Use of Non-iteration Dimensional-split Semi-Lagrangian (NDSL) in Central Weather Bureau Global Forecast System

Central Weather Bureau Global Forecast System(CWB-GFS) is planned to improve the spatial resolution into 15km in the near future. For Eulerian model, we have to use shorter time step to satisfy the CFL(Courant-Friedrichs-Lewy) condition, which increases more computational resources to meet the needs of operation center. Considering to save the computational resources, implementation of Semi-Lagrangian scheme should be the next. Semi-Lagrangian method usually be used on U (zonal wind), V (wind direction), T (temperature) and Q (tracers) as prognostic variables. However, it will be a great change for CWB-GFS which uses vorticity, divergence, virtual potential temperature and tracers as the prognostic variables. NDSL (Non-iteration Dimensional-split Semi-Lagrangian) provides us an simpler and more efficiency way to upgrade dynamic core from Eulerian to Semi-Lagrangian. it only needs to estimate the advection terms via NDSL from Eulerian form, instead of rewriting the governing equations from divergence/vorticity to winds. The results of 16-days integration showes that NDSL method can capture the characteristics of Eulerian model within 4 times larger time step, and also saving about 30% computational wall-time.