Pollination Networks: A Web Built by Foraging Bees

A network you cannot see from any single flower

A meadow of a dozen flowering plant species and a dozen pollinator species implies, in principle, over a hundred possible plant-pollinator connections. No individual bee experiences this network directly — it experiences only the flowers it happens to encounter on a given foraging trip. And yet, aggregated over a season, the pattern of who visits whom is neither random nor arbitrary: it has a specific, reproducible mathematical structure, generated entirely by simple, local rules that individual foragers follow.

Foraging as an optimisation problem

A foraging bee is, in effect, solving a continuous cost-benefit problem: which flower patch to visit next, given the nectar reward it remembers finding there, the energy cost of flying to it, and how depleted that patch is likely to be after recent visits from itself and other foragers. This is close enough to a formal optimisation problem that ecologists model it using optimal foraging theory, treating the forager as maximising net energy gain per unit time. In practice this shows up as bees learning to prefer patches with a track record of good reward, avoiding recently visited flowers whose nectar has not yet replenished, and abandoning a patch once its average return drops below the expected return of moving on — the same logic behind the marginal value theorem used to model foraging across many animal species, not just pollinators.

Flower constancy: specialising without being told to

The single behaviour with the largest effect on network structure is flower constancy: once a bee has learned to efficiently extract nectar from one flower species — figuring out where to land, which direction to probe, how to trigger the specific mechanism that species uses to dispense pollen and reward — it tends to keep visiting that same species on subsequent trips, even walking past other, sometimes more rewarding, flowers to do it. The likely reason is cognitive: switching between differently shaped flowers requires relearning a distinct handling routine each time, and that relearning cost outweighs the benefit of occasionally finding a slightly richer patch of a different species. Flower constancy is good for the plant too, arguably the whole point from the plant's perspective: pollen delivered to the wrong species' stigma is wasted, so a bee that stays loyal to one species is a far more effective pollen vector than one that samples indiscriminately.

forager rule, repeated independently by many bees:
  if currently constant to species S and S still has reward:
      keep visiting S (low switching cost, learned handling)
  else:
      sample nearby patches, weight choice by remembered reward
      and inverse distance; switch constancy if a species is
      consistently better and switching cost is amortised
  carry pollen collected from the current flower to the next
  visited flower of the SAME species -- cross-pollination

Nested, not random

When ecologists draw the full bipartite network of which pollinator species visits which plant species across a whole community, a consistent pattern shows up across ecosystems worldwide: the network is nested. A small number of generalist plants and generalist pollinators interact with almost everyone, while specialist species — pollinators that visit only one or two plant species, plants pollinated by only one or two insect species — tend to interact specifically with a subset of those same generalists, rather than pairing off with other specialists. Drawn as a matrix with species sorted by how many partners they have, this produces a distinctive triangular fill pattern rather than the scattered, roughly uniform fill a random network would show.

Nestedness matters ecologically because of what it does for robustness. If a specialist species is lost from a nested network, the rest of the community stays connected through the generalist hubs, and the damage stays localised. Random or perfectly modular network structures do not have this property to the same degree — losing a well-connected node in a random network can fragment it much more severely. This is one reason ecologists watching pollinator declines pay close attention to network structure, not just species counts: a community can lose diversity while still functioning, right up until it loses one of the generalist hub species holding the nested structure together, after which the loss of function can be abrupt rather than gradual.

Recruitment adds a second layer

In social bees, individual foraging decisions are not fully independent — a successful forager returning to the hive can perform the waggle dance, encoding the direction and distance of a good general food-source area relative to the hive and the sun, recruiting other foragers to search nearby. This adds colony-level reinforcement on top of individual optimal foraging and flower constancy: a genuinely rich patch draws in more foragers over time, in a feedback loop reminiscent of the trail-reinforcement seen in ant foraging and slime mold networks, though here the shared signal is a dance inside the hive rather than a chemical left in the environment.

Frequently asked questions

Try it live

Everything above runs in your browser — open Pollination Network and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.