Data assimilation challenges posed by nonlinear operators: A comparative study of ensemble and variational filters and smoothers

Abstract The ensemble Kalman Filter (EnKF) and the 4D variational method (4DVar) are most commonly used filters smoothers in atmospheric science. These methods typically approximate prior densities using a Gaussian solve linear system of equations for posterior mean covariance. Therefore, strongly nonlinear model dynamics measurement operators can lead to bias estimates. To improve performance regimes, minimization 4DVar cost function follows multiple sets iterations, known as an “outer loop”, which helps reduce caused by assumptions. Alternatively, “iterative methods” follow similar strategy periodically re-linearizing operators. come with different, possibly more appropriate, assumptions drawing samples from density, but have seen little attention numerical weather prediction (NWP) communities. Lastly, particle (PFs) present purely Bayesian filtering approach state estimation, avoids many made above methods. Several strategies applying localized PFs NWP been proposed very recently. current study investigates intrinsic limitations data assimilation methodology applications that require In doing so, it targets specific problem is relevant remotely-sensed measurements, such radar reflectivity all-sky radiances, pose challenges Gaussian-based systems. This comparison includes approaches designed recently nonlinear/non-Gaussian applications, well those currently NWP.