Evaluating global spectral unmixing techniques using imaging spectroscopy data for retrieval of green, non-photosynthetic vegetation, and soil fractional cover

Global estimates of fractional cover of green vegetation (GV), non-photosynthetic vegetation (NPV), and soil provide valuable information about the Earth system. As the new generation of Earth visible-to-shortwave infrared (VSWIR) imaging spectrometers take orbit, global fractional cover data will be obtainable with new and improved spectral unmixing algorithms. Using an ASD Field Spectrometer and the spectral line point intercept (SLPIT) method, we estimate fractional cover from field measurements to compare to contemporaneous airborne (Airborne Visible/Infrared Imaging Spectrometer; AVIRISNG) and spaceborne (Earth surface Mineral dust source InvesTigation; EMIT) imaging spectroscopy data. Field data were collected in multiple biomes in the US Southwest at 84 validation sites. We used EndMember Combination Monte Carlo (E(MC)²) and Multiple-Endmember Spectral Mixture Analysis (MESMA) to derive fractional cover, two candidate spectral unmixing algorithms used by current (e.g., EMIT) and future imaging spectroscopy missions. Field data exhibited strong agreement between ground and spaceborne/airborne measurements. Our best spectral unmixing approach, E(MC)², produced mean absolute error of ≤ 0.06 for NPV, GV, and soil with uncertainties ≤ 0.08 for all classes. We further investigated the performance of global vs. local endmember libraries and found that the global library outperformed the local endmember library. We investigated normalization techniques and their effectiveness with our ground and image fractions. Additionally, we calculated uncertainties from both ground and image fractions. Field results aligned well and within uncertainty predictions from previously reported simulation work. These findings support the use of E(MC)² to be used in current and future imaging spectroscopy missions.