Two‐step ensemble data assimilation on the simplex: Application to sea‐ice concentration

Abstract Ensemble‐based data assimilation is widely used in atmospheric and oceanic sciences to improve modeled state estimates by incorporating observations. Ensemble Kalman filters are a class of data‐assimilation methods that assume the joint distribution in observation‐state space is Gaussian. Two‐step ensemble data assimilation is an alternative framework that relaxes Gaussianity assumptions. It works by first making a scalar update in observation space and then updating the state‐space variables accordingly. This article develops a non‐Gaussian parametric approach to the second step, the state‐space update, that is designed specifically for variables constrained to the simplex. The method assumes the joint observation‐state space prior is a mixed Dirichlet and constructs an analysis ensemble using transport methods. Results from the assimilation of sea‐ice concentration into a single‐column sea‐ice model (Icepack) show that, for ensemble sizes and , the new method outperforms existing approaches.