Modifications of two convective schemes used in the NCEP Eta Model

Brad Ferrier



The convective parameterization of Betts-Miller-Janjic (Janjic, 1994; BMJ), which is currently used in the operational Eta model, has been identified as being primarily responsible for overly active shallow convection (Baldwin et al., 2002), early triggering of deep convection over too large an area, and a low bias in higher forecast precipitation accumulations. The smooth texture of the Eta precipitation fields documented by Baldwin and Wandishin (2002) has been found to be primarily a result of the first-guess cloud top being determined by the highest equilibrium level of the parcel, in which convective triggering is determined based on column-integrated enthalpy and entropy considerations.

The forecast precipitation from a modified BMJ scheme becomes much more focused with greater temporal and spatial structure when convection is considered only when limiting the amount of lifting of (dry) air parcels from their source level to their condensation level to no more than 25 mb, and further lifting the amount of lifting to their level of free convection to no more than 50 mb. Two additional changes were needed in order to prevent the occurrence of spurious grid-scale precipitation maxima. The first includes adding the effects of ice, primarily through increased diabatic heating an lower ice saturation mixing ratios, in determining the cloud updraft parcel characteristics, as well as in the enthalpy conservation and entropy evaluations. Promising results were achieved when adjusting profiles towards a reference state saturated with respect to water (T>=0C) and ice (T<0C). The second change involves including the effects of grid-scale ascent in calculating the departure of the reference profiles from saturation, following eq. (10) of Betts (1986).

Extensive tests and changes were also made to the Kain and Fritsch (1993; KF) convective scheme to address a tendency of the scheme to produce larger precipitation amounts than are observed. It was found that detraining all of the rain and snow calculated in the updrafts onto the grid alleviated this problem. Other changes that helped were (1) Including grid-scale hydrometeor fields as input into the model, (2) modifying shallow convection so that hydrometeors remain in the column except for converting rain into cloud water, and (3) turning off convective downdrafts. Many sensitivity experiments also resulted in an optimal relationship for describing the updraft radius (which affects entrainment rates) as a function of cloud-base vertical motion.

These modified versions of the BMJ and KF schemes have been developed for possible implementation in the operational Eta model and in the Short-Range Ensemble Forecasting (SREF) system. Preliminary results will be shown for two case studies from spring 2002 run using the 10-km hydrostatic Eta model, as well as more recent cases from the past several weeks based on 32-km real-time parallel runs with full cycling and 12-km runs without cycling.