Professional Documents
Culture Documents
net/publication/229383008
CITATIONS READS
180 393
3 authors, including:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Estudo geofísico das estruturas de impacto de Vista Alegre (PR) e Vargeão (SC) View project
All content following this page was uploaded by Alvaro Penteado Crósta on 29 May 2019.
foot on the ground. Therefore, techniques that require Bodie has characteristics of a bonanza-type lode
no a priori input of ground data were assessed in their quartz vein deposit and a large-scale stockwork system.
ability to convert image data to ground reflectance and Gold occurs in quartz veins and quartz–adularia–
for establishing the spectral similarity between image chalcedony stockworks in the center of the district. Pet-
pixels and reference minerals. rographic work by Herrera et al. (1993) showed vertical
The area selected comprises the Bodie mining dis- and lateral zoning in the alteration with silicification in
trict and the southern part of the neighboring Paramount the center surrounded by zones of potassic, argillic, and
district located in Mono County, California, USA (Fig. 1). sericitic alteration and an outer zone of propylitic alter-
AVIRIS data were acquired on 20 August 1992, at 11:22 ation. Little information is available for the Paramount
a.m. (local time) by NASA’s ER-2 aircraft. The scene used district due to the lack of important mineralization, but
in this study consists of 614⫻512 pixels (10.44 km⫻9.70 the district has recently been prospected for bulk ton-
km) in 224 bands at 16-bit radiometric resolution. nage precious metal mineralization (Herrera et al., 1993).
Geological characteristics observed at Paramount are very
similar to those in Bodie, and the origin of the alteration
GEOLOGY AND MINERALIZATION OF THE
is believed to be similar (Silberman et al., 1995).
BODIE AND PARAMOUNT DISTRICTS
Bodie was an important gold-silver mining district in the
DATA CALIBRATION TO
second half of the 19th century, whereas only a mercury
SURFACE REFLECTANCE
mine was exploited at the Paramount district. Miocene
(7.8–9.4 Ma) intermediate to basic volcanic rocks (daci- Identification of surface materials using hyperspectral
tic, andesitic, and rhyolitic lavas, plugs, tuffs, and brec- data is based on quantitative comparisons between pixel
cias) underlie both districts. Hydrothermal alteration as- and reference spectra. In order to make such compari-
sociated with precious metal concentration at Bodie is sons, imaging spectrometer data must be processed to
probably the result of intense thermal spring activity, retrieve ground reflectance values from radiance mea-
which still occur along fault zones in this region (Silber- surements. A number of methods have been proposed
man et al., 1995). for such conversion, some of which require the input of
Mapping Hydrothermal Alteration Minerals 311
ground data collected simultaneously with the sensor’s tors’ length. SAM produces an image with the pixels it
overflight. Radiative transfer methods rely on the use of manages to classify assigned to the respective reference
atmospheric models and allow the retrieval of apparent minerals, constrained by a user-specified threshold, to-
ground reflectance from the radiance values provided by gether with a set of “rule images,” one for each reference
AVIRIS. A comparison of several methods by Clark et mineral used. DN values in these “rule images” are the
al. (1995) showed that a hybrid method, which combines expression of the angle itself (smaller DNs indicate
simultaneous ground-based data and the radiative trans- greater similarity to the reference mineral). SAM, on the
fer method, provides the best results for reflectance re- other hand, is not well suited in determining small spec-
trieval in hyperspectral data, followed by the sole use of tral differences between mineral species and is highly
the radiative transfer method developed by Green et dependent on the wavelength ranges and on the thresh-
al. (1993a). olds selected, which are arbitrary.
We used the latter method, which uses AVIRIS lab- We used the version of SAM in the ENVI image pro-
oratory-calibrated radiance data (Chrien et al., 1990), in cessing software package (Research Systems, Inc., 1995).
conjunction with in-flight calibration data obtained over
Rogers Dry Lake at the beginning of the 1992 AVIRIS USGS Tricorder Algorithm
data acquisition season (Green et al., 1993b), and the Tricorder is still under development at the Spectroscopy
MODTRAN radiative transfer code. The method com- Laboratory of the U.S. Geological Survey. It was de-
pensates for AVIRIS-derived estimates of water vapor, signed to compare spectra of materials from the USGS
aerosol, and surface on a pixel by pixel basis. Compared Digital Spectral Library to image spectra acquired by
to other radiative transfer methods, it provides a better hyperspectral sensors, analyzing simultaneously for multi-
correction for H2O and other atmospheric gases, as a ple minerals, using multiple diagnostic spectral features
function of elevation throughout a scene (Clark et al., for each mineral (Clark and Swayze, 1995).
1995). Tricorder works by first removing a continuum from
spectral features in the reference library spectra and also
IMAGE PROCESSING METHODS FOR from each spectrum in the image data set. Continuum-
MINERAL MAPPING removed library and image spectra are then compared
using a least square procedure. The algorithm uses a lin-
In this study we compared two different methods for ear correlation coefficient to determine the fit by individ-
mineral mapping at Bodie and Paramount: Spectral ually fitting all diagnostic absorption features of all refer-
Angle Mapper (Kruse et al., 1993) and Tricorder (Clark ence spectra in the library to each pixel spectrum. An
et al., 1990; Clark and Swayze, 1995). Both methods overall fit is then calculated, by weighing each individual
compare spectra from pixels in the scene and reference fit based on the area under its continuum-removed spec-
spectra from a spectral library, but use different algo- tra. The overall fits are then compared and the mineral
rithms to compare and measure spectral similarity. The from the library with the best fit is assigned to the im-
two methods are relatively insensitive to illumination dif- age pixel.
ferences due to topography in areas of low to moderate One of the strengths of Tricorder is that it considers
relief and high sun angle. The library used for this as- three attributes in the analysis: the depth of particular
sessment was the public domain USGS Digital Spectral absorption features, the “goodness of fit,” and the reflec-
Library which contains nearly 500 reference spectra of tance level of the continuum at the center of the feature.
minerals, vegetation, and other surface materials (Clark By doing this, it can analyze feature shapes using all data
et al., 1993). points and resolve even complex features, such as dou-
blets in minerals like kaolinite, dickite, and halloysite and
Spectral Angle Mapper Classifier small variations in mineral species.
The Spectral Angle Mapper (SAM) is a supervised classi-
fication technique that measures the similarity of image ALTERATION MAPPING AT BODIE
spectra to reference spectra from a spectral library, or AND PARAMOUNT
from field and laboratory spectra. SAM measures similar-
ity by calculating the angle between the two spectra, The results of alteration mapping discussed in this sec-
treating them as vectors in n-dimensional space, with n tion uses the locations of Bodie Bluff and Silver Hill,
being the number of bands (Kruse et al., 1993). The both in the Bodie district, and Paramount as references
angle is the arc-cosine of the dot product of the two (see Fig. 1).
spectra. Smaller values for the angle indicate higher sim-
ilarity between pixel and reference spectra. This method SAM Results
is not affected by gain (solar illumination) factors, since Two spectral regions were processed separately: the
the angle between two vectors is independent of the vec- visible/near-infrared (VNIR) region (0.4–1.3 lm) and the
312 Crósta et al.
shortwave infrared (SWIR) region (2.0–2.4 lm). The and three reference spectra on a low-end Unix workstation
VNIR bands were analyzed for minerals with diagnostic (Sun IPX).
electronic transition features (hematite, goethite, jarosite,
etc.) and the SWIR bands were analyzed for hydroxyl- Tricorder Results
bearing minerals and carbonates. Tricorder is run through command files, which are set to
In the SWIR, 46 AVIRIS bands were used with 15 analyze specific groups of minerals. The command file
reference spectra. The threshold value used for the maxi- used included 126 materials (minerals, vegetation, and
mum angle between spectral vectors was established em- ice). Among the minerals analyzed were some of the
pirically at 0.065 radians. Values above this produced a most common alteration minerals. All the minerals ana-
large number of classified pixels in only one or two classes lyzed using SAM were included in Tricorder’s com-
and values below it produced images of mostly “nonclassi- mand file.
fied” pixels. SAM classified most of the alteration at Bodie Tricorder also analyzes separately for multiple min-
and Paramount as Na-montmorillonite, Ca-montmorillo- erals in the VNIR and SWIR spectral regions. Therefore,
nite, and a mixture of kaolinite⫹smectite (Fig. 2). This the analysis of the whole spectrum for a single pixel may
mixture was concentrated predominantly at Paramount, find two different minerals, one for each spectral region.
with smaller occurrences at three locations at Bodie: Silver Figures 4 and 5 show the mineral maps produced
Hill, near the center of district, and Bodie Bluff. At Para- by Tricorder for the SWIR and VNIR, respectively. Com-
pared to results from SAM, these produced a greater
mount, a clear zonation was revealed with most of the
number of spectral classes relatively to mineral species.
western part classified as a mixture of kaolinite⫹smectite,
In the SWIR region, Tricorder distinguished between
surrounded by pixels classified as Na-montmorillonite, and
Na- and Ca-montmorillonite in most of the Bodie district
the eastern part classified as Na- and Ca-montmorillonite.
and at the eastern part of Paramount. It also distinguished
SAM identified the two types of kaolinite used as refer- two types of kaolinite (well and poorly crystalline) in the
ences (well and poorly crystalline) in a very small number western part of Paramount, a mixture of kaolinite⫹smec-
of pixels. Illite and Ca-montmorillonite were mapped tite in the central part of the Bodie district, halloysite in
mostly as isolated pixels scattered throughout the scene the western part of Paramount, plus a few pixels as Al-
with no meaningful pattern. We treated these as misclassi- muscovite mostly at Silver Hill and the center of the Bo-
fications and removed them from the final result. die district, and illite in mine tailings at Silver Hill. Dolo-
Computation time for the SWIR 15-mineral classifica- mite was given a good fit by Tricorder, with a fair number
tion using 46 AVIRIS bands on a low-end Unix worksta- of pixels being mapped as such, but these pixels were iso-
tion (Sun IPX) was 8 min. lated and scattered all over the image without any mean-
In the VNIR, 94 AVIRIS bands were used with nine ingful geological pattern. We treated these as misclassifi-
reference spectra of common Fe2⫹- and Fe3⫹-bearing cations and removed them from the final result, together
minerals. Initial results were not satisfactory because SAM with other minerals with low fits and small number of
classified most pixels in the entire scene as a mixture of pixels: dolomite, phlogopite, hectorite, diaspore, clintonite,
hematite and quartz. Even after experimenting with differ- palygorskite, elbaite, and nontronite.
ent values for the minimum angle and removing the he- In the VNIR, Tricorder mapped two different types of
matite⫹quartz spectrum from the reference spectra, SAM hematite: the first one, a fine-grained variety, was mapped
classified most of the pixels in the scene as a single refer- around hydrothermally altered areas at Bodie and Para-
ence mineral. We then decided to use only three refer- mount, and the other, a mixture of hematite and quartz,
was distinctively associated with the alteration at Bodie
ence spectra: hematite, goethite, and jarosite. Some em-
Bluff. It also mapped goethite in the center of the Bodie
pirical work had to be done to find the threshold value
district and in a nearby tailings pond. A few pixels scat-
for the maximum angle, which was 0.25 radians. We no-
tered across the Bodie district were mapped as K-jarosite,
ticed that changes in the value of this threshold produced while some others were classified as Fe2⫹-bearing miner-
significant changes in the number of pixels classified as als (mostly cummingtonite) at the center of district and
one of the reference minerals. at Silver Hill.
The results showed goethite in altered areas at Bodie The algorithm took around 4 h to process the 141
Bluff and Silver Hill, together with jarosite in mine dumps MB scene for 126 minerals in the VNIR and SWIR run-
and in a tailings pond (Fig. 3). A few pixels were mapped ning on a medium-performance Unix workstation (HP
as hematite at Silver Hill and in an area to the north of 9000) at the USGS Spectroscopy Laboratory.
the major alteration. SAM also mapped goethite in the
Paramount district as well as jarosite, which also appeared Comparison of Results Obtained by SAM
scattered throughout the western part of the scene, and and Tricorder
along dirt roads. A spatial comparison between the areas of major alter-
It took 15 min to run SAM using 94 AVIRIS bands ation identified by both methods reveals that Na-mont-
Mapping Hydrothermal Alteration Minerals 313
Figure 2. SAM classification results for the SWIR bands (2.00–2.45 lm) of the AVIRIS scene over Bodie and Paramount.
Figure 3. SAM classification results for the VNIR bands (0.42–1.34 lm) of the AVIRIS scene over Bodie and
Paramount.
314 Crósta et al.
morillonite was identified by both methods at approxi- match to the kaolinite spectrum from the library, sug-
mately the same locations at Bodie and the eastern part gesting a possible mineral zonation of the alteration. Table
of Paramount. SAM, however, mapped a larger number of 1 shows the location of these sites on the scene, a com-
pixels as Ca-montmorillonite in both districts, particularly parison of the minerals assigned to them by Tricorder
at eastern Paramount, in areas, which Tricorder assigned and SAM and the result of XRD analysis.
to halloysite, and both types of kaolinite. The mixture of Rock samples from each site were spectrally ana-
kaolinite⫹smectite shows good correlation between both lyzed using a Beckman UV-5240 spectrophotometer. The
methods only at the center of the Bodie district, whereas resulting spectra were convolved to AVIRIS bandwidths.
another major area of occurrence for this mineral mapped The results of the analyses are shown in Figures 6–10.
by SAM at the western part of Paramount was mapped Each sample spectrum was plotted together with the
by Tricorder as both types of kaolinite and as halloysite. spectrum of the corresponding AVIRIS image pixel. The
The separation achieved by Tricorder, between Na- AVIRIS spectra were edited to remove atmospheric H2O
montmorillonite, halloysite, and poorly crystalline kaolin- and noisy bands. The reference spectra of the mineral
ite at the eastern part of Paramount, did not appear in assigned to the pixel by the classification algorithm(s)
the SAM results. were also plotted for comparison.
In the VNIR, major differences show up in the com- Figure 6 shows the spectra of sample BO-17 (from
parison between results obtained by both methods for Paramount) and of the AVIRIS’s pixel corresponding to
those minerals that could be mapped by SAM (hematite, its location, together with the library spectra for Na-
goethite, and jarosite). SAM did not find much hematite montmorillonite. The main hydroxyl absorption feature
in the whole scene, whereas Tricorder mapped signifi- near 2.2 lm is broader in the AVIRIS spectrum than in
cant clusters of pixels as hematite at both districts. How- the sample and library spectra. A secondary feature near
ever, some of the pixels mapped as hematite by Tri- 2.3 lm can be seem in the AVIRIS spectrum, which does
corder were assigned to goethite by SAM. On the other not appear in the sample spectrum, suggesting the occur-
hand, pixels that Tricorder mapped as goethite were as- rence of a mixture of illite/muscovite. XRD analysis showed
signed to jarosite by SAM. the presence of illite and kaolinite in sample BO-17 (Ta-
ble 1), instead of montmorillonite. Although it is difficult
to generalize for an entire pixel the results obtained from
SPECTRAL ANALYSIS
a single sample, it appears that both classifiers were not
In order to assess the results, we selected five sites for able to identify the correct mineralogy at this location
ground checking and collected samples for laboratory based on the pixel spectrum. For instance, the presence
spectral analysis. Sample analysis also included X-ray dif- of kaolinite, identified by XRD, did not show on the
fraction (XRD). These sites correspond to areas where one AVIRIS spectrum for this pixel.
or more alteration minerals were mapped in the VNIR Sample BO-11a (Fig. 7) corresponds to the site at
and SWIR. Two of the sites are located at Bodie, corre- Bodie Bluff classified as montmorillonite by Tricorder
sponding to a mine excavation at Bodie Bluff (sample and SAM. Tricorder assigned this pixel to Ca-montmoril-
BO-11a) and to a nearby tailings pond (sample BO-05). lonite, whereas SAM classified it as Na-montmorillonite.
The other three sites are located at Paramount (BO-14a, The pixel spectrum is quite similar to the one in Figure
BO-14b, and BO-17), corresponding to natural outcrops 6, suggesting also the occurrence of subpixel mixtures of
of altered rocks. Samples were broken off with a ham- illite/muscovite and quartz. XRD analysis revealed only
mer, except for sample BO-05, collected from fine- the presence of cristobalite, with a very strong peak. How-
grained sediments on the floor of a dry tailings pond. A ever, the fact that XRD did not show any hydroxyl-bearing
sixth site, located at Paramount, had been selected for mineral could be related to the strength of the cristobalite
sampling, but could not be reached on the ground. The peak, which may have masked other minerals.
pixel spectrum of this sixth site showed a remarkable Figure 8 shows the spectrum of sample BO-14b
Table 1. Comparison of Results between SAM and Tricorder for Selected Sitesa
Pixel Location
Sample (row/line) Tricorder SAM X-Ray Diffraction
BO-17 481/435 (Paramount) Na-montmorillonite Na-montmorillonite Illite, kaolinite
BO-11a 161/352 (Bodie) Ca-montmorillonite Na-montmorillonite Cristobalite
BO-14b 589/445 (Paramount) Kaolinite (pxyl) & goethite Kaolinite/smectite & jarosite Kaolinite
BO-14a 485/442 (Paramount) Halloysite Na-montmorillonite Montmorillonite, kaolinite, (illite)
BO-05 78/344 (Bodie) Na-montmorillonite & goethite Na-montmorillonite & jarosite Goethite
* 524/352 (Paramount) Kaolinite Kaol./smectite *
a
Spectra for each site are shown in Figures 6–11, respectively. *The location with an asterisk did not have a sample collected.
316 Crósta et al.
Figure 6. Plot showing the Beckman spectrum for sample Figure 8. Plot showing the Beckman spectrum for sample
BO-17 (from Paramount), the reference spectrum for Na- BO-14b (from Paramount), the reference spectrum for
montmorillonite extracted from the USGS Digital Spectral poorly crystalline kaolinite and the AVIRIS spectrum identi-
Library and the AVIRIS pixel spectrum identified as Na- fied as such by Tricorder (SAM indicated a mixture of kao-
montmorillonite by SAM and Tricorder. linite and smectite). Spectral features characteristic of goe-
thite are also present in the AVIRIS and Beckman spectra,
identified as such by Tricorder and as jarosite by SAM.
CONCLUSIONS
We managed to recognize and map a variety of minerals
associated with hydrothermal alteration processes using
AVIRIS data at the Bodie and Paramount mining districts,
California. The method selected for calibrating the data
and for retrieving reflectance from radiance was precise
Figure 10. Plot showing the Beckman spectrum for sample enough to allow mineral mapping without the use of inde-
BO-05 (from Bodie), the reference spectrum for Na-mont- pendent ground and atmospheric data. This represents a
morillonite and for goethite and the AVIRIS spectrum identi- major advantage for mineral exploration activities, in
fied as goethite in the VNIR bands and as Na-montmorillo-
nite in the SWIR bands. (SAM indicated Na-montmorillonite which ground data acquisition is usually not possible dur-
and jarosite). ing the initial phases of these activities.
Results obtained for the SWIR region showed a rela-
tively good coincidence between the two methods for
suggests the presence of a doublet near 2.2 lm, which mapping the alteration minerals of major occurrence in
is characteristic of halloysite, as seem on the library spec- the study area. Both methods showed an east–west zona-
trum. The match between the sample and pixel spectra, tion in the alteration at Paramount and also some zoning
however, is not good at this wavelength. The broadness of at Bodie. The zonation at Paramount is clearer and, as
the absorption feature on the AVIRIS spectrum does sug- shown by Tricorder, has predominantly kaolinite and hal-
gest some sort of mineral mixture at subpixel level. This loysite in the western and southwestern portions and
is confirmed by XRD, which identified the presence of predominantly Na-montmorillonite in the eastern por-
montmorillonite, kaolinite, and some illite in this sample. tion, with a dividing line coinciding with Atastra Creek.
Figure 10 shows sample BO-05 (from Bodie), col- This shows that only the eastern part of the Paramount
lected at the tailing pond where the AVIRIS spectrum in- District exhibits an alteration similar to the Bodie Dis-
dicates the presence of Fe3⫹ and O—H-bearing minerals.
This location was identified by Tricorder as containing Na-
montmorillonite and goethite. SAM also assigned the oc-
currence of Na-montmorillonite and jarosite, instead of Figure 11. Plot showing the reference spectrum for well-
goethite. A comparison of the library spectrum of Na- crystalline kaolinite and the AVIRIS pixel spectrum identi-
fied as such by Tricorder and as a mixture of kaolinite and
montmorillonite and the pixel spectrum does not show a smectite by SAM. No sample was collected at this locality.
good match for the feature at 2.2 lm in terms of shape,
although the position seems to be correct. Again, the
broadness of this feature as seem on the pixel spectrum
is likely to be related to subpixel mixtures. The main Fe3⫹
feature at approximately 0.92 lm shows a very good match
to the goethite feature. This is confirmed by XRD, which
identified goethite in this sample and no hydroxyl-bear-
ing mineral.
Figure 11 shows the AVIRIS pixel and library spectra
for the location that could not be sampled on the ground
(marked with an asterisk in Table 1). Both classification
algorithms identified the presence of kaolinite at this loca-
tion, based on the well-defined doublet absorption near
2.2 lm, characteristic of well-crystalline kaolinite. An in-
spection of other pixel spectra near this location also re-
vealed similar well-defined doublet absorption. This sug-
gests a predominance of kaolinite over other alteration
318 Crósta et al.
in 1992 and 1993. In Summaries of the Fourth Annual JPL Silberman M. L., Breit, F., and Lawrence, E. F. (1995), Geology
Airborne Earth Science Workshop, 25–29, October, JPL and Ore Deposits of Bodie Hills, Northern Mono Basin Re-
Publ. 93-26, Jet Propulsion Laboratory, Pasadena, CA, Vol. gion, Special Publ. 22, Geological Society of Nevada, 49 pp.
1, pp. 69–72. Research Systems, Inc. (1995), ENVI User’s Guide, Version 2.0.
Herrera, P. A., Closs, L. G., and Silberman, M. L. (1993), Al- RSI, Boulder, CO.
teration and geochemical zoning in Bodie Bluff, Bodie min- Taranik, J. V., and Crósta, A. P., (1996), Remote sensing for
ing district, eastern California. J. Geochem. Exploration 48: geology and mineral resources: an assessment of tools for
259–275. geoscientists in the near future. In XVIII Congress of the
Kruse, F. A., Lefkoff, A. B., Boardman, J. W., et al. (1993), The International Society of Photogrammetry and Remote Sens-
Spectral Image Processing System (SIPS)—interactive visual- ing, Vienna, Austria, 9–19, July, International Archives of
ization and analysis of imaging spectrometer data. Remote Photogrammetry and Remote Sensing, Vol. XXXI, Part B7,
Sens. Environ. 44:145–163. pp. 689–698.