Requires covariance forecasts
Prohibitive for large systems without
simplifications
Requires tangent linear model for traditional form
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
We use a 1D Kalman filter – computational efficiency, likely
little spatial correlation (through topography, soil
properties and atm forcing), 1D calculations