EduDA — data assimilation simulator

Run seven data assimilation algorithms on the chaotic Lorenz '96 model and compare them as you change the conditions. Everything runs in your browser.

EKF — Extended Kalman Filter: Explicitly updates covariance via tangent linear approximation. High computational cost in large systems.
POEnKF — Perturbed Observation EnKF: Monte Carlo ensemble with perturbed observations. Efficiently captures flow-dependent error covariance.
EnSRF — Ensemble Square Root Filter: Deterministic square root update without observation perturbations, eliminating noise sampling errors.
LETKF — Local Ensemble Transform Kalman Filter: State-of-the-art operational weather forecasting method combining local parallel transforms and flow-dependent covariance.
3DVar — 3D Variational Data Assimilation: Uses a static, time-invariant background error covariance matrix (B). Computationally light but lacks flow dependency.
4DVar — 4D Variational Data Assimilation (L-BFGS): Optimizes initial states across a time window with dynamical consistency using L-BFGS quasi-Newton adjoint gradient optimization.
PF — Particle Filter (SIR / LPF): Represents fully nonlinear and non-Gaussian distributions. Supports standard SIR and Local Particle Filter (LPF).