Vigilance-Based Predator Detection Model
A Probabilistic Framework for Mixed-Species Groups
A mathematical modeling project exploring how group size, individual vigilance, and behavioral synchronization influence a mixed-species group's ability to detect predators.
Spatial simulation, hero visual
Swap in graphic from deck p.11
“Do mixed-species groups detect predators better? How does group size influence detection — and what behavioral factors limit the benefit of adding more members?”
The Problem
Animals often form groups to reduce predation risk — individuals can benefit from the vigilance of others, spending less time personally scanning while still maintaining collective awareness. But adding more individuals doesn't necessarily make a group proportionally safer: if members scan at the same time, their behavior becomes redundant rather than complementary.
Model Pipeline
Individual Vigilance → Independent Watchers → Collective Vigilance → Detection
Individual Vigilance
v(N) = v₀1 + kN
The first model represents how much time a single individual spends actively scanning for predators. v₀ is baseline vigilance when solitary, N is group size, and k controls how quickly vigilance decreases as the group grows — individuals can afford to scan less when surrounded by other potential watchers.
Larger groups allow individuals to reduce their own vigilance because detection responsibility is shared.
Individual Vigilance
Swap in graphic from deck p.4 — individual vigilance vs. group size for different values of k
Effective Independent Watchers
Neff(N, s) = N1 + s(N − 1)
Simply counting group members overestimates collective awareness, since members may scan simultaneously. The synchronization parameter s corrects for this: at s = 0, scanning is independent and every member contributes unique coverage; at s = 1, the group is fully synchronized and effectively behaves like a single watcher.
Group size alone isn't enough — the independence of individual behavior determines how much additional awareness each member actually contributes.
Effective Independent Watchers
Swap in graphic from deck p.5 — independent vs. partially vs. fully synchronized scanning
Collective Vigilance
C(N, s) = 1 − (1 − v(N))Neff(N,s)
Combining individual vigilance with the effective number of independent watchers gives the probability that at least one member is actively scanning at a given moment. This rises quickly with group size at first, then flattens — and synchronization reduces the benefit considerably, since overlapping members repeatedly provide the same coverage.
Collective vigilance shows diminishing returns: adding individuals helps most when their behavior stays relatively independent.
Collective Vigilance
Swap in graphic from deck p.6 — collective vigilance curves separating by synchronization level
Predator Detection Probability
Pdet(N, t, s) = 1 − e−μtC(N,s)
The final stage translates collective vigilance into the probability that an approaching predator is actually detected, modeled as a Poisson process. μ is the detection rate while vigilant and t is the observation window. Detection probability rises with collective vigilance and time, but again with diminishing returns as the group grows.
Larger groups generally improve predator detection, but the marginal benefit of each new member becomes increasingly small.
Predator Detection Probability
Swap in graphic from deck p.7 — detection probability over time
What the Model Found
Key Findings
Mixed groups detect predators better
Groups benefit from additional members as long as scanning behavior isn't completely synchronized.
The biggest gains happen early
Moderate-sized groups capture much of the available improvement before the detection curve saturates.
3–5 members already provide a meaningful advantage
Relatively small groups can capture a large portion of the collective-vigilance benefit.
Behavioral independence matters
Adding more individuals becomes far less valuable once their vigilance patterns start to overlap.
We realized the first model wasn't enough.
Taking the Model Further
Extended Model
The original framework assumes identical behavioral characteristics, a constant group size, uniform detection ability, and no spatial or communication structure. To make it more realistic, we extended it so each individual carries its own parameters:
Baseline vigilance
v₀,ᵢ — how vigilant an individual naturally is
Relaxation rate
kᵢ — how quickly its vigilance changes as the group grows
Detection ability
μᵢ — its effectiveness at detecting a predator
Position & radius
xᵢ, Rᵢ — whether the predator is close enough to observe
Pairwise synchronization
sᵢⱼ — how strongly its behavior correlates with another individual
Communication delay
τᵢ — how long detection info takes to reach the group
Pdet(t) = 1 − exp(−Σᵢ μᵢ wᵢ(N) δᵢ(xₚ) tieff)
Instead of assuming every member contributes equally, an individual now contributes only when it can realistically detect the predator — depending on its vigilance, position, detection radius, synchronization, and available reaction time.
Extended Simulation
Simulation Results
Spatial group configuration
Swap in graphic from deck p.11
Low- vs. high-synchronization detection curves
Swap in graphic from deck p.11
Detection probability kept increasing with group size, but the improvements grew progressively smaller. More importantly, low-synchronization groups substantially outperformed high-synchronization groups even at similar sizes — diversity in behavior can be more valuable than simply adding more individuals.
More isn't always better.
Independence matters.
Groups benefit not only from additional watchers, but from having watchers whose behaviors and information are meaningfully independent.
▸View Model Assumptions
The initial model assumes identical behavioral parameters across species, partially but not perfectly synchronized scanning, a relatively small vigilance-relaxation rate, and predator detections that follow a Poisson process based on collective vigilance. Group size is modeled from 1 to 20 individuals per species.
Limitations & Future Work
The original framework assumes a uniform detection rate across species, doesn't explicitly represent spatial organization or communication networks, and assumes a constant group size.
Potential improvements include species-specific detection rates, predator movement and dynamics, spatial heterogeneity, and more detailed reaction or communication delays.
What I Learned
This project showed me how mathematical modeling can translate complex behavioral systems into a series of manageable assumptions. The most interesting part was realizing that simply increasing the size of a system doesn't necessarily improve its performance — the relationships and dependencies between its components matter just as much. By progressively extending the model, we moved from an idealized probability framework toward a more realistic simulation involving heterogeneous individuals, spatial constraints, and information delays.