From: Data Assimilation Methods for Neuronal State and Parameter Estimation
 | UKF | 4D-Var |
---|---|---|
Implementation choices | initial covariance (\(P_{xx}\)) | model uncertainty (\(Q^{-1}\)) |
sigma points (λ) | type of optimizer/optimizer settings | |
process covariance matrix (Q) | state and parameter bounds | |
Data requirements | Pro: can handle a large amount of data | Pro: may find a good solution with a small amount of data |
Con: may not find a good solution with a small amount of data | Con: cannot handle a large amount of data | |
Run time | Minutes | Days, hours, or minutes depending on choice of optimizer and settings |
Scalability to larger models | Harder to choose Q | Search dimension is (N + 1)L + D |
EnKF may use a smaller number of ensemble members | Sparse Hessian can be exploited during optimization |