ict and
Mono
is area
ft. The
0.44 x
lution,
Covers
of the
arked
cedony
ork by
in the
' Zones
‚one of
or the
zation,
onnage
logical
9 those
similar
ANCE
Identification of surface materials using hyperspectral data is
based on quantitative comparisons between pixel and reference
spectra. In order to make such comparisons, imaging
spectrometer data must be pre-processed to retrieve ground
reflectance values from radiance measurements. ^ Several
methods have been proposed for such conversion, some of
which require the input of ground data preferentially collected
simultaneously with the sensor's overflight. Radiative transfer
methods rely on the use of atmospheric models and allow the
retrieval of apparent ground reflectance from the radiance
values provided by AVIRIS. A comparison of several methods
by Clark et al. (1995) showed that a hybrid method which
combines simultaneous ground-based data and the radiative
transfer method developed by Green et al. (1993a), provides the
best results for reflectance retrieval in hyperspectral data,
followed by the sole use of Green's radiative transfer method.
We selected Green's radiative transfer method because it meets
the requirements of this study that no independent ground or
atmospheric measurement be utilized. This method uses the
AVIRIS laboratory-calibrated radiance in conjunction with in-
flight calibration data obtained over Rogers Dry Lake at the
beginning of the 1992 AVIRIS data acquisition season, and the
MODTRAN radiative transfer code (Green et al. 1993b). The
method compensates for AVIRIS derived estimates of water
vapor, aerosol and surface on a pixel by pixel basis. Compared
to other radiative transfer methods, it provides a better
correction for H,O and other atmospheric gases, as a function
of elevation throughout a scene (Clark et al, 1995).
Disadvantages of this method include: (i) errors in the ground
measurements during the in-flight calibration at the beginning
of the season will propagate into the derived surface
reflectance; (ii) computation time is considerable and (iii) the
method is still under development at JPL and not yet available
for general use.
6. IMAGE PROCESSING METHODS FOR MINERAL
MAPPING
In this study we compared two different image processing
techniques for mineral mapping at Bodie and Paramount:
Spectral Angle Mapper (Kruse et al, 1993) and Tricorder
(Clark et al, 1990; Clark and Swayze, 1995). Both methods
compare spectra from pixels in the scene and reference spectra
from a spectral library, using different algorithms to compare
and measure spectral similarity. The two methods are relatively
insensitive to illumination differences due to topography in
areas of low to moderate relief and high sun angle. The library
used for this assessment was the public domain USGS Spectral
Library which contains nearly 500 reference spectra of
minerals, vegetation and other surface materials (Clark et al.,
1993).
6.1 Spectral Angle Mapper Classifier
Spectral Angle Mapper (SAM) is a supervised classification
technique that measures the spectral similarity of image spectra
to reference spectra which can be obtained either from a
spectral library or from field and laboratory spectra. SAM
defines spectral similarity by calculating the angle between the
two spectra, treating them as vectors in n-dimensional space,
With 7 being the number of bands used (Kruse et al., 1993).
163
Small values for the angle represent higher spectral similarity
between pixel and reference spectra. This method is not
affected by gain (solar illumination) factors, since the angle
between two vectors is invariant with respect to the lengths of
the vectors. SAM produces an image with the pixels it manages
to classify assigned to the respective reference minerals,
constrained by a user-specified threshold, together with a set of
"rule images", one for each reference mineral used. DN values
in these "rule images" are the expression of the angle itself
(smaller DNs indicate greater similarity to the respective
reference mineral) and these images can be used to assess
individual results for each reference mineral.
The advantages of SAM are that it has already been
implemented in commercial imaging processing packages, it is
easy to use and it is computationally fast. SAM's
implementation used for this study is part of ENVI image
processing software (Research Systems, Inc, 1995).
6.2 USGS Tricorder Algorithm
Tricorder is still under development at the Spectroscopy
Laboratory of the U.S. Geological Survey and is expected to be
released soon for general use. It was designed to compare
spectra of materials from the USGS Digital Spectral Library to
image spectra acquired by hyperspectral sensors, analyzing
simultaneously for multiple minerals, using multiple diagnostic
spectral features for each mineral (Clark and Swayze, 1995).
Tricorder works by first removing a continuum from spectral
features in the reference library spectra and also from each
spectrum in the image data set. Both continuum-removed
spectra are then compared using a modified least square
procedure.
One of the strengths of Tricorder is that it considers several
attributes in the analysis, such as the depth of particular
absorption features, the "goodness of fit" and the reflectance
level of the continuum at the center of the feature. By doing
this, it can analyze feature shapes using all data points and
therefore resolve even complex feature shapes such as doublets
in minerals like kaolinite, dickite and halloysite. The method,
like SAM, is also not affected by illumination differences in
areas of low to moderate relief.
7. ALTERATION MAPPING AT BODIE
AND PARAMOUNT
The results of alteration mapping discussed in this section uses
the locations of Bodie Bluff (BB) and Silver Hill (SH), both in
the Bodie district, and Paramount (PA) as references (see
Figure 1).
7.1 SAM Results
Two spectral regions were processed separately in SAM: the
visible/near-infrared (VNIR) region (0.4 to 1.3 mm) and the
shortwave infrared (SWIR) region (2.0 to 2.4 mm). The VNIR
bands were analyzed for minerals with diagnostic electronic
transition features (hematite, goethite, jarosite, etc.) and the
SWIR bands were analyzed for hydroxyl-bearing minerals and
carbonates. Spectra used as reference were obtained from the
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
