navigation image mapnext pagetable of contentsprevious page

The page is devoted mainly to examples of AVIRIS images, plus a pair of images from two other systems. Special attention is given to one site, Cuprite, Nevada, which is a splendid case study for illustrating the efficacy of hyperspectral imaging.


Additional Examples of Imaging Spectrometer Products; Multisensor Analysis

By now, hundreds of missions designed to test and use imaging spectrometers have flown, with many impressive images generated from the data. Still one of the most successful and instructive among these was the NASA AVIRIS flights over Cuprite, Nevada. The JPL AVIRIS team and the Spectroscopy Group at the U.S. Geological Survey in Denver reduce and manipulate the data. Be advised to check out the USGS page that will come up as it will offer the option of clicking on "Maps" which then brings up the Cuprite site as a choice. The images you will find at that site are superior to those below (the legends are readable), as these latter are degraded in downloading.

The Cuprite mining district lies near Tonopah, NV, in the southwestern part of the state. Gold and copper have been mined from here for more than a century. This area is a valuable geological study site to evaluate remote sensing, in particular with hyperspectral data sets, for mineral exploration, because of the wide variety of telltale alteration and other mineralization.

Here is a view made from three AVIRIS channels in pseudotrue color of the main area displaying significant alteration. The area shown is approximately 17 by 10 km.

 

Psuedotrue color AVIRIS image of Cuprite, Nevada.

By way of comparison, we got the next image from the HYDICE sensor. This false color combination consists of Red = Channel 175 (2,200 nm), Green = Ch. 125 (1,650 nm), and Blue = Ch. 50 (650 nm). The area here includes part of the AVIRIS scene but also includes other features.

False color HYDICE image of the Cuprite region.

13-28: Comment on the "purity" of the above two images compared with those of Landsat. ANSWER

The next image (at rather blurred resolution) shows part of the AVIRIS scene (bottom) in a color-coded classification of four minerals and a Thematic Mapper band ratio (R = 5/7; G=3/1; and B = 2/4) image of the same area, together with plots of the spectral curves obtainable by each system:

Comparison diagram of a Landsat TM classification and an AVIRIS classification of the Cuprite region.

We can select specific AVIRIS channels (one, or several contiguous ones, lumped together) to rely on specific absorption bands, from which we can derive values images from combinations of three channel sets. Here is an image in which the colors represent the iron minerals Hematite, Goethite, and Jarosite. The Goethite occurs mainly as stain or pigmentation in the alluvial materials in the valley and along slopes. (Black is unclassified).

AVIRIS classification of the Cuprite scene showing the distribution of Hematite, Goethite, and Jarosite.

By judiciously selecting channels extending into absorption bands that are indicators of particular mineral species or groups, we have identified and mapped most of the alteration-related minerals at Cuprite. This next map extends the location of iron-bearing minerals, which have many of their diagnostic bands in the 400 to 1,200 nm range.

AVIRIS classification of the Cuprite scene showing the distribution of iron-bearing minerals.

This map differs from the previous one (above) in subdividing Hematite and Goethite according to crystallinity and in showing other iron minerals as distinguishable chemical phases.

We can recognize an even more diverse assembly of minerals using bands within the 2,000-2,500 nm range. AVIRIS images have shown a large group of silicates, carbonates, and sulphates, as shown in this map.

AVIRIS classification of the Cuprite scene showing the distribution of A large group of silicates, carbonates, and sulphates.

13-29: Compare the above three images that show variations in certain elements that are then translated into their mineral variations with the natural and false color composites towards the page top. Do you see evidence in those latter two images of these mineral variations? ANSWER

Because the image is from the Internet, its resolution was only 72 dots per inch (dpi), which blurs the names of the minerals in the legend. But, of special interest is the mineral Buddingtonite, shown in fuchsia (a peach pink), which occurs in a few patches on the map. Buddingtonite is a rare form of the common potash-feldspar group. The ammonium ion, NH3+, partially replaces the potassium ion, K+. This map is an exciting example of high resolution hyperspectral data to reveal a notable diversity of minerals in alteration zones and fresh rock, as well, at a detail that could require years of field mapping to duplicate.

The Cuprite site has now been overflown at low and high altitudes to see how altitude affects the ability to distinguish mineral distributions. In the image pair below, the left shows a part of the Cuprite scene at 2.3 x 7 m resolution; the right, obtained from a higher altitude, is the same area at 18 x 18 m resolution. Judge for yourself from the patterns the extent of better or new information obtained at low altitude/high resolution.

Part of the Cuprite test site, imaged at high (left) and low (right) resolution; the browns denote a kaolinite + alunite mix and the blues indicate calcite.

Turning now to examples from other disciplines and subjects, we display several images touching on these, with minimal comment.

This next AVIRIS image, which we first saw on page 3-1, shows a group of crops in both circular and rectangular fields. The area is near Summitville, Colorado; a false color image of this site was included on page I-24. Note that the identified crops correspond to those shown on page 13-6, as a plot displaying spectral curves for a series of crops, first, and just below it, a continuum-removal plot of the same series. These came from the same data set used to generate this image.

Classification of AVIRIS data for a scene within the San Luis Valley of south-central Colorado, showing effective recognition of several crop types.

13-30: What do you think is meant by "nothing mapped"? ANSWER

In the next image, we show natural and false color AVIRIS images of field crops around Greybull, Wyoming, in the Bighorn basin. The offset spectra in the middle match with the several crops and features, labelled by letter, present in the enlargements, but the source of this image does not identify the crops.

AVIRIS observations, focusing on water features, appear in the next two images. The top is a natural color view depicting part of Key West, Florida, and the shallow waters around it that cover coral reefs. Below it is a view of snow (yellow) and water vapor (blue, but land features persist beneath it) capping and surrounding Mount Rainier in northwestern Washington State.

Natural color AVIRIS image of Key West, Florida, and the shallow waters surrounding it. Colored AVIRIS image of Mt. Rainier in Washington State.

 

The DAIS instrument, as earlier stated, has thermal channels as well as ones in the Visible and Short Wave-IR (SWIR). Here is an intriguing aerial oblique view made with that system, showing a natural color image on the left and a SWIR-thermal image on the right, of Mount Etna in Sicily, during one of its active periods. Smoke obscures some details of volcanic features in the natural image that "shine through" in the SWIR-Thermal-IR rendition.

Natural color and SWIR-thermal DAIS image of Mt. Etna in Sicily.

13-31: What does the thermal image tell you that is not evident in the visible image? ANSWER

Before closing this subsection that highlights the rapidly growing use of imaging spectrometers, we briefly mention another burgeoning, and loosely related, field of remote sensing. We now call it "multisensor analysis." The term refers to combining data obtained by more than one type of sensor on a spacecraft or, more commonly, by sensors on different spacecraft. For example, we may image a study area at various times by Landsat, SIR-C, TIMS, SPOT, MOMS, and AVIRIS. Of course, we may independently examine each data set, and imagery derived therefrom. Or we can lay visual products side by side. From this multiset, we can interpret the scene by simply looking at the various aspects of scene content. This process is standard procedure in conventional photointerpretation.

Or instead, as we have seen in several images shown elsewhere in this tutorial, we can merge, or register, two or more data sets to form a single image, composed of combined images. Landsat and radar data are good examples. Landsat provides a color rendition of the surface cover and radar provides a sense of topography or relief. Another example uses Digital Elevation Map topographic data, in digital format, to create a quasi-3D, perspective view of a SPOT scene.

Somewhat more sophisticated is the approach that uses each sensor data set as input to classification. Thus, we can combine visual and SWIR bands from AVIRIS with TIMS thermal data, so that 10 to 12, or more, band values contribute to the multivariate analysis that leads to a classified scene or map, likely to have improved accuracy. And, of course, we can digitize and combine other kinds of data by using aerial photography or thematic maps (described in the review of Geographic Information Systems in Section 15).

The trick to doing meaningful multisensor analysis lies in properly registering a variety of data sets. These data sets may come from sensors mounted on several platforms, which leads to multidimensional data, characterized by different pixel sizes, viewing geometries, orbital or flight line paths, times of year, angles of illumination, etc. Techniques for registering images have evolved over the years. One example superposes multitemporal Landsat data. Another combines day and night HCMM images (from orbits inclined to each other). We can now do Automated Multisensor Registration conveniently, using computer-based algorithms that register to a common geocoded base, integrate tie-point features, make geometric corrections (rectification), and resample pixels to a common size.

We clearly observe that, with the proliferation of sensors and their platforms (satellites) in space, the systematic combining of data, acquired over a wide range of wavelengths, scales, and temporal conditions, will result in a strongly synergistic use of the valuable data sets each operating system provides. Plans are well along for putting hyperspectral remote sensing routinely into space. This has already been started with the orbit of EO-1 (see first page of the Overview). ASTER, which acts like a broader band hyperspectral system, was launched with Terra in 1999 (see page 16-10). Another satellite under development, with launch in the next few years, is OrbView-4, operated by Orbital Imaging Corp.

navigation image mapnext pageprevious page


Primary Author: Nicholas M. Short, Sr. email: nmshort@ptd.net