The proliferation of data coming in daily from ocean remote sensing observational networks is getting bigger and will likely to get a lot bigger. This fact makes it natural to search for methods and strategies that can make the best use of this wealth of information. In this work, we investigate the utility of historical datasets to missing data interpolation issues. We state missing data interpolation as a data assimilation issue and present a data-driven strategy for the reconstruction of missing data in remote sensing observations series. Our data-driven strategy exploits a Hidden Markov Model (HMM) formulation. We report numerical experiments for simulated geophysical dynamics and real SST observation series, which demonstrate the relevance of the proposed framework