Spectral Mixture Analysis


The optical reflectance properties (visible and infrared color) of Earth surfaces can be represented as both reflectace spectra and as locations within a high dimensional mixing space.  Mixing spaces are defined by spectrally distinctive endmembers and varying mixtures of these endmembers.  Projections of lowdimensional mixing spaces allows pure endmembers and mixed spectra to be represented as clouds of pixels where each spectrum represents a discrite point within the cloud.  An example of a global mixing space with its bounding spectra is shown in the figure above. The mixing space is is derived from Landsat ETM+ images of 30 diverse Earth environments, each 30x30 km in area.  Each 30x30 km area is imaged by 1000x1000 pixels in which the color of each 30x30 m sample of the Earth surface is represented by both a reflectance spectrum and by a point in the 3D  mixing space. The mixing space allows pixels of homogeneous, optically distinctive spectral endmembers (like water, vegetation, ice, snow, soil and rock) and pixels imaging mixtures of these endmembers to be represented as a 3D cloud depicting the optical characteristics of the Earth surface.  The topology of this mixing space provides a basis for the development of spectral mixture models.  Inversion of mixture models yields estimates of the relative abundance of each endmember within each pixel.  Endmember abundance estimates can be used to map spatiotemporal variations in Earth surface properties.   For more details see:

  Small, C., The Landsat ETM+ Spectral Mixing Space, Remote Sensing of Environment, v.93, p.1-17, 2004.


Landsat ETM+ Imagery Examples

The image areas are 30x30 km.

All images are shown at full resolution (30 m pixel) so the scale (on your screen) is equivalent.

In these composites, the Red, Green and Blue layers correspond to

Short Wavelength Infrared, Near Infrared and Visible radiance measurements

(ETM+ Bands 7, 4 & 2 respectively).

The full pdf file of all 30 images is >30 MB.

Plan accordingly...