Quantifying the efficacy and retrieval boundaries of spectral unmixing algorithms using various endmember reduction techniques in the EMIT Science Data System

Dust is a major contributor to radiative forcing in arid regions, and consequently has major impacts on the climate cycle, including altering precipitation patterns, rates of shrub encroachment, and agricultural production. However, the exact role that suspended aerosols play across the Earth's atmosphere remains highly uncertain. The Earth Surface Mineral Dust Source Investigation (EMIT) imaging spectrometer will fill that knowledge gap with visible-to-shortwave infrared (VSWIR) measurements of the surface from the International Space Station. Global applications of retrieval techniques - even those considered relatively mature - require an unprecedented level of careful consideration, accountability, and uncertainty characterization. Our team used simulation studies to explore the limits of current retrieval efficacy, utilizing products from multiple algorithms to determine retrieval boundaries. Using a new composite of spectral libraries of drylands from around the globe, we generated millions of candidate spectra with different random fractions of surface composition. These synthetic mixtures were then analyzed to evaluate algorithm parameter selection - with emphasis on endmember library selection and Spectral Mixture Analysis metaparameter selection. Using monte-carlo sampling methods, we estimate mixture fraction uncertainty that incorporates modeling errors, instrument noise, and atmospheric processes, and show how the true error budget is fully encapsulated. The synthetic mixtures were further used to quantify the uncertainty in mineralogical retrievals induced by different quantities and types of vegetation. Our results provide a critical link for EMIT, as imaging spectroscopy data are used for the first time to characterize the Earth's surface with uncertainty accounted for end-to-end.