Presets
Playback
Shell, launches and replay
Camera
Layers
Quality
Model check
The science
What you are looking at
The Earth (Blender-baked continents, oceans, ice caps and cloud layer) is surrounded by one orbital shell 50 km thick. Every dot is a tracked object: grey fragments, plus some intact satellites and spent rocket bodies. Over a 200-year replay the number of dots follows the population N(t) from the model below. Orange flashes are collisions drawn at the model’s rate. Objects are drawn far larger than real ones, and when N is large one dot stands for several objects (the readout says how many).
Model (identical to the 2D page)
The population obeys dN/dt = L − N/τ + (F−2)·½σvN²/V. L is the launch rate, τ the atmospheric-drag lifetime, σ the mean collision cross-section, v ≈ 10 km/s the mean relative speed, V = 4π(RE+h)²·50 km the shell volume and F the number of trackable fragments per collision (two parents are lost, hence F−2). The collision rate ½σvN²/V is the kinetic theory of a gas: pairs grow as N². The equation is integrated with RK4 (h = 0.05 yr).
Why a threshold
Loss is linear in N, the collisional source is quadratic. Setting dN/dt = 0 gives two roots: a stable level N* and an unstable critical level Ncrit. Starting above Ncrit fragments are made faster than drag removes them (the Kessler syndrome). If 1/τ² − 4(F−2)cL < 0 there is no equilibrium at all. The last graph shows how the two roots move with altitude: higher shells clean themselves far more slowly, so the safe region shrinks.
Honest simplifications
Drag lifetime is the rough scaling τ = 25 yr·exp((h−600 km)/110 km) (real values vary by an order of magnitude with the solar cycle and area-to-mass ratio). All objects are alike, fragments do not spread to other altitudes, the relative speed is fixed. Orbit directions and speeds of the dots are only decoration (all in one visual time-lapse), and positions of the flashes are random among the visible dots. This shows the mechanism, not a forecast; it is not for risk assessment.
Checks run in the page
- The closed-form checks of the 2D page: RK4 step halving, long-time limit equal to the analytic stable root, and the unstable root acting as a true threshold.
- The collision events the 3D layer would draw over the whole replay (Poisson draws at rate c·N²) are compared with the integral of c·N² dt, within 4 standard deviations.
- The rendered dots are read back from the instance matrices: each must lie within (6371 + altitude ± 25 km)/6371 Earth radii, and dots × objects-per-dot must equal N.
Credits
Blender 5.2 Earth with Cycles-baked textures, satellite, debris and rocket-body models; Three.js r160 instancing.