This work addresses the joint analysis of multi-source and multi-resolution remote sensing data for the interpolation of high-resolution geophysical fields. As case-study application, we consider the interpolation of sea surface temperature fields. We propose a novel statistical model, which combines two key features: an exemplar-based prior and second-order statistical priors. The exemplar-based prior, referred to as a non-local prior, exploits similarities between local patches (small field regions) to interpolate missing data areas from previously observed exemplars. This non-local prior also sets an explicit conditioning between the multi-sensor data. Two complementary statistical priors, namely a prior on the spatial covariance and a prior on the marginal distribution of the high-resolution details, are considered as sea surface geophysical fields are expected to depict specific spectral and marginal features in relation to the underlying turbulent ocean dynamics. We report experiments on both synthetic data and real SST data. These experiments demonstrate the contributions of the proposed combination of non-local and statistical priors to interpolate visually-consistent and geophysically-sound SST fields from multi-source satellite data. We further discuss the key features and parameterizations of this model as well as its relevance with respect to classical interpolation techniques.