Accounting for Model Uncertainties with Stochastic Forcing and the Impact on Ensemble Performance

The NRL's regional ensemble system consists of NCOM and NCODA/ 3D-Var DA system. Due to a number of recent improvements, the probabilistic forecast performance has been enhanced significantly in terms of accuracy, skill, and reliability based on many deterministic and probabilistic verification metrics. These improvements in the ensemble system include: (1) adding stochastic forcing to the mixing parameters in the horizontal mixing parameterization based on the Smagorinsky scheme, and the vertical mixing parameterization based on the Mellor-Yamada Level 2 scheme, and (2) calibration of initial perturbations.

Despite the performance enhancements, the ensemble is still under-dispersive, i.e. lacking spread and spread growth. To overcome this difficulty, we design and implement a stochastic forcing model to add stochastic forcing to the tendencies of the important prognostic variables at all the grid points. One of the advantages of this model is the ability to generate the stochastic forcing with the desired statistical characteristics such as the spatial, temporal de-correlation scales and the amplitudes of the forcing in order to target various uncertainties with different scales. The stochastic forcing model with different spatial and temporal scales and amplitude has been tested and compared, including the ensemble without stochastic forcing. The results including issues, future opportunities and research areas will also be discussed and addressed.