Spectral unmixing is the estimation, from the spectrum of a mixed pixel, of the spectral signatures of the materials present (endmembers) and of their relative contributions to that pixel (abundances) 1 2.

A pixel is a mixture when it covers more than one pure material, which arises from limited spatial resolution or from strong mixing of materials 2. A mixing model describes analytically how the endmembers combine to form the measured spectrum, and unmixing amounts to inverting that formation process, so the abundances parametrise the model 1.

Each endmember should ideally be the spectrum of a pure material in the scene, although what counts as pure depends on the task 2. The unmixing problem is naturally ill-posed and admits a wide range of admissible solutions, so results depend on the model and constraints chosen 1.

NASA’s processing-level description does not mention unmixing, so the retained sources do not assign it to a data level 3. They also do not establish an accuracy range or a preferred algorithm.

Footnotes

  1. arXiv (Dobigeon et al.), “Nonlinear Unmixing of Hyperspectral Images: Models and Algorithms”. https://arxiv.org/pdf/1304.1875 ↩ ↩2 ↩3

  2. arXiv (Gimenez Zapiola et al.), “Hyper-spectral Unmixing algorithms for remote compositional surface mapping: a review of the state of the art”. https://arxiv.org/pdf/2507.14260v1 ↩ ↩2 ↩3

  3. NASA Earthdata, “Data Processing Levels”. https://www.earthdata.nasa.gov/learn/earth-observation-data-basics/data-processing-levels ↩