Daniel Sousa, Christopher Small

This analysis presents standardized substrate, vegetation, and dark (S, V, & D) endmembers (EMs) for spectral mixture analysis (SMA) of MODIS daily nadir-looking BRDF-adjusted reflectance (daily NBAR) data. MODIS daily NBAR EMs are derived from a diverse collection of over 43 million MODIS spectra spanning 6 continents and all non-polar biomes. EM spectra found in the study are comparable to those of previous studies using decameter Landsat and Sentinel imagery. Sensitivity analysis of SVD mixture models based on 351 pairs of 27 possible single-pixel EM combinations shows mean S, V, and D fraction estimates to differ by 4 ± 3%, 3 ± 2%, and 3 ± 2%, respectively. In addition to the SVD EMs, an additional snow EM is also identified. This snow EM is deemed tentative pending a more detailed analysis of cryospheric environments. Vicarious validation based on unmixing of coincident Landsat 8 spectra shows complementary -7 and +11% biases in S and D EM fractions, but less than 0.5% bias in V EM fractions. Similar 6 to 7% dispersion in Landsat vs MODIS estimates is observed for all three fractions. Model misfit for the 4-EM SVD+snow MODIS model is low (> 99.9% of pixels with RMSE < 5%). V fraction is compared to Normalized Difference, Enhanced and Soil Adjusted Vegetation Indices (NDVI, EVI and SAVI). NDVI has over 2x the dispersion (29% vs 13 to 14%) and over 4x the bias (+18% vs 4%) of EVI and SAVI when compared against V fraction. Combined with previous studies, these results extend the scaling linearity and low misfit of the global SVD model from 2 m up to 500 m.