AECSF: Adaptive Ensemble Conditional Score Filtering for High-Dimensional Nonlinear Data Assimilation

This paper presents AECSF for Bayesian state estimation in high-dimensional nonlinear dynamical systems. The method builds on score-based diffusion filters as a sampling-based alternative for posterior estimation and uses adaptive ensemble conditional score filtering rather than relying only on heuristic likelihood corrections. Its motivation includes particle-weight collapse and the inability of Gaussian ensemble updates to capture non-Gaussian posterior structure.