Requires covariance forecasts
Prohibitive for large systems without simplifications
Requires tangent linear model
May be unstable for highly non-linear problems
Basis for most sequential methods ie OI, nudging, PSAS etc
We will concentrate on the Kalman filter in this presentation