International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
-
3.3. Spectral Indices Method
Spectral indices method is a kind of orthogonal
transformation to be used for surface mineralogical
mapping by using the five ASTER SWIR bands
(Yamaguchi and Naito, 2003). The spectral indices (SI)
method contains mainly five indices, namely, brightness,
alunite, kaolinite, calcite and residual indices,
respectively.
In general, spectral indices in 6-dimensional (SWIR
channels only) spaces can be defined as values measured
by projected data points onto imaginary axes with
appropriate unit vector directions. Jackson (1983) states
that it is kind of orthogonal transformation and the
transformed axes are designated to denote specific
spectral characteristics. To find out alunite and kaolinite
minerals, coefficients of the SWIR bands are obtained in
Table 2 (Yamaguchi and Takeda, 2003).
Table 2. Transform Coefficients of spectral indices for
the ASTER SWIR bands (Yamaguchi and Takeda, 2003)
Spectral
index
Alunite — -0.511 -0.003 0.802 0.059 -0.304
Kaolinite 0663. -03360 “0337 -0.311 - 0.191
Band5 Band6 Band7 Band8 Band9
All SWIR channels were multiplied with their respective
special coefficients for alunite and kaolinite minerals,
then all calculated SWIR channels were added to obtain
indices for those minerals.
4. RESULTS AND DISCUSSIONS
After applying these two methods, the results were
checked using existing occurrence data of the study area
(Figure 1). According to reference data (occurance map),
19 hydrothermally altered locations containing both
alunite and kaolinite together were compared using
existing data and results of two different methods.
Resultant imageries have 32-bit DN values, then, to
distinguish the minerals from the surrounding pixels or
materials, threshold values for each resultant images were
computed statistically by using mean value and standard
deviations. The 13 of the samples were identified as
alunite by using SI method. On the other hand, in BR
method, kaolinite and alunite were discriminated in 11
sites and 8 sites, respectively. As a result, SI and BR
techniques detected alunite and kaolinite minerals. SI
method gives better results for detecting alunite than that
of kaolinite. On the contrary, BR method can be more
applicable to discriminate kaolinite from alunite.
Table 3 gives the DN value (32-bit) of 19 locations for
each method and mean values of the related images.
Underlined values in the table are the values above the
corresponding threshold level of the each resultant image.
The resultant whole image data are normally distributed
therefore, threshold level is accepted as mean value of the
80
related images during the study (Fung and LeDrew,
1988). It is also possible to get optimum threshold value
as mean + standard deviation value. However, as the aim
is to obtain the lowest threshold value to check the
resultant data, mean values are used as the threshold
value. Threshold values of the SI method for alunite and
kaolinite mineral distinction are 4.034 and 4.54,
respectively. According to corresponding threshold
values, total 13 samples were clearly discriminated as
alunite via SI method. Sample number 3, 6, 13, 14, 17,
and 18 have DN values lower than mean value.
Therefore, SI method fails on these 6 samples for alunite
mineral detection. In addition, SI method could not
correctly distinguish kaolinite mineral in selected 19
sites. Using the mean value as threshold level, only 3
samples (sample no: 3, 4, and 10) were correctly detected
as kaolinite in this method.
On the other hand, by using band ratio (BR) method 8
samples having the threshold value of 1.304 and 11
samples having threshold value of 0.819 were correctly
identified as alunite and kaolinite, respectively. These
values are also the mean values of corresponding images
of band ratio results.
SI method is more successful for discrimination of the
alunite minerals than band ratio method. On the contrary,
the results show us that band ratio method is superior to
SI method in discriminating kaolinite minerals.
The results could be tested by changing the threshold
value. For instance, new threshold value as mean +
standard deviation value, 7 and only | samples are
correctly defined as alunite and kaolinite minerals,
respectively. Similar situation also appears in BR
method. Correctly detected samples for alunite and
kaolinite decrease from 8 to 4 and from 11 to I,
respectively.
5. CONCLUSIONS
In this manuscript, two different approaches are
examined to compare for detecting alunite and kaolinite
minerals on the study area. One of the techniques is band
ratioing method which is performed by division of
reflected channel to absorption channel. In our study,
optimum ratio for detection of alunite minerals is band 4
over band 5. Furthermore, the optimum band ratio for
kaolinite minerals is the division of band 7 to band 6.
Another technique tested in this study is spectral indices
method (SI) which is kind of orthogonal projection of
SWIR channels. This method is quite successful for
detection of alunite minerals. However, it failed in
discrimination of kaolinite minerals. During the SI
method application, coefficients of the alunite and
kaolinite indices are taken from Yamaguchi, Y., and
Takeda (2003). Nevertheless, their study was carried out
in Nevada, USA. This area almost desert area. Therefore,
coefficients of SI should be re-calculated for non-desert
or semi-vegetated areas.
Internation
Tie tiet t
Figui
mine
mine
