Introduction of GRAPES_VAR

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 27 January 2011: 2:45 PM
Introduction of GRAPES_VAR
2A (Washington State Convention Center)
Wei Han, NOAA, College Park, MD; and J. Xue

GRAPES-Var is the variational data assimilation component of GRAPES, including GRAPES-3DVar (3 Dimensional Variational) and GRAPES-4DVar (4 Dimensional Variational). The global version of GRAPES-3DVar inherits most subroutines from the regional version, which was developed in 2001-2002, except that a spectral filter takes the place of a recursive filter in the latter for preconditioning of control variables. Assimilation cycles combining both regional and global versions of GRAPES-3DVar and GRAPES prediction model are built up and an extensive series of pre-operational trials are conducted. The GRAPES-3DVar is capable of directly assimilating both conventional and nonconventional observation data, such as radiances from polar orbital satellite NOAA15,NOAA16,NOAA17,NOAA18, NOAA19,METOP and FY3, atmospheric motion vectors from geostationary satellites and Velocity Azimuth Display (VAD) data from Doppler radar, etc. The use of remote sensing data in GRAPES-3DVar is of great significance in the alleviation of the sparseness of conventional observational data in NWP. GRAPES-4DVar is the extension of GRAPES-3DVar in that the time integration of the prediction models is introduced into the system to assimilate all observations distributed within a time window, thereby obtaining the flow-dependent background error statistics. Preliminary results of experiments of GRAPES-4DVar with real conventional and non-conventional observational data are encouraging, showing substantial improvements of forecasts in comparison with those of GRAPES-3DVar. GRAPES-4DVar is the first four-dimensional variational assimilation system capable of implementation in the real operational environment developed in China, and one of few variational assimilation systems based on non-hydrostatic prediction models currently available. The tangent linear and adjoints of the GRAPES prediction model in GRAPES-4DVar were completed by using an automatic differential tool, YHTAD, developed by mathematicians participating in the GRAPES project. The tangent linear and adjoints of GRAPES are also used in studies of the sensitivity of forecast errors to the initial conditions and other model parameters, as well as in ensemble forecast systems.