New Directions in 4D-Var: The ECMWF Perspective

4D-Var is a central component of ECMWF NWP system and has been key to its continuous success. Over the past 10 years the evolution of the ECMWF 4D-Var has been mainly driven by the increasing convergence with Ensemble DA methods. This development has been motivated by the need to provide reliable estimates of analysis and forecast uncertainty; by current and foreseeable developments in supercomputer architectures; and by a broader appreciation of the fundamentally probabilistic nature of the estimation and forecasting problem. While this trend is expected to continue, other facets of the data assimilation problem have become more prominent. This is true both for the scientific aspects of DA, e.g. nonlinear and non-Gaussian errors, model error representation, coupled DA; and for the computational aspects of 4D-Var, e.g., scalability and efficiency. In this talk we will discuss some of the new solutions that ECMWF is developing to tackle these challenges and why we still believe that a 4D-Var based approach is well positioned to give meaningful answers to these important DA questions.