Uncertainty Quantification in Remote Sensing

Remote sensing data sets produced by NASA and other space agencies are a vast resource for the study of Earths (and indeed, other planets) physical processes. However, no remote sensing instrument actually observes these processes directly; the instruments collect spatially aggregated spectra. Inference on physical state based on these spectra occurs via a complex ground data processing infrastructure featuring a retrieval algorithm that depends on a forward model of physical process thought to generate the spectra. Data produced this way should be accompanied by uncertainties associated with these inferences. In fact, some retrieval methods based on Bayes Rule do produce nominal uncertainties, but in the usual implementation they are underestimated, by construction. Moreover, this approach views the retrieval as an instance of inverse uncertainty quantification rather than in a forward UQ context. In this talk I review the formalism used in the applied math and statistics UQ community (e.g, the National Research Councils 2012 report, Assessing the Reliability of Complex Models: Mathematical and Statistical Foundations of Verification, Validation, and Uncertainty Quantification) and argue that the retrieval process can and should be seen in different ways depending the science goal that motivates the retrieval in the first place.