Neural-Network-based Kalman Filters for the Spatio-Temporal Interpolation of Satellite-derived Sea Surface Temperature

International audienceIn this work we address the reconstruction of gap-free Sea Surface Temperature (SST) fields 1 from irregularly-sampled satellite-derived observations. We develop novel Neural-Network-based 2 (NN-based) Kalman filters for spatio-temporal interpolation issues as an alternative to ensemble 3 Kalman filters (EnKF). The key features of the proposed approach are twofold: the learning of 4 a probabilistic NN-based representation of 2D geophysical dynamics, the associated parametric 5 Kalman-like filtering scheme for a computationally-efficient spatio-temporal interpolation of Sea 6 Surface Temperature (SST) fields. We illustrate the relevance of our contribution for an OSSE 7 (Observing System Simulation Experiment) in a case-study region off South Africa. Our numerical 8 experiments report significant improvements in terms of reconstruction performance compared with 9 operational and state-of-the-art schemes (e.g., optimal interpolation, Empirical Orthogonal Function 10 (EOF) based interpolation and analog data assimilation)