List view
In reality no matter what we will be doing in future, we will need to rely on 10s of ensemble members in high dimensional spaces. So will require more investigation for areas such as sample-error corrections, use of historical information, structured input covariances etc.
No due dateAs it stands, our current range of filters explicitly build the covariance matrix in the data space. This scales poorly in the data dimension. However one can calculate updates without building the full covariance matrix, working instead with the samples matrix. This is something that has not been a priority for us as typically it is good practice to dimensionally reduce data before calibrations. Furthermore the forward model is typically far more expensive. However we may wish to utilize data of size O(10^4)+ in future and so this should be extended
No due date•1/1 issues closed- No due date
- No due date•1/3 issues closed
- No due date•3/3 issues closed
- No due date•0/1 issues closed