Representing Uncertainties in the NOAA/NCEP Next Generation Global Prediction System Associated with Unresolved Convection

In this study we address the challenge of representing uncertainties in numerical weather prediction (NWP) models at the physical process level associated with unresolved convection. The method utilizes cellular automata (CA) (deleted in order) to describe sub-grid scale fluctuations and self-organization of convective cloud population (number and size) within a NWP grid-box. In order to make the sub-grid distribution of convective clouds physically based, we condition the rules that govern the evolution of the CA on perturbed model fields, such as the vertical velocity, specific humidity and CAPE. The perturbations to the model fields (which we condition the cellular automata with) are specified according to a general class of non-Gaussian, stochastically generated skewed (SGS) distributions that are derived from large eddy simulations and observations. We carry out experiments to explore the use of this method in the Chikira-Sugiyama convection scheme developed for NOAA's Next Generation Global Prediction System (NGGPS). In these experiments, we let each individual sub-grid cluster formed by CA-SGS cells in a NWP grid box represent a cloud-element in the Chikira-Sugiyama scheme such that the number of sub-grid cloud elements is varying in space in a stochastic manner. In addition we use a sepearate distribution of CA-SGS conditioned on the humidity field to treat also triggering aspects of the convection in a stochastic manner. The impact of this method on ensemble medium-range weather prediction, as well as in sub-seasonal weather prediction, is evaluated in terms of the precipitation distribution and organization of convection.