Other methods leveraged the spectral characteristics of EM variability to design improved algorithms (e.g., in spectral transformations and robust SU methods). However, there is still a noticeable dependence between the quality of the unmixing solutions and the necessary amount of user supervision in the algorithms. Many recent techniques need considerable tuning to reach their full potential, with a significant portion of algorithm design being left to the user. The lack of more extensive data with reliable ground truths has also made the evaluation of the algorithms somewhat difficult. In the following, we detail some aspects that we think deserve further consideration: Vegetation 1 0.2 0.4 0.4 0.5 0.2 0.4 0.2 0.2 0.4 0.6 0.4 0.2 0.51 1.5 2 Wavelength (µm) 0.5 0.1 0.05 1 0.5 0.2 0.4 0.6 0.5 1 1.5 2 Wavelength (µm) 0.1 0.05 0.2 0.15 1 0.1 0.05 0.15 ◗ As discussed, one important research direction is to improve the robustness of the methods to the selection of their parameters and to develop informed adjustment methodologies. This could be performed, for instance, by leveraging metadata (e.g., external classification maps) that are available in many applications. This point applies to the majority of SU algorithms reviewed in this article and would make those methods more readily employable as out-of-the-box solutions in practical scenarios. ◗ Most SU algorithms that address spectral variability depend strongly on spectral libraries and reference EM Soil 0.15 0.5 0.1 0.05 1 0.15 Water 0.1 0.5 1 1.5 2 Wavelength (µm) FIGURE 19. Spectral signatures returned by the algorithms that estimate the EM spectra for each image pixel. DECEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE 257 DeepGUn ELMM Fractional MESMA Reference