Joint Characterization of the Cryospheric Spectral Feature Space

Multispectral and hyperspectral feature spaces are useful for a variety of remote sensing applications ranging from spectral mixture modeling to discrete thematic classification. In many these applications, models used project the higher dimensional continuum reflectances (or radiances) onto lower mappings image target’s physical properties or categorical composition. such cases, characterization space dimensionality, geometry topology can provide fundamental guidance effective model design. Utility this characterization, however, hinges on identification appropriate basis vectors space. The objective study is compare contrast two fundamentally different approaches identifying via dimensionality reduction. so doing, we illustrate how be combined render joint that reveals not apparent using either approach alone. We use diverse collection AVIRIS-NG reflectance spectra ice snow utility facilitate both classification reflectance. Joint also shown assist with interpretation inferred spectra. Spectral combining principal components (PCs) t-distributed Stochastic Neighbor Embeddings (t-SNEs) physically interpretable dimensions representing global structure cryospheric as well local manifold structures revealing clustering resolved within continuum. distinct continua snow-firn gradients parts Greenland Ice Sheet multiple clusters common glacier sea in locations. revealed t-SNE spaces, extended distinguishes subtle differences curvature specific spatial locations accumulation zone, BRDF effects related view geometry. ability PC + produce while preserving hyperspectra suggests type might much all terrestrial land cover.