Combining Analog Method and Ensemble Data Assimilation: Application to the Lorenz-63 Chaotic System

摘要: Nowadays, ocean and atmosphere sciences face a deluge of data from space, in situ monitoring as well numerical simulations. The availability these different sources offers new opportunities, still largely underexploited, to improve the understanding, modeling, reconstruction geophysical dynamics. classical way reconstruct space-time variations system observations relies on assimilation methods using multiple runs known dynamical model. This framework may have severe limitations including its computational cost, lack adequacy model with observed data, modeling uncertainties. In this paper, we explore an alternative approach develop fully data-driven framework, which combines machine learning statistical sampling simulate dynamics complex system. As proof concept, address chaotic Lorenz-63 We demonstrate that nonparametric sampler catalog historical datasets, namely, nearest neighbor or analog sampler, combined stochastic scheme, ensemble Kalman filter smoother, reaches state-of-the-art performances, without online evaluations physical