navigation image mapnext pagetable of contentsprevious page

Very believable portrayals of the different alteration types at White Mountain were achieved by ratioing, principal components analysis, and maximum likelihood supervised classification, acting on both Landsat TM and Bendix scanner data. Also, a preview of the power of hyperspectral remote sensing is introduced on this page.


Ratio, PCA, and Maximum Likelihood Analysis of the Utah Site


The question now becomes: what special processing will accentuate or enhance the alteration types? Using this summer, 1984, Landsat 4 image, we try three types: ratioing, Principal Components Analysis (PCA), and Supervised Classification.

Examine in succession, four ratio images, about which we will briefly comment.

Ratio (3/1) image of White Mountain TM subscene.

Ratio (4/2) image of White Mountain TM subscene.

The ratio ofTM Band 3 to Band 1 (3/1) renders most of the area in rather dark grays, but several areas are whitish (brighter). These probably correspond to zones of strong hematitic alteration (very reflective in band 3 but dark in band 1). The ratio 4/2 is similar but the bright areas appear displaced. These may mark local areas of denser vegetation.

Ratio (7/5) image of White Mountain TM subscene.

Ratio (1/7) image of White Mountain TM subscene.

The 7/5 image has a unique pattern, in which a hook-shaped dark area within a scene otherwise light-toned coincides closely to the general altered zone. Band 7 is an excellent detector of hydrous minerals such as the clays, alunite, etc., because these absorb radiation (hence significantly reduce reflectance). Ratio 1/7 shows bright areas that are approximately the same as the basalts and some of the andesites.

5-6: If you had to choose a single ratio image as the one to take with you into the field to check out alteration, which would you pick and why? ANSWER

A ratio color composite made up of Blue = 7/5, Green = 1/7, and Red = 3/1 does not separate the two volcanic rock types (both are blue-green) but shows White Mountain as purple and renders some of the k/a (kaolinite/alunite) areas yellow.

Ratio color composite of White Mountain, with B = 7/5; G = 2/7; R = 3/1

A second ratio composite (second image above), with Blue = 1/7, Green = 4/2, and Red = 3/1, produces a much different result. The deep blue closely expresses the basalt outcrops with the andesites now in a different shade of blue. White Mountain is a distinct orange-brown, but note that the same color appears north of the basalt. The k/a zones are a purplish-red, and quite distinct from a different red for the more iron-rich zones. So, ratio images seem to improve on standard composites in terms of alteration detection.

Ratio color composite of White Mountain, with B = 1/7; G = 4/2; R = 3/1

5-7: Which of the two ratio composites would you take in the field? ANSWER

Let's evaluate the utility of Principal Components Analysis (PCA). The first principal component provides, as we've seen before, a view much like a black and white aerial photo. Lighter tones mark k/a alteration areas. There are several very bright, small spots. These are probably pits dug by prospectors (one pit exposing exceptional quality alunite was visited by the writer but was not entered because about 30 rattlesnakes were slithering about - there are limits to one's ambitions as a remote senser!).

First PC image of the White Mountain TM subscene.

The second principal component is darker overall, with some alteration especially dark. The image again shows the bright spots and a bright area west of the basalt hills is part of the area noted in the regional scene as light-toned alluvium.

Second PC image of the White Mountain TM subscene.

The third principal component seems meaningless, except that the black spots probably correspond to certain alteration zones.

 Third PC image of the White Mountain TM subscene.

A glance at the fourth principal component shows the same dark hook-like pattern that was observed in the ratio image 7/5. White Mountain is set apart by its light tones, with similar tones north of the basalt deposits.

Fourth PC image of the White Mountain subscene

A principal component color composite, consisting of blue = PC2, green = PC4, and red = PC1 is resplendent with information. The basalt rocks appear blue-green, whereas the andesites tend to be dark blues. White Mountain is a distinct yellow, as is the area above the basalts. This strongly implies that these are outcrops of limestone similar to those at White Mountain, however, no geologic map to prove that supposition is available. The k/a zones appear in wine purple color. The hematitic zones are deep reds and yellows. The areas covered by alluvium tend to be multi-colored, with uncertain boundaries.

Principal Components 2,4,1 composite image of White Mountain TM subscene.

A second PCA composite where Blue = PC4, Green = PC5, Red = PC2 is less definitive. Basalts are purple and andesites may be green and/or yellow. White Mountain is not distinct. The k/a zones are bright red, but part of these is bounded by a black pattern, whose nature is puzzling. Nothing like it is evident in the individual PCA images, but the nebulous PC5 may be contributing.

Principal Components 4,5,2 composite image of White Mountain TM subscene.

Earlier, a PCA composite (shown below) was made from a 24-channel Bendix aircraft flight over White Mountain. For this, they included eight non-thermal channels and combined components 1, 2, and 4 into the composite shown below. Most of the rock units mapped in the field and identified in the Landsat images seem to show up but in some instances occupy somewhat different areas and have dissimilar sizes of outcrop. But, we safely conclude that, using the Bendix image as a standard, the Thematic Mapper (TM) PCA composites match fairly well. At least, they are good enough to stand alone as successful guides to the principal rock and alteration types defined by field work.

Principal Components 1,2,4 (channels on the Bendix Scanner) composite image of White Mountain.

5-8: All three above PCA products each seem to have useful information. Evaluate them beyond the information already offered in the preceding paragraphs. What in particular is separated rather well in the two TM images? ANSWER

We come now to a highlight of Section 5, a Supervised Classification of the Landsat TM data made by IDRISI with training sites based on the maps and on field observations by the author. Ten classes were established and then identified in training sites used to run a Maximum Likelihood classifier. Here is the end result (unfortunately, the legend initially created has been lost from the display program; see descriptions below).

Maximum Likelihood supervised classification (10 classes) of White  Mountain made using all 7 bands in the Landsat TM data set.

This is a colorful and a believable product. The classes designated Basalt (dark blue) and Andesite (green) are largely where they should be in field terms. White Mountain is well separated but its legend color (whitish) also is found where additional limestone outcrops are postulated north of the basalts. The Kaolinite/Alunite zone (purple) coincides well with the map information. The class designated as Ironrich (brown) is broadly equivalent to the geologic map unit called Moderately Hematized, whereas the class called Hematite (red) matches at least some of the map units called Strongly Hematized. Arbitrarily, four different classes of alluvium were set apart, based on photointerpretative and geologic reasoning. The class MixAlluv (gray) is partly within the altered zone and we assume it is a mix of altered rock and volcanic rock debris. DrkAlluv (dark gray) is a differentiable deposit consisting mostly of weathered volcanic residue. The class LsAlluv (light blue), we presume contains considerable contributions from White Mountain and other limestone sources. BrtAlluv (yellow) refers to alluvium west of the western basalt hills, which probably received much of its input from the Wah Wah Mountains. Its brightness (in individual bands and color composites) implies a variety of light-colored detritus (fragmented debris) and clays.

We also classified the White Mountain data collected by the Bendix scanner, using 7 of its channels. We consulted the same map to pick training sites, but these sites were not necessarily the same as we selected for the above Supervised Classification. Plus, we employed the IDIMS program for the processing algorithm, once again applying the maximum likelihood classifier. In the image above is the resulting classification and a color code for the selected classes (note that these are not the same as for the first classification). The major differences between the two classifications are: 1) we subdivided the volcanic units in part by degree of vegetative cover, and 2) we treated the unconsolidated cover (alluvium) as a single unit. In general, the correspondence between the two classification images and between this classification and the published map is good in both instances.

Maximum Likelihood supervised classification (9 classes) of White Mountain using 7 Visible-IR channels in the Bendix 24-channel data set.

5-9: Comparing the two classifications, how does the Bendix classification differ from the IDRISI one? Account for this. ANSWER

This case study with its accompanying images should convince you that satellite remote sensing has practical value. We already insinuated this idea in the Overview and the content of the first four Sections. Besides, remote sensing can lead to possibly sensational payoffs. We grant that the White Mountain example is almost a sure thing. The major types of alteration are distinctly different, so much so, that the color aerial photo is almost sufficient to produce an accurate map (remember, we said that the images seem superior to the pre-Landsat field map). But, the special processing products make these differences even more obvious.

Imagine that we have a one-time opportunity to stake several claims anywhere in Utah but must do so in 30 days. Its a big state! But, with Landsat and other space imagery, properly processed, we can shrink this huge area to just those small patches that apparently display abundant gossan. We can visit the most promising of these patches in brief trips, using rapid reconnaissance to seek signs of minerals. We can take samples for quick assay to determine grade or concentration (amounts of useful metals present per unit volume). Favorable results mean that we ought to file a claim by the deadline. Then, we must drill and map in detail to determine whether any mineralization we have detected is in enough gross volume to warrant developing and mining. With any luck, we’ll learn to fully realize the merits of prospecting from space.

Our chances of finding promising signs of mineralization will improve sharply if we use a spectrometer rather than multiband imagery. This improvement comes from many subtle variations in composition, as well as key information that helps to identify individual mineral species, that are present in detailed spectral curves but may be lost in undersampled multiband spectral data. We can now fly spectrometers on air and space platforms, providing hyperspectral rather than multispectral images and plots.

navigation image mapnext pageprevious page


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