Spatio-temporal filtering using wavelets

In this paper, a class of spatio-temporal processes with first-order autoregressive temporal structure and functional spatio-temporal interaction is introduced. The spatial second-order regularity is allowed to change over time and is characterized in terms of fractional Sobolev spaces. The associated filtering problem is considered, assuming that observations are defined by spatial linear functionals of the process of interest, being affected by additive noise. Conditions under which a stable solution to this problem is obtained are studied. A functional least-squares linear estimate fusion method is derived to calculate this solution A multiscale finite-dimensional approximation to the problem is obtained from the wavelet-based orthogonal expansions of the time cross-section spatial processes, which allows the numerical inversion of the linear operator involved.