Towards ensemble data assimilation for the Environment Canada Regional Ice Prediction System

A short-range high-resolution sea ice prediction system has been developed at Environment Canada. This study describes the first steps towards transitioning this system from a simple deterministic data assimilation system based on the three-dimensional variational (3D-Var) approach into a data assimilation system based on an ensemble of ensemble-variational (EnVar) analyses. First, an ensemble of 3D-Var analyses using static background-error covariances is implemented and used to evaluate different strategies for simulating model uncertainties during the ensemble forecast step; these range from perturbing parameters within the sea ice model to completely disabling the sea ice dynamics or thermodynamics in some of the ensemble members. The experiments show a good ensemble spread–error relationship in areas with low or high ice concentration, though more work is needed to better simulate uncertainties in areas with intermediate ice concentration. Second, results from idealized experiments with EnVar analyses using ensemble covariances are presented. They demonstrate the potential improvement of sea ice analyses from using state-dependent multivariate ensemble covariances when assimilating ice concentration observations to correct both ice concentration and unobserved variables such as ice thickness and ocean temperature.