EduDA - Educational Data Assimilation Platform / データ同化学習・シミュレーションプラットフォーム
EduDA (Educational Data Assimilation) is an open-source interactive web platform designed to facilitate intuitive understanding and empirical comparison of data assimilation algorithms (EKF, POEnKF, EnSRF, LETKF, 3DVar, 4DVar, Particle Filter) on the standard chaotic Lorenz '96 testbed.
Governing Equations: Lorenz '96 Chaotic Dynamics
dx_j / dt = (x_{j+1} - x_{j-2}) x_{j-1} - x_j + F (j = 1, ..., 40, F = 8.0, periodic boundary conditions)
Supported Data Assimilation Algorithms
- EKF (Extended Kalman Filter): Linearized tangent-linear covariance update with model process noise Q.
- POEnKF (Perturbed Observation EnKF): Monte Carlo ensemble sampling with perturbed synthetic observations.
- EnSRF (Ensemble Square Root Filter): Deterministic square root mean and perturbation update avoiding observation noise sampling error.
- LETKF (Local Ensemble Transform Kalman Filter): Parallel local grid-space low-dimensional ensemble transforms.
- 3DVar (3D Variational): Static Gaspari-Cohn background error covariance matrix (B).
- 4DVar (4D Variational): Adjoint model gradient optimization over a time assimilation window.
- PF (Particle Filter / SIR): Sequential Importance Resampling for non-Gaussian distributions.
Pre-Designed Educational Preset Labs
- Lab 1: Effects of Inflation - Preventing ensemble shrinkage and filter divergence under limited ensemble size.
- Lab 2: Effects of Localization - Cutting spurious distant correlations under small ensemble sizes.
- Lab 3: Static (3DVar) vs Flow-Dependent (LETKF) Covariance - Error propagation into unobserved domains under sparse observation.
- Lab 4: High-Dimensional Particle Filtering Limits - Experiencing weight collapse and the curse of dimensionality.
Please enable JavaScript to run and visualize the simulations interactively.