Imaging spectrometer record continuous reflectance and emission spectra in high spectral resolution and consequently resolve individual absorption bands which would remain undetected in broad-band multispectral imagery. Such data therefore nicely complements data from other sensor modalities such as multi-spectral scanner, Radar and Lidar. Unfortunately, however, the current temporal revisit frequency is insufficient as only a few orbital platforms have been launched.

In my talk, I will show a few successful applications derived from hyperspectral data cubes. This will include examples using both empirical approaches as well as physical-based approaches. I will discuss limitations of both approaches when analyzing data from imaging spectrometer with a particular focus on the ill-posed inverse problem, model transferability, as well as epistemic uncertainty as a result of model over-simplifications. I will conclude by presenting some recent progress in solving these issues.