The GRAPES global four-dimensional variational data assimilation system (GRAPES-4Dvar) was put in operation on 1 July 2018 , which is developed by the Numerical Prediction Center of China Meteorological Administration (CMA) on its own, consists of a number of core techniques for data assimilation, including GRAPES non-hydrostatic full-compressible global tangent-linear and adjoint models, linearized physical parameterization schemes, global 4Dvar parallelization algorithms and global 4DVar observational data profiling, incremental analysis with multiple outer loops, preconditioned Lanczos-CG minimization algorithm, digital filter as a weak constraint, and adaptive bias correction scheme for satellite observation. These techniques can effectively assimilate the conventional data with high frequency in time and various types of satellite data, and they have increased the data utilization by about 50% compared with 3D variational assimilation system.

 The results of the back-test and parallel operational test show that the 4Dvar-based GRAPES global forecast has been improved in all fronts in the short and medium range. To be specific, the forecast of the Northern Hemisphere within 3 days and that of the Southern Hemisphere from 1 to 10 days have been significantly improved, the forecasting skills for heavy rain belts and large-scale precipitation have been enhanced, and the forecasting error for typhoon tracks has been significantly reduced (by about 15%). According to the synoptic verification by the Central Meteorological Observatory, the 4Dvar-based GRAPES global forecast has seen the improved stability and accuracy in the dynamic field and large-rainbelt forecast and the improved accuracy of predicting the range and intensity of temperature drop.

Since sustained over the Shuguang High Performance Computer and the China Integrated Meteorological Information Service System (CIMISS) hosted by CMA, which are sufficient to meet its requirement on computing resources and speed, the operational GRAPES 4Dvar can be running easily and steadily. As GRAPES 4Dvar has made breakthroughs in analysis quality and computational efficiency, and its performance indicators for assimilation and forecast generally outperforms the current operational global 3D variational assimilation system, the GRAPES global forecast system( GRAPES-GFS) which has been upgraded from Version2.1 to Version 2.2, has been running four times (00,06,12,18UTC)a day.

Reporter:Ren Lu