Hyperspectral Remote Sensing, Imaging Spectrometers - Ground Truth; The "Multi" Concept; Imaging Spectroscopy - Remote Sensing Tutorial - facegis.com
Hyperspectral Remote Sensing, Imaging Spectrometers

Almost all sensors that are multispectral in function have had to sample the EM spectrum over a relatively wide range of wavelengths in each discrete band. These sensors therefore have low spectral resolution. This mode is referred to as broad-band spectroscopy. Spectral resolution can be defined by the limits of the continuous wavelengths (or frequencies) that can be detected in the spectrum. In remote sensors an interval of bandwidth of 0.2 µm in the Visible-Near IR would be considered low spectral resolution and 0.01 µm as high resolution. (The term has a somewhat different meaning in optical emission spectroscopy, where it refers to the minimum spacing in µm or Angstroms between lines on a photographic plate or separable tracings on a strip chart.)

It is now possible to operate remote sensors that can have high spectral resolution. Terms referring to high spectral resolution systems is hyperspectral remote sensing, hyperspectral spectroscopy , imaging spectroscopy, and narrow-band imaging.

Since higher spectral resolution leads to much more definitive information on the composition and certain physical properties of different materials and objects, being able to build and fly sensors with such resolutions has been a goal since remote sensing was in its infancy. Beginning in the 1980s, Dr. Alexander F.H. Goetz and his colleagues at the Jet Propulsion Laboratory began a revolution in remote sensing by developing a powerful new instrument called AVIRIS (for Airborne Visible-Infra Red Imaging Spectrometer).* This instrument took advantage of new detector technology to extend ground-based spectrometers into the air on moving platforms. Thus, the distinct value of obtaining hyperspectral curves has made it possible to acquire detailed data on the materials and classes present on the Earth's (or other planets) surfaces. Essentially these curves are continuous spectral plots that measure reflectances from the ground, water, or the atmosphere in the wavelength region responding to solar illumination (Visible-NearIR-Shortwave IR). The plots also record the fine details of absorption phenomena.

With these hyperspectral curves, it is practical now do rigorous analysis of surface compositions over large areas. Moreover, data can be displayed either as spectral curves with detail similar to those on the preceding page or as images similar to those obtained by Landsat, SPOT, etc. With spectral curves we capture the valuable information associated with diagnostic absorption troughs, and with images we get relatively pure scenes, colorized (through color compositing) from intervals that represent limited color ranges in the visible or in false color for the near-IR (NIR).

Images can be constructed to show variations in reflectance covering very narrow spectral intervals (e.g., 0.01 micrometers). This means that color composites made from bands that lie astride, or near, significant spectral indicators, such as absorption troughs, can contain color patterns closely tied to specific features. This next diagram - a plot of spectra for four common minerals - helps to elucidate how one would go about making informative images from hyperspectral data.

Hyperspectral curves for four minerals.

Each mineral has an absorption band in the 2.30 - 2.36 µm interval. The low point of the band varies by about 0.02 µm - just the band width that AVIRIS reads - from each other (Epidote and Antigorite almost coincide). Constructing a 3-band RGB composite using three contiguous narrow bands from the 0.06 µm interval would likely discriminate the four minerals if all are present (this singling out would be better if the minerals were spatially separated rather than co-mixed). Note that even better discrimination would result if one of the colors was used for a narrow band between 2.0 and 2.2 µm.

As an example of narrow band imagery, consider this pastoral scene produced as a true color image from data obtained by another imaging spectrometer (AVIRISwiss'91).

True color image obtained from the AVIRISwiss'91 imaging spectrometer.

Your reaction to this image is that it looks almost identical to a typical image made from Landsat and other broad band satellites. This is true, largely because broader spectral intervals generally have about the same levels of response (e.g., reflectance), so that if one takes a narrow band sample to make a color composite, it has about the same contribution as the broad band in which it is included, except if there is a particular absorption trough that is sampled instead. Images made from combinations of bands that are closely sampling the fine structure of a spectral curve will usually be more sensitive to the diagnostic deviations that broad bands tend to miss. Here is an example that makes this point:

Both images show a closed anticlinal fold in the Namib Desert. The left image is a natural color composite made from Landsat data. The right image was acquired with an imaging spectrometer in which 3 narrow hyperspectral bands were combined as a color composite. That image seems to show much more information, although an interpretation was not offered on the website from which both images were extracted.

Examples given on pages 13-8 through 13-10 will help to substantiate these ideas. For now, we will give one concrete example of how narrow band spectroscopy can be quite powerful in identifying materials at the "species" level. AVIRIS was flown over a test area - the Dragon Mine in the East Tintic mineral district of Utah. Various clay minerals are associated with different ore minerals. Kaolinite and a close relative, Halloysite, are two principal species; the precise location of each would be an aid to propecting for the ores elsewhere in the scene. These two minerals have similar spectra but with one diagnostic absorption band located about 0.3 µm apart. First, we show an image made from AVIRIS bands within a 2.0 - 2.4 µm interval. Below that are two "maps" showing the spatial dispersion of Halloysite (purple-red) and Kaolinite (green) in a part of the scene:

AVIRIS image of the Dragon Mine
Distribution of Halloysite Distribution of Kaolinite.

Thus, hyperspectral remote sensing in this case is able to recognize and locate minerals so similar that they could probably not be separated in the field (unless one has a spectrometer handy; small portable spectrometers are available commercially).

As experience accrued through operating AVIRIS, instrument designers built other imaging spectrometers, using different spectral dispersion devices, detector sizes, and other variables. We shall discuss the basics of instrument design later, along with a list of many of these operational spectrometers. As a result a large number of service organizations have emerged or expanded to offer hyperspectral remote sensing from aircraft as a commercial product. By the turn of the millenium, imaging spectrometers had entered space. At one point, the Space Shuttle manifest included a spaceborne version of AVIRIS, known as SISEX, but it fell victim to technological and budgetary constraints. A HyperSpectral Imager (HSI) was launched in 1997 onboard the Lewis satellite, but that system failed to reach a functional orbit.

The obvious improvements in information content gained by operating a spectrometer from air/space platforms are evident in this plot of the spectral response for the mineral Alunite (potassium-aluminum sulphate). Neither the TM nor MODIS, with their broader spectral bands are capable of picking out the diagnostic absorption band that identifies Alunite. The bottom plot has this information; curves produced from imaging spectrometer data would point to the band, so that the spectral data for that band could be used to make images or otherwise analyze for Jarosite locations within a scene containng this mineral.

Three spectral curves for the mineral Alunite; top and middle curves generalized from data obtained by Landsat TM and MODIS (an instrument on Terra-1; see Section 16); the bottom curve was obtained by a laboratory spectrometer and thus shows far more structure (e.g., absorption troughs).

Before describing these various imaging spectrometers and examining imagery and applications associated with them, we need to take a more extensive look into the fundamentals of spectroscopy than we presented in the Introduction of this Tutorial. To do this, we provide, in the next three pages, a condensation of a valuable review produced by a leading spectrometrist, Dr. Roger N. Clark of the U.S. Geological Survey.

* Note: A benchmark paper at the outset of imaging spectroscopy's appearance as a viable remote-sensing tool is: Goetz, A.F.H, G. Vane, J.E. Solomon, and B.N. Rock, 1985, Imaging Spectroscopy for Earth Remote Sensing, Science, v. 228, pp. 1147-1153.

Source: http://rst.gsfc.nasa.gov