Hierarchical methods for global-scale estimation problems

There is a substantial signal-processing challenge associated with large-scale (especially global-scale) remote-sensing problems: solving the statistical inverse problem (i.e. deducing the properties of the sensed field from measurements) by brute force, that is by covariance matrix inversion, is completely impractical for fields involving millions of pixels. This paper reports on the ongoing development of an alternative technique, in which the statistical problem is modeled on a multiscale tree, applied to estimating sea-surface temperature (SST) based on infrared radiance observations from the along-track scanning radiometers (ATSRs).