The Sentinel 2 MSI Spectral Mixing Space

A composite spectral feature space is used to characterize the mixing properties of Sentinel 2 Multispectral Instrument (MSI) spectra over a wide diversity landscapes. Characterizing linearity and identifying bounding endmembers allows Substrate Vegetation Dark (SVD) mixture model previously developed for Landsat MODIS sensors be extended MSI sensors. The utility SVD its ability represent variety landscapes in terms areal abundance their most spectrally physically distinct components. Combining benefits location-specific models with standardized indices, based offers simplicity, consistency, inclusivity applicability land cover mapping applications. In this study, set 110 image tiles compiled from hotspots worldwide provide basis characterization, identification that span space. resulting these 13,000,000,000 effectively 3D, 99% variance 3 low order principal component dimensions. Four continua are identified: Snow:Firn:Ice, Reef:Water, Evaporite:Water Substrate:Vegetation:Dark (water or shadow). first exhibit complex nonlinearities, but geographically dominant continuum conspicuous mixing. Bounding endmember identified continuum. subset 80 landscapes, excluding nonlinear (reefs, evaporites, cryosphere), linear produces fraction estimates modeled <6% RMS misfit. Two sets sensors, allowing unmixed globally compared across time light apparent disparity between 11D statistically 3D space, relative contribution 11 bands information content quantified using both parametric (Pearson Correlation) nonparametric (Mutual Information) metrics. Comparison (principal component) (Uniform Manifold Approximation Projection) projections reveal interpretable not resolved projection.