Ensemble Weather Forecasting — Why Forecasts Lose Skill
Interactive numerical weather prediction model: a nonlinear pressure field advects itself across a periodic grid, and a small ensemble of near-identical starting states is run side by side to watch forecast error grow with lead time — the real reason weather models can't predict past ~2 weeks.
Real weather models don't run once — they run dozens of times from slightly different starting conditions, because the atmosphere is chaotic: any two runs that begin almost identically will drift apart at a roughly exponential rate until they saturate. This simulator builds a small numerical weather model from scratch — a pressure field advecting itself across a periodic grid via its own geostrophic wind — and launches a handful of ensemble members from near-identical initial states next to one deterministic "control" run. Watch the terrain-like pressure surfaces (the ensemble "spaghetti plot") peel apart from the control as simulated forecast days pass, while live readouts track the ensemble spread and estimate the model's predictability horizon — the same diagnostic operational centers like ECMWF use to decide how far out a forecast can be trusted.
A nonlinear pressure field advects itself across a periodic grid while a small ensemble of near-identical starting states is run side by side, showing forecast error grow with lead time — the real reason weather models lose skill after roughly two weeks.
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install