4.1
The Development of NCEP 4D-Var System: Experimental Results Compared to those of 3D-Var
X. Zou, Florida State Univ., Tallahassee, FL; and H. Liu, J. Derber, J. G. Sela, R. Treaton, and B. Wang
Abstract:
We introduce the NCEP (National Centers for Environmental Prediction) global spectral forecast model and its ``full-physics'' tangent linear and adjoint models into the NCEP's SSI (Spectral Statistical Interpolation) analysis system and thus developed a 4D-Var system. The tangent linear and adjoint models used in 4D-Var include all the physical processes except radiation, i.e., the surface processe, vertical diffusion, shallow convection, gravity wave drag parameterization, large-scale precipitation and cumulus convection. Radiation is treated as an external forcing in these models. Numerical experiments were carried out at a resolution of 28 vertical layers and 62 waves triangularly truncated. The 4D-Var analysis is formulated similarly as the 3D-Var analysis and minimizes a cost function consisting of (i) the distance between the desired analysis and background field which is the 6-h forecast from the previous analysis ($J_b$), (ii) the distance between observations and the 6-h forecast from the desired analysis ($J_o$); and (iii) a penalty term constraining the norm of the time tendency of the divergence to eliminating large-amplitude gravity waves ($J_c$). The only difference between the 3D-Var and the current implementation of 4D-Var formulation is in the $J_o$ term in which the times of observations are taken care of more precisely in 4D-Var than in 3D-Var and the NCEP spectral model is used in 4D-Var as a strong constraint describing the time evolution of the analysis increment during the assimilation window. A parallel run of the 3D-Var and 4D-Var experimentations was undertaken to examine the performance of the 4D-Var system and to assess the computational requirement of 4D-Var. Results from a one week testing are encouraging. Compared with the 3D-Var parallel runs, 4D-Var showed good convergences, smaller analysis increments, better fits to data, and improved forecasts verified by the targeted dropwinsondes from the 1998 NORPEX (NORth Pacific EXperiment).
Session 4, Assimilation
Tuesday, 11 January 2000, 2:15 PM-4:45 PM
Next paper