Model Error Representation in the Canadian Regional Ensemble Prediction System

State-of-the-art ensemble prediction systems typically use deterministic physical parameterizations (single or multiple choices) and ad hoc techniques for sampling errors and uncertainties originating from the subgrid-scale processes, truncation in the numerical discretizations and diffusion. This may involve perturbing poorly constrained physics parameters or adding perturbations directly to the physics tendencies. While these ad hoc methods are relatively simple to apply they are rather unsatisfactory from a more fundamental perspective. In the long term, development of inherently stochastic physics appears to be a more appropriate approach to represent model errors originating from the unresolved-scale processes.
In the first part of the presentation we will introduce the ongoing work aimed at developing stochastic parameterizations of deep convection and turbulence, for future implementation in the Environment and Climate Change Canada’s Regional Ensemble Prediction System (REPS). In the second part of the presentation, we will discuss the development of a new scale-separation verification method for applications in limited-area ensemble prediction systems, such as the REPS. The method allows for studying the impact of various sources of perturbations on the spread-error relationship as a function of spatial and temporal scale and aims at quantifying the value added by the inherently stochastic parameterizations.