NUMERICAL EXPERIMENTS ON MULTI-LEVEL STATISTICAL ESTIMATION OF DYNAMIC BALANCE CONSTRAINTS IN GRAPES-3DVAR

This paper further explores the estimating and expressing of dynamic balance constraints using statistical methods in GRAPES-3DVAR(Version GM). Unlike the single-level scheme which only considers the coupling between mass and wind at one level, the multi-level scheme considers the coupling between their vertical profiles and calculates the balanced mass field at each layer using the rotational wind at all model levels. A reformed ridge regression method is used in the new scheme to avoid the multicollinearity problem and reduce the noises caused by unbalanced mesoscale disturbances. The results of numerical experiments show that the new scheme can get more reasonable vertical mass field, reduce the magnitude of the adjustment by the initialization, and improve the potential temperature analysis performance. Furthermore, the results of forecast verification in January(winter) and July(summer) both confirm that the new scheme can significantly improve the temperature forecast accuracy and bring slight positive effects to the pressure and wind forecast.