search

jgomezdans / KaFKA

Kascading Fast Kalman Assimilation
GNU General Public License v3.0
5 stars 8 forks source link

Refine assimilation #3

Closed jgomezdans closed 7 years ago

jgomezdans commented 7 years ago

Currently, the assimilation is done very fast by assuming the prior covariance is diagonal. A second stage would use this fast solution as a way of preconditioning the problerm, or as a first guess. There are quite a few options on how to do this efficiently in the literature, but it's hard to see which one fits the use case we have.

jgomezdans commented 7 years ago

However, if the matrix is block diagonal, it might just be easier to solve the small n*n problems directly...