Fig. 1.
Schematic diagram of the filtering matrix ρ.
Fig. 2.
Spatially averaged (a) height RMS error and (b) wind RMS error for the four data assimilation methods [the PODEn4DVar (solid line), En4DVar (long-dashed line), LETKF (short-dashed line) and 4D-LETKF (dash-dot line)] without model error (h 0=250 m), respectively.
Fig. 3.
Spatially averaged (a) height RMS error and (b) wind RMS error for the four data assimilation methods [the PODEn4DVar (solid line), En4DVar (long-dashed line), LETKF (short-dashed line) and 4D-LETKF (dash-dot line)] for the imperfect model (h 0=0 m).
Fig. 4.
Spatially averaged (a) height RMS error and (b) wind RMS error for the PODEn4DVar for the imperfect model (h 0=0 m) with the ensemble number N=100, 90, 80 and 70, respectively.
Fig. 5.
Spatially averaged (a) height RMS error and (b) wind RMS error for the PODEn4DVar with the hybrid localisation scheme (referred to as “Hybrid”) and the implicit one (referred to as “Implicit”) for the imperfect model (h 0=0 m).
Fig. 6.
Spatially averaged (a) height RMS error and (b) wind RMS error for the PODEn4DVar with the square root ensemble method (referred to as “New”) and the original one (referred to as “Ori”) for the imperfect model (h 0=0 m).
