As we know, for the ensemble Kalman filter (EnKF), we need to create a set of samples in the beginning and then to run the predict and analysis step. But for now I have a question of how to create the ensemble samples.

For example, the initial state of some system is $x^0$,suppose we also know the true initial state $x^t$, then we can compute the covariance by $P^0=cov(x^0,x^t)$, and we can create an ensemble that follows the multivariate normal distribution,$N(x^0,P^0)$.

Q1:I am not sure if this implement is a Monte Carlo method?

Q2:It seems that using MCMC method or Gibbs sampling can also produce a certain ensemble to follow a certain distribution. But for the above case, is there any difference? Does one use the $N(x^0,P^0)$ to create the samples or use MCMC method to create samples following $N(x^0,P^0)$?

Q3: However, if we only know the $x^0$, and do not know the true state, or covariance $P^0$, how can we estimate the $P^0$, or create the initial ensemble samples for EnKF.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.