Uncertainty-Weighted Spectral Similarity for Hyperspectral Imaging Spectroscopy

Spectral similarity measures such as the Spectral Angle Mapper (SAM) treat all spectral bands equally, an assumption difficult to justify for modern imaging spectroscopy data where measurement uncertainty varies strongly with wavelength. We introduce wSAM, an uncertainty- weighted extension of SAM that incorporates per-band observational uncertainty as inverse- variance weights. On 47 EMIT–RadCalNet observation pairs across five sites, wSAM (w= 1/σ2) reduces mean spectral angle by 49% relative to SAM and outperforms both hard band masking and alternative weighting schemes, while a cross-site discriminability analysis confirms that weighting sharpens rather than erases spectral differences between surface types.

IEEE Geoscience and Remote Sensing Letters, 23, 2503905