Skip to main content

Crown Shyness — Jointly Generating a Whole Canopy

Authors: Thomas Yang @ Phi Lab Foundation · Fengchen Liu @ UC Berkeley

A stand grown from bare stems to equilibrium — 45 seconds.

Open the interactive showcase →

Overview​

Neighbouring trees of certain species stop growing just short of touching, leaving narrow channels of sky between their crowns. This pattern, known as crown shyness, belongs to no single tree: remove either neighbour and the channel between them disappears.

This project asks whether a canopy showing crown shyness can be generated jointly, as a whole stand, rather than by growing each tree independently and pruning it afterwards. That constraint is the point of the study. A model that reduces to independent per-tree generation followed by procedural trimming is treated as disqualified, however convincing its output looks.

Approach​

The simulation grows a stand from bare stems to equilibrium under two interacting local rules:

  • Occupancy — a branch tip that reaches into space a neighbour already fills stops where it is. Taller trees claim contested ground first, so every boundary carries the history of both crowns that shaped it.
  • Abrasion — where crowns do meet, wind grinds the contacting tips away. Each contacting branch retreats by a fixed fraction of its own length.

No gap width is specified anywhere in the code. The channel between crowns is not drawn, it settles: an equilibrium between growth pressure and contact loss. Because of this the spacing scales with the crowns themselves rather than with a hard-coded distance.

Measurement Discipline​

Every quantity reported by the project is produced by a single canonical estimator, applied per frame, under an enforced three-standard rule that keeps the generating pipeline, the measurement, and the success criteria from drifting apart. Gates are written before implementation and are never relaxed to fit a result; where a test cannot discriminate against its own noise floor, the outcome is reported as inconclusive rather than as a pass.

Interactive Showcase​

The showcase runs the simulation in the browser rather than replaying a recording. The die-back dial changes the abrasion fraction and the stand re-renders as it moves; the timeline replays the full growth arc; and stem density can be switched to watch the same rule act at four different scales. Median crown-edge clearance and canopy cover are plotted per frame from the same fixed estimator.

Open the interactive showcase →   Back to Research Portfolio