceived risk increases [5]. The authors argue this can be understood as managing a trade-off between sensitivity and robustness according to how risky and noisy the environment is, considering the costs of both true and false alarms [5].
This context-dependence helps explain why a school can sometimes mount a dramatic flash expansion and other times show only a local ripple. The cascade's reach is tuned to the situation: a clear, close threat pushes the school closer to full responsiveness, while ambiguous or distant cues may produce only a small avalanche [5]. The result is a system that can spread information quickly when it matters but avoids overreacting to every shadow.
What is still unclear about cascades and evasion
The link between cascade dynamics and successful predator evasion is not fully worked out. The scale-free avalanche study describes how small groups trigger large direction changes but does not explicitly connect these cascades to coordinated predator evasion [2]. Similarly, a recent modeling paper on risk perception and startle cascades addresses predator avoidance and collective escape but does not clearly explain how startle cascades produce coordinated evasion [4]. The hardyhead study documents the escape maneuvers and their effectiveness against terns [1], and the startle-order study shows that leaders and followers exist during attacks [3], but how the subcritical tuning observed in undisturbed schools [5] plays out during a real attack remains an open question.
Sources used in this answer
Fishy Tales: Behaviour of Schooling Fish Escaping from Terns
Hardyheads escaping diving terns used flash expansion and formed a vacuole around the bird, an escape maneuver effective about 80% of the time, and also used the C-startle response underwater and in mid-air.
Scale-free behavioral cascades and effective leadership in schooling fish
Avalanches of large heading changes in fish schools show scale-free size and duration distributions and are generally triggered by a small number of effective leaders, a pattern reproduced qualitatively by a leader-based Vicsek model.
Schooling fish under attack are not all equal: some lead, others follow
High-speed video shows that schooling fish under attack startle in a non-random order, with some individuals leading and others following.
Modeling and simulation of the fish collective behavior with risk perception and startle cascades
A model of fish collective behavior incorporates risk perception and startle cascades in the context of predator avoidance and collective escape.
Subcritical escape waves in schooling fish
Startle cascades in fish schools are subcritical, and the distance to criticality decreases when perceived risk increases, consistent with a trade-off between sensitivity and robustness.