Professional Documents
Culture Documents
Alteration mineral mapping with ASTER data by integration of coded spectral ratio
imaging and SOM neural network model
1, 1 2
Mohammad Hassan TAYEBI *, Majid HASHEMI TANGESTANI , Robert Keller VINCENT
1
Department of Earth Sciences, Faculty of Sciences, Shiraz University, Shiraz, Iran
2
Department of Geology, Bowling Green State University, USA
Abstract: This study applied coded spectral ratio imaging and a self-organizing map (SOM) neural network model for subpixel
classification of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data to map alteration minerals of the
Masahim volcano, SE Iran. Spectra of samples were measured using a FieldSpec spectroradiometer, and were interpreted for finding
key alteration minerals. Key minerals including sericite and kaolinite were selected. Thirty-six nonreciprocal spectral ratios of 9 ASTER
reflective bands were created. Each spectral parameter (of 9 spectral bands and 36 spectral ratios) for the field and library mineral
spectra was divided into deciles. Each decile of spectral bands or ratios was labeled from 9 for the highest down to 0 for the lowest decile.
Spectral ratio code signature patterns of field samples were then compared to the US Geological Survey (USGS) spectral library to find
3 spectral ratio images for each key mineral. Based on these 3 ratio images, color ratio images were created and training areas were
selected by a geographic information systems (GIS) overlay approach and were used to train the SOM model. This model was applied
to the 5, 15, and 36 ratio image data sets as chemical composition images, and the results were compared using confusion matrices
and coefficient of determination (R2). The results of SOM applied to 15 and 36 ratio image data sets showed good identification of key
minerals. We achieved overall accuracies of 87% for SOM applied to the 36 ratio images, 86% for 15, and 85% for 5 ratio image data sets.
The R2 produced from 15 ratio image data sets for sericite and kaolinite were 0.35 and 0.52, whereas they were 0.64 and 0.68, for 36 ratio
image data sets. It was concluded that using SOM applied to ratio images could, therefore, be useful in alteration mapping of a volcano.
Key words: Alteration mapping, Masahim volcano, SOM neural network model, coded spectral ratio imaging, GIS
illumination, and sensor electronic gain and offset that are and geomorphometric feature analysis (Ehsani and
not always perfectly recorded or accounted for (Vincent, Quiel, 2008). The SOM is a 2-layer structure, including
1997). A second approach is an empirical approach, 1 input layer and 1 output layer. Therefore, the dilemma
whereby the corrections for atmospheric absorption and of determining the hidden layer size is avoided (Hu and
scattering, solar illumination variations with topographic Weng, 2009). It is capable of coping with presence-only
slope, and sensor electronic gain and offset are made data (Li and Eastman, 2006), it is not affected by the local
from the image data itself, and then selected spectral minima problem and it is insensitive to the structure of the
parameters of each pixel are compared to the reflectance codebook vector map (Lee and Lathrop, 2006).
spectrum of laboratory and field spectra of user-selected There were several questions raised at the onset of this
objects to search the image for objects that have a similar study. Is it efficient for us to use the SOM model for subpixel
chemical composition. Dark-object–corrected spectral classification of ratio images as chemical composition
ratio imaging is such an empirical approach, in which the images? How many ratio images do we need to input to the
shadow/slope factor and additive factor (the dark object, SOM to get good subpixel classification results? The study
or DO) from atmospheric haze and electronic offset of the area is the Masahim composite volcano, which is situated
sensor, almost constant for the entire data set, are removed in western Kerman, Iran. In this study, a coded spectral
(Vincent, 1997). This becomes less accurate if the elevation ratio imaging method was used to find suitable training
differences within the scene of interest are very large, but areas and a SOM model was applied to the different ratio
this problem also occurs in the theoretical approach. image data sets, including 5 and 15 short wave infrared
It has been shown that dark-object–corrected spectral (SWIR) and 36 visible near infrared and short wave
ratio images enhance chemical compositional differences, infrared (VNIR+SWIR) ratio images, respectively.
while suppressing variations in slope, solar illumination,
and atmospheric haze (Vincent, 1972, 1997; Rowan et al., 2. Geological setting
1974; Perry and Vincent, 2009). Triplets of spectral ratios The Masahim composite volcano (55°07ʹ–55°30ʹE;
displayed in colors of red, green, and blue (as a ratio color 30°10ʹ–30°30ʹN) lies 265 km west of Kerman, SE Iran
composite) are commonly used to enhance the appearance (Figure 1); it is situated in the Dehaj–Sarduiyeh district,
of specific minerals of interest, where they are exposed which is a part of the Urumieh-Dokhtar Magmatic Arc
on the ground (Perry and Vincent, 2009). However, the (UDMA). The Masahim volcano was formed between 6.8
problem is that the color ratio images cannot show the ± 0.4 (older stratigraphic level) and 6.3 ± 0.9 (younger
fractions of desired minerals. stratigraphic level) Ma ago (Hassanzadeh, 1993). Its
Detection and quantification of fractions of the target eruptions have been mainly phreato-magmatic (including
materials for each pixel can be important for dark-object– interaction of water and magma) to phreato-plinian (same
corrected spectral ratio mages. Therefore, evaluating interaction with higher plumes), including lava flows,
a model like neural network to map and estimate pyroclastic surges, pyroclastic falls, and pyroclastic flows
abundance fractions of minerals present in ratio images (Ahmadipour, 1994) as the main components.
can be useful for the mapping of volcanoes. Artificial Ahmadipour (1994) divided the Masahim volcano
neural networks (ANNs) were originally designed according to petrographic and structural characteristics,
as pattern-recognition (DuBose and Klimasauskas, as well as distances to their origins, into 4 main facies:
1989) and data analysis tools that mimic the neural central (caldera), proximal, medial, and distal, with several
storage and analytical operations of biological neural units in each. The central facies mainly consists of lava
networks (DuBose and Klimasauskas, 1989; Villmann breccias, which are rhyodacitic-dacitic in composition and
et al., 2003). ANN approaches have a distinct advantage in some part display intensive hydrothermal alteration, as
over statistical classification methods in that they are well as pre- and postcaldera stage intrusives. Precaldera
nonparametric, require little or no a priori knowledge of stage intrusive rocks generally contain plagioclase, biotite,
the distribution model of input data and no assumption hornblende, and quartz (andesitic-dacitic) that have been
about the probabilistic models of data, are robust in noisy exposed by collapse and erosion, and have undergone
environments, and have the ability to learn complex intensive alteration (Ahmadipour, 1994). At the central
patterns (Ji, 2000) and perform generalizations (Zhuang et facies mafic constitutents are completely altered, as
al., 1991; Geaga et al., 1992; Emaruchi et al., 1997). well as feldspar, and silicification represents the final
Among the ANN techniques, the self-organizing map process (Dimitrijevic, 1973). However, postcaldera stage
(SOM) neural network model has been used in different intrusives (plugs and dikes) lack hydrothermal alteration
fields such as land-use classification (Ji, 2000; Bagan et (Dimitrijevic, 1973), and only display evidence for surficial
al., 2005; Jianwen and Bagan, 2005), lithological mapping weathering. Alluvial and colluvial deposits cover portions
(Mather et al., 1998; Grebby et al., 2010; Bedini, 2012), of the central facies, and are especially concentrated
628
TAYEBI et al. / Turkish J Earth Sci
55°10'0"E 55°20'0"E
dc-a
Ngvc Ec
Etab Iran
dc-a
El
Study area
a
Kerman
Et
a
Pe
Ngp
rs
ia
30°20'0"N
30°20'0"N
G
ul
Ettb
f
dc-a
Urumieh Dokhtar Magmatic Belt
Ngar d Ev
d
Ngpy
Qal
Ngp
dc-a
Meymand
a
Ngvc Ngp
gd Q
MF
Qt2
Sf Q DF
0 1.5 3 6 9
Km
Recent alluvium
Etb Trachyandesitic-
Qt2 Younger gravel fans CF Ets
Ers
Trachybasaltic lava flows
Tuffs-tuffites
Ev
d Diorite
Ngar
Altered rocks
Eab Andesite-basaltic rocks
Ngpy Pyroclastics El Limestones
Hornblende phenoandesite Limestones and marls
NEOGENE
a Elm
(lava flow) Fault
Ers Red sandstones
Ngvc Volcanomict conglomerates
Conglomerates Road
(pyroclastics fall and flow) Ec
Sample sites
Ngp Pyroclastics surge Elf Flysch Residential
area
gd Granodiorite
Figure 1. Location map of the study area. Geological map of northwest Shahr-e-Babak, Iran, showing the
Masahim caldera at center (modified from 1:100,000 geological map of Shahr-e-Babak, Geological Survey
of Iran, 1973).
along the inner slopes of the caldera rim. The lower-most Pyroclastic falls are poorly sorted layers of welded, lithic
exposures in the proximal facies are pyroclastic surge tuff rich in lithic fragments (andesitic-dacitic), which in
deposits (tuff-breccias) that mantle the older intrusive particular show enrichment in lithic fragments towards the
country rocks (Ahmadipour, 1994). top of each layer. Lava flows overlie a thick accumulation
Pyroclastic flows include block and ash with of pyroclastic flows and falls of the volcano and mainly
stratigraphic layering that reflects the history of the consist of andesite and andesite-basalt rocks (Ahmadipour,
volcano, as well as abundant angular blocks embedded 1994). Medial and distal facies mainly contain pyroclastic
in volcanic ash and lapilli (Figure 1, Ahmadipour, 1994). falls and flows.
629
TAYEBI et al. / Turkish J Earth Sci
3. Materials and methods (Research Systems Inc, 2003). The scattering is removed by
VNIR and SWIR datasets of one ASTER (Level 1B, subtracting band minimum from every pixel in the band.
0410120707401112190002) scene, acquired on 5 October Then 3 sets of spectral ratio image data including 36 ratio
2004, were used to map alteration units of the Masahim images from VNIR+SWIR bands, a subset of 15 images
composite volcano. The ASTER instrument measures from SWIR bands alone, and yet smaller subset of 5 ratio
reflected radiation in 3 bands between 520 and 860 nm images from SWIR bands but based on characteristic ratio
(VNIR) and 6 bands from 1600 to 2430 nm (SWIR), with codes of key representative minerals were produced and
15- and 30-m resolution, respectively (Fujisada, 1995). used as inputs to the SOM neural network (Table 1).
The following steps (Figure 2) show brief descriptions of a (2) Field observation, sample collection, spectral
procedure along which an integration of ASTER data and reflectance measurement of fresh and weathered surfaces of
SOM neural network model was implemented to deal with the rock samples using a Fieldspec Pro spectroradiometer
the objectives of the research: and X-ray diffraction (XRD) analysis, and identification
(1) Pre-processing of ASTER data, including geometric of key representative minerals, which are composed of the
correction using 1:25,000 topographic maps and mineral assemblages.
atmospheric correction using the dark object subtraction (3) Calculation of spectral ratio codes for the field
method. Dark object subtraction is a simple empirical samples and the USGS spectral library, determination of 3
procedure (Gupta, 2003) that adjusts the brightness ratio images for each representative mineral, and creation
values (digital number) of a given scene based on image of color ratio composite images.
information. The digital number to subtract from each band (4) Training area selection using overlay (AND and
can be the band minimum, an average, or a specific value XOR logic) method for representative minerals.
Feature maps
GIS overlay method Producing
(AND and XOR) 3 ratio value layers
Accuracy as s es s ment
Figure 2. The flowchart of process steps. After preprocessing, training area is selected using field samples and
representative alteration minerals. Spectral ratio codes of USGS library and field samples are calculated and
compared for finding 3 ratio images for each mineral and color ratio image is created. GIS overlay method
is used to select training area. SOM is then separately applied to 5, 15, and 36 spectral ratio image data sets.
Feature maps are generated for each input. Finally, the results are compared using confusion matrices and of R2
coefficients for accuracy assessment.
630
TAYEBI et al. / Turkish J Earth Sci
Table 1. Ratio image data sets which were input to the SOM neural network.
(5) Application of SOM to spectral ratio image data near 1.4 µm, 1.9 µm, 2.20 µm, and 2.35 µm. However,
sets (5, 15, and 36 ratio image data sets) for subpixel rock sample 100 does not exhibit strong absorption
classification and generation of feature maps. features in 1.4 µm or 2.20 µm. For example, the spectrum
(6) Accuracy assessment and comparison of the of sample 263 shows alteration to sericite+illite, sample
output results via a confusion matrices approach and R2 219 to sericite+illite+hematite, and samples 228 and 245
coefficient. The flowchart of processes is shown in Figure 2. to sericite+goethite (Figure 3a). The resampled spectra
3.1. Training area selection of all rocks exhibit Al-OH absorption bands around
For subpixel classification of ratio image data sets with 2.20 µm, which falls within the spectral limits of ASTER
a neural network, the analyst must select representative band 6 (Figure 3b). Rock samples 218, 222, 266, and 264
pixels as a training area for each of the key minerals. If exhibit multiple absorption features near 0.48 µm, 0.92
the multispectral image contains sufficiently distinct visual µm that can be attributed to the goethite, and near 1.39
cues, it may be possible to find suitable training areas by µm, 1.41 µm, 1.9 µm, 2.17 µm, and 2.20 µm that can be
visual examination. Frequently, however, one must resort attributed to kaolinite (Hunt and Salisbury, 1970). Rock
to additional sources of information, such as field data sample 239 exhibits multiple absorption features near 1.39
or existing maps, to find representative areas for each µm, 1.41 µm, 2.17 µm, and 2.20 µm that can be attributed
class. The process of finding training areas can, therefore, to kaolinite (Hunt and Salisbury, 1970). The spectra of
be rather labor intensive (Schowengerdt, 2007). Since samples 218 and 264 indicate mixtures of kaolinite+mo
coded spectral ratio imaging (Vincent, 1997) information ntmorillonite+goethite, rock sample 239 is a mixture of
enhances chemical compositional differences by using the montmorillonite+kaolinite, and sample 222 is a mixture
most unique spectral characteristics (absorption features) of montmorillonite+kaolinite+goethite (Figure 3c). The
of a given mineral or rock and a neural network has no resampled spectra of rocks (Figure 3d) exhibit an Al-OH
way of finding the most characteristic spectral parameters absorption band around 2.20 µm (coincided on band 6
like the ratio codes, ratio codes information can be used to of ASTER), and a high reflectance in band 4 of ASTER,
select training areas for a given mineral or rock, especially which is typical in argillic (kaolinite) and phyllic (sericite)
when we do not have access to the area and only have alterations. However, the difference between kaolinite and
access to archived library spectra of the USGS and field sericite is in the reflectance of band 5, which is lower in
samples (collected from the study area) as references for kaolinite relative to sericite (Figure 3d).
material identification. The rock samples 229, 232, 233, 236, 237, 238, 259,
3.1.1. Spectral features of caldera (central facies) and 262 typically exhibit multiple absorption features
The spectral reflectance of field samples was measured near 1.4 µm 1.9 µm, and 2.20 µm, and weaker features
using a Fieldspec Pro spectroradiometer (ASD) with near 2.35 µm, 1.83, 2.1 µm, 0.88 µm, and 2.44 µm that can
spectral range of 350–2500 nm, and a quartz-tungsten- be attributed to the sericite+illite. The spectra of samples
halogen (QTH) lamp was used as light source (Figures 236, 238, 259, and 262 show mixtures of sericite+illite+
3 and 4). USGS spectral library (Clark et al., 1993) and hematite (Figures 3e and 4a); samples 237 and 232 show
XRD analysis were used for comparative analysis and goethite, in addition to the preceding minerals (Figure
characterization of field samples. Sample site locations are 3e); samples 233 and 229 show mixtures of sericite+illite
shown in Figure 1. (Figures 3e and 4a), and sample 257 reveals a mixture of
The result of characterizing field samples by jarosite+sericite (Figure 4a). The resampled spectra of all
interpretation of reflectance spectra showed alteration to rocks exhibit Al-OH absorption bands around 2.20 µm,
sericite+illite+goethite+hematite+jarosite (phyllic zone), which falls within the spectral region of ASTER band 6
kaolinite+ montmorillonite+goethite+hematite (argillic (Figures 3f and 4b). Rock samples 250 and 254 exhibit
zone), and chlorite (Figures 3 and 4). Rock samples 219, multiple absorption features near 0.48 µm, and 0.92 µm
228, 263, 245, and 100 display multiple absorption features that can be attributed to goethite, and 1.39 µm, 1.41 µm,
631
TAYEBI et al. / Turkish J Earth Sci
a c e
100 232
264
245
Value (offset for claritiy)
233
266
236
239
263
237
228 222
238
219 218
2000 2500 500 1000 1500 2000 2500 500 1000 1500 2000 2500
500 1000 1500
b 100 d f
232
245 264
Value (offset for claritiy)
266 233
263 239
236
228
222
219 237
218
238
Figure 3. Reflectance spectra of volcano samples measured in the laboratory. (a) Reflectance spectra of lava breccias exposed at
the caldera, and (b) spectra after being resampled to ASTER VNIR+SWIR bands. (c) Reflectance spectra of altered lava breccias
exposed at the caldera and (d) spectra after being resampled to ASTER VNIR+SWIR bands. (e) Reflectance spectra of precaldera
stage intrusive rocks exposed at the caldera and (f) spectra after being resampled to ASTER VNIR+SWIR bands.
1.9 µm, 2.17 µm, and 2.20 µm that can be attributed to tried to map sericite (phyllic zone) and kaolinite (argillic
kaolinite (Figures 4c and 4d). zone) in the study area.
Rock samples 252, 224, and 258 (plugs and dikes, 3.1.2. Coded spectral ratio imaging
Figures 4e and 4f) do not exhibit strong absorption Coded spectral ratio imaging (Vincent, 1997) was
features in the VNIR-SWIR region (Figure 4e). However, developed to enhance chemical compositional differences
some of them exhibit weaker absorption features near by using the most unique spectral characteristics
0.48 µm (goethite), 0.90 µm, 2.2 µm, 2.3 (chlorite), 1.4 (absorption features) of a given mineral or rock. Each
µm, and 1.9 µm. Absorption minima at 1.4 µm and 2.20 digit in a ratio code of a particular dark-object-subtracted
µm, with shallower depth at 1.9 µm, indicate the presence spectral ratio represents the decile of dark-to-bright in that
of hydroxyl groups, rather than just water of hydration, spectral ratio image in which a particular mineral or rock
and can be attributed to weak alteration (Figure 4e). The would be displayed. For instance, for a given spectral ratio
resampled spectra of these rocks also do not exhibit strong image, a ratio code of 0 signifies that a particular mineral
absorption features (Figure 4f). The phyllic and argillic or rock is in the darkest 10% (decile) of all the materials in
alteration units (sericite and kaolinite, respectively) were the spectral library for that spectral ratio image. Likewise,
typically mixed at the center of the volcano. Therefore, we a ratio code of 9 signifies that a particular rock or mineral
632
TAYEBI et al. / Turkish J Earth Sci
a c e 258-E
254 224-E
Value (offset for claritiy)
258-B
257
250
259 224-B
251
262 264-B
229
500 1500 2000 2500 500 1000 1500 2000 2500 500 1000 1500 2000 2500
1000
b d f 258-E
Value (offset for claritiy)
254 224-E
258-B
257
250
259
224-B
264-B 251
262
229
1000 1500 2000 2500 1000 1500 2000 2500 1000 1500 2000 2500
X a xis : Wa ve le ng th (nm)
Figure 4. Reflectance spectra of volcano samples measured in the laboratory. (a) Reflectance spectra of precaldera stage intrusive
rocks exposed at the caldera and (b) spectra after being resampled to ASTER VNIR+SWIR bands. (c) Reflectance spectra of
precaldera stage intrusive rocks exposed at the caldera and (d) spectra after being resampled to ASTER VNIR+SWIR bands.
(e) Reflectance spectra of postcaldera stage intrusive rocks exposed at the caldera and (f) spectra after being resampled to ASTER
VNIR+SWIR bands.
is brighter in that spectral ratio image than at least 90% of ratio code 9); and for kaolinite were 7/5(with ratio code 9),
the materials included in the spectral library. To produce a 7/6 (with ratio code 9) and 6/5 (with ratio code 1). Figure 6
color image that shows the mineral you wish to map as red, shows the resulting color ratio image for enhancing sericite
you must find 3 spectral ratios (from the ratio code tables as yellow (Figure 6a, where the ratios with ratio codes of
for that particular sensor) that have ratio codes of 9, 0, and 9 were displayed as red and green, and the one with the
0 and display them as red, green, and blue, respectively. lowest ratio code was displayed as blue), and kaolinite as
yellow (Figure 6b, similarly displayed).
3.1.3. Selecting training area using GIS
The brighter (ratio code of 9) or darker pixels (ratio
The spectral ratio codes of 13 spectra of sericite, 8 spectra code of 0) in the spectral ratio images indicate a better
of kaolinite from the USGS spectral library (Table 2), and match to the reference material in those pixels. However,
field samples were compared to select the best ratio images higher and lower values could be artifacts of the shadow
to provide contrast between the minerals of interest and topographic effects that remained near cliffs after the
(Figure 5). Three spectral ratios with very high ratio values resampling of 30 m pixels to 15 m pixels, and should not
(ideally, a ratio code of 9) and spectral ratios with very low be used. The actual values of each ratio image were divided
values (ideally, with a ratio code of 0) were selected. The into 7 intervals or slices using density slicing, and the
best ASTER spectral ratios for mapping sericite were 7/6 output pixels exported into a GIS environment as a shape
(with ratio code 9), 6/5 (with ratio code 0), and 8/6 (with file layer, yielding 3 ratio value layers for each mineral.
633
TAYEBI et al. / Turkish J Earth Sci
Table 2. ASTER VNIR+SWIR ratio codes for muscovite (sericite) and kaolinite (USGS spectral library).
Ratio image/Mineral 2/1 3/1 4/1 5/1 6/1 7/1 8/1 9/1 3/2 4/2 5/2 6/2 7/2 8/2 9/2 4/3 5/3 6/3
muscovi1 8 7 7 5 2 5 4 4 7 7 4 2 4 3 3 7 3 2
muscovi2 6 6 5 4 2 4 4 4 6 5 3 2 4 3 3 5 3 2
muscovi3 3 4 7 7 6 6 6 6 5 7 7 6 6 7 6 8 7 6
muscovi4 7 7 6 4 2 5 4 4 6 6 3 2 4 3 3 6 3 2
muscovi5 4 5 5 5 2 4 4 4 6 6 5 2 4 4 3 6 4 2
muscovi6 6 6 7 7 5 6 6 5 7 7 6 5 6 6 5 7 6 4
muscovi7 3 3 5 3 2 3 3 3 2 5 3 2 3 3 3 6 3 2
muscovi8 4 5 6 6 3 5 5 4 6 6 6 3 5 5 4 6 6 3
muscovi9. 6 6 5 3 2 4 3 3 6 5 3 2 3 3 3 5 3 2
Table 2. (Continued). ASTER VNIR+SWIR Ratio codes for muscovite (sericite) and kaolinite (USGS spectral library).
Ratio image/Mineral 7/3 8/3 9/3 5/4 6/4 7/4 8/4 9/4 6/5 7/5 8/5 9/5 7/6 8/6 9/6 8/7 9/7 9/8
muscovi1 4 3 3 2 1 3 3 2 0 4 2 2 9 9 8 2 1 2
muscovi2 4 3 3 3 2 4 4 4 0 4 3 3 9 9 7 3 3 2
muscovi3. 7 7 6 5 3 3 4 4 0 1 2 2 7 7 5 6 4 2
muscovi4 4 3 3 2 1 3 3 2 0 8 3 3 9 9 8 2 2 2
muscovi5 4 4 3 3 2 4 4 3 0 2 3 3 9 9 7 4 3 2
muscovi6 6 5 4 4 2 3 4 3 0 1 2 2 8 8 6 4 3 2
muscovi7 4 3 3 2 1 3 3 2 0 3 3 3 9 9 8 3 2 2
muscovi8 5 5 4 4 2 4 4 4 0 2 3 3 8 8 6 4 3 2
muscovi9. 3 3 3 3 2 3 3 3 0 3 3 3 9 9 7 3 2 2
Ratio image/Mineral 2/1 3/1 4/1 5/1 6/1 7/1 8/1 9/1 3/2 4/2 5/2 6/2 7/2 8/2 9/2 4/3 5/3 6/3
muscovia 6 6 6 4 2 4 4 4 6 6 4 2 4 3 3 6 3 2
muscovib 5 6 7 6 3 5 5 4 7 7 6 3 6 5 4 7 6 3
muscovic 6 7 7 6 4 6 5 5 7 7 6 3 6 5 4 7 6 3
muscovid 6 6 6 6 5 6 6 6 7 6 6 5 6 6 5 6 6 4
kaolini1 4 4 4 1 1 1 1 1 5 4 1 1 1 2 2 4 1 1
kaolini2 6 6 5 1 1 2 2 2 6 4 1 1 2 2 2 4 1 2
kaolini3 6 6 5 1 1 2 2 2 5 4 1 1 2 2 2 4 2 2
kaolini4 6 6 4 1 1 2 2 1 6 3 1 1 2 2 2 2 1 1
kaolini5 6 6 5 1 1 2 2 2 7 5 1 1 2 2 2 4 1 1
kaolini6 6 6 5 1 1 2 2 2 6 4 1 1 2 2 2 3 1 1
kaolini7 6 6 5 1 1 2 2 2 6 4 1 1 2 2 2 4 1 1
kaolini8 6 6 4 1 1 2 2 2 6 3 1 1 2 2 2 2 1 1
Ratio image/Mineral 7/3 8/3 9/3 5/4 6/4 7/4 8/4 9/4 6/5 7/5 8/5 9/5 7/6 8/6 9/6 8/7 9/7 9/8
muscovia 3 3 3 3 2 3 4 3 0 3 3 3 9 9 7 3 3 2
muscovib 5 4 4 2 1 3 3 2 0 2 2 2 9 9 7 3 1 2
muscovic 5 4 4 3 2 3 3 2 0 1 2 2 9 9 6 3 2 2
muscovid 6 5 5 5 3 5 5 4 0 2 3 4 8 8 7 5 4 2
kaolini1. 2 1 2 1 1 1 1 1 2 9 5 4 9 7 4 2 1 2
kaolini2 2 2 2 1 1 2 2 2 2 8 6 5 9 8 5 2 2 2
kaolini3 2 2 2 1 1 2 3 2 1 8 5 4 9 8 5 2 2 2
kaolini4 2 2 2 1 1 2 2 2 0 8 4 3 9 8 5 2 1 1
kaolini5 2 2 2 1 1 2 2 2 1 9 5 4 9 8 5 2 1 2
kaolini6 2 2 2 1 1 2 2 2 1 9 5 4 9 8 5 2 1 2
kaolini7 2 2 2 1 1 2 2 2 1 9 5 4 9 8 5 2 1 2
kaolini8 2 2 2 1 1 2 2 2 1 8 5 4 9 8 5 2 1 2
634
TAYEBI et al. / Turkish J Earth Sci
1: 2 /1 5 : 6 /1 9 : 3 /2 1 3: 7 /2 1 7 : 5 /3 2 1 : 9 /3 2 5: 8 /4 29 : 8 /5 33 : 9 /6
2: 3 /1 6 : 7 /1 1 0: 4 /2 1 4 : 8 /2 1 8 : 6 /3 2 2: 5 /4 2 6 : 9 /4 3 0 : 9/5 3 4 : 8/7
3: 4 /1 7 : 8 /1 11 : 5 /2 1 5 : 9/2 1 9 : 7 /3 23 : 6 /4 2 7 : 6/5 3 1 : 7 /6 3 5 : 9 /7
4: 5 /1 8 : 9 /1 1 2: 6 /2 1 6 : 4 /3 2 0 : 8 /3 2 4: 7 /4 2 8 : 7 /5 3 2 : 8/6 3 6 : 9/8
10 10
9 259 expo.000.sco.rt 9 b muscovi 7
a
8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
Sample 259 (6 /5 7 /6 8 /6 ) Muscovite 7 (U SGS Library) (6 /5 7 /6 8 /6 )
10 10
9 c 218 expo.000.sco.rts 9 d kaolini
8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
0 0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35
Sample 218 (7 /5 7 /6 6 /5 ) Kaolinite (U SGS Library ) (7 /5 7 /6 6 /5 )
Figure 5. Spectral ratio code signature pattern of sample 259 (a), muscovite (sericite) (b), sample 218 (c), and kaolinite (d). The
best ratio images of sericite and kaolinite were selected (red circles) by comparing spectral ratio code signature pattern of field
spectra and USGS spectral library.
AND and XOR logic were used according to Eqs. (1), (2), where T sericite is the training area layer for sericite, T kaolinite
and (3) to create a training area layer. The intersection is the training area layer for kaolinite, Rv i/j are the 3 ratio
(AND logic) method was performed on the ratio layers to value layers for each mineral, and i,j are the 2 spectral band
select and generate new layer (T sericite) for sericite and (T numbers used in the Ri/j spectral ratio. Then another GIS
kaolinite
) for kaolinite. This is an integration of 3 ratio layers overlay (XOR logic) method was used to create a training
that preserves pixels that fall within the spatial extent area layer (TA) according to Eq. (3). This is an integration
common to 3 ratio layers. Therefore, pixels that pass this of T sericite and T kaolinite that preserves pixels that fall within
AND-logic method are pixels covered by the sericite that the spatial extent not common to these 2 layers.
have ratio codes 9, 0, and 9 and covered by kaolinite that
have ratio codes 9, 9, and 1. When a color ratio image is TA= (T sericite + T kaolinite) – (T sericite * T kaolinite), (3)
made with display colors of red, green, and blue assigned
to the first, second, and third ratios, respectively, it yields where TA is training area. The training area was then used
a color of yellow for kaolinite (red+green) and fuchsia/ for training the SOM with all of the inputs.
purple for sericite (red+blue). The AND logic equations
for the sericite and kaolinite layers are given by Eqs. (1) 3.2. Mapping alteration units
and (2): 3.2.1. Self-organizing map (SOM) model
A SOM (Kohonen, 1995) is a 2-layer network that maps
T sericite = Rv7/6 ∩Rv6/5 ∩Rv8/6 or T sericite = Rv7/6 * Rv6/5* Rv8/6 the n-dimensional feature vector (Mather et al., 1998) or
(1)
reflectance bands (input layers, which is 5, 15, and 36 ratio
T kaolinite = Rv7/5 ∩Rv7/6 ∩Rv6/5 or T kaolinite = Rv7/5 * Rv7/6* Rv6/5, image data sets) into a typically 2-dimensional array of
(2) nodes referred as the competitive layer (output layer) or
635
TAYEBI et al. / Turkish J Earth Sci
a b
)"")") ")")")")
")") )"
")
") ")")
)"") )"
") )"")
") ")")")
") ")
")
)"")")
")
)"
)" S a mple 259
") ")
")")
S a mple 218
)"")")")")")")")")")")
)"")
R a tio c o lo r c o mpo s ite R a tio c o lo r c o mpo s ite
(R GB : 7/6 8/6 6/5) (R GB : 7/5 7/6 6/5)
)"
S a mple s ite s
P lug s a nd dike s
c Alte re d ro c ks d e
Alte ra tio n units Alte re d ro c ks
La v a b re c c ia s
Figure 6. Color ratio images displayed as RGB with sampling sites as red dots in Figure 6a only. (a) Muscovite (sericite) is
displayed as yellow (RGB: 7/6 8/6 6/5), (b) Kaolinite is displayed as yellow (RGB: 7/5 7/6 6/5). Color field photos of the caldera
showing alteration units, pre- and postcaldera stage intrusive rocks (c, d, and e).
636
TAYEBI et al. / Turkish J Earth Sci
637
TAYEBI et al. / Turkish J Earth Sci
layer more precisely. Pixels within the training areas were (Figure 8d). The result of SOM applied to the 36 ratio
again presented to the SOM and the output neuron with image showed that the fraction image of sericite at the
the minimum Euclidean distance between a training pixel caldera mainly consists of 62%–75% (yellow pixels) and
and its weight vector was selected as the best matching 87%–100% (red pixels, Figure 9b). Meanwhile, the fraction
unit. The feature maps of alteration units of the volcano image of kaolinite at the caldera mainly consists of 37%–
derived from 5, 15, and 36 ratio image data sets are shown 50% (cyan pixels, Figure 9d).
in Figures 7–9. Maps include the fraction and spatial To evaluate the results of SOM classifications maps,
distribution of sericite (phyllic zone) and kaolinite (argillic the accuracy of the alteration units was assessed using
zone) in the Masahim volcano. The zoomed feature maps ground reference information by determining the overall,
of the caldera are likewise shown. user’s, and producer’s accuracies and kappa coefficients
The result of the SOM applied to the 5 ratio images (Tables 3–5). The overall accuracy of the SOM applied
showed that the fractions of sericite at the caldera mainly to the 36, 15, and 5 ratio image data sets was 87%, 86%,
consisted of 62%–75% (yellow pixels) and 87%–100% (red and 85%, respectively. The confusion matrix of the SOM
pixels) (Figure 7b), and the fraction of kaolinite mainly applied to the 36 ratio image data set (Table 3) shows good
consisted of 75%–87% (green pixels) and 87%–100% (red accuracy for sericite and kaolinite. Errors of commission
pixels) (Figure 7d). The result of SOM applied to the 15 and omission (errors of C and O) for sericite in the SOM
ratio image showed that the fraction image of sericite at applied to the 36 ratio image data set (Table 3) were 0.07
the caldera mainly consisted of 75%–87% (green pixels) and 0.15, and the user accuracy and producer accuracy
and 87%–100% (red pixels) (Figure 8b). However, the were 93% and 85%, respectively. Sericite has 15% error of
fraction image of kaolinite at the caldera mainly consisted O and 7% error of C with kaolinite, which again means that
of 50%–62% (magenta pixels) and 62%–75% (red pixels) 15% of sericite pixels are omitted from the sericite class
b
0.37 to 0.50
0.50 to 0.62
0.62 to 0.75
0.75 to 0.87
a b
0.87 to 1.00
d
0.37 to 0.50
0.50 to 0.62
0.62 to 0.75
0.75 to 0.87
c d
0.87 to 1.00
Figure 7. Feature and zoomed feature maps of Masahim volcano derived from 5 ratio images. Maps include the fractions
of sericite (a and b) and kaolinite (c and d). The yellow solid line indicates the geologic unit boundaries (Figure 1).
638
TAYEBI et al. / Turkish J Earth Sci
b
0.37 to 0.50
0.50 to 0.62
0.62 to 0.75
0.75 to 0.87
a 0.87 to 1.00 b
d
0.37 to 0.50
0.50 to 0.62
0.62 to 0.75
0.75 to 0.87
c d
0.87 to 1.00
Figure 8. Feature and zoomed feature maps of Masahim volcano derived from 15 ratio images. Maps include the fraction
of sericite (a and b) and kaolinite (c and d). The yellow solid line indicates the geologic unit boundaries (Figure 1).
and 7% of the pixels that should be included as sericite and kaolinite. For example, errors of C and O for sericite
are assigned to kaolinite. User accuracy and producer were 0.05 and 0.20, and the user and producer accuracies
accuracy for kaolinite were 79% and 90% respectively, of that were 95% and 80%, respectively. Sericite has 20%
with 15% error of O and 7% error of C with sericite. The error of O and 5% error of C with kaolinite, which means
results also indicated a kappa coefficient of 0.73 for the that 20% of sericite pixels were omitted from the sericite
SOM (Table 3). class and 5% of the pixels that should have been included
Errors of C and O for sericite in the SOM applied to as sericite were assigned to kaolinite. Errors of C and O
the 15 ratio were 0.12 and 0.10, respectively (Table 4). for kaolinite were 0.26 and 0.07, and its user and producer
Meanwhile, user accuracy and producer accuracy were accuracies were 74% and 93%, respectively. Kaolinite
88% and 90%, respectively. Sericite has 10% error of O and has 7% error of O and 26% error of C with sericite. The
12% error of C with kaolinite, which means that 10% of results also indicated that the operating SOM has a kappa
sericite pixels are omitted from the sericite class and 12% coefficient of 0.69 (Table 5).
of the pixels that should be included as sericite are assigned The most accurate classification is achieved by applying
to kaolinite. Errors of C and O for kaolinite were 0.17 and the SOM to the 15 and 36 ratio data sets (Figures 8 and 9).
0.20, and the user accuracy and producer accuracy were Output results of the SOM applied to the 15 and 36 ratio
80% and 83%, respectively. Moreover, kaolinite showed image data sets generally enhanced similar areas spatially.
20% error of O and 17% error of C with sericite. The However, the fractions of pixels in SOM applied to 15 ratio
results also indicated that the operating SOM has a kappa image data sets are higher than the output result of 36 ratio
coefficient of 0.70 (Table 4). image data sets (Figures 8 and 9). The accuracy of fractions
The confusion matrix of the SOM applied to the 5 ratio estimated by the SOM applied to 15 and 36 ratio image
image data set (Table 5) showed good accuracies for sericite data sets were also assessed using R2 coefficient. A total of
639
TAYEBI et al. / Turkish J Earth Sci
b
0.37 to 0.50
0.50 to 0.62
0.62 to 0.75
0.75 to 0.87
a b
0.87 to 1.00
d
0.37 to 0.50
0.50 to 0.62
0.62 to 0.75
c 0.75 to 0.87 d
0.87 to 1.00
Figure 9. Feature and zoomed feature maps of Masahim volcano derived from 36 ratio images. Maps include the fraction of
sericite (a and b) and kaolinite (c and d). The yellow solid line indicates the geologic unit boundaries (Figure 1).
Table 3. Confusion matrix of SOM for the classification image of 36 ratio images input.
Ground truth
40 sites of 60 m × 60m were sampled and used for this 0.64 and kaolinite 0.68 of R2 coefficients in the SOM
assessment. The scatter plots of the accuracy assessment applied to the 36 ratio images.
are shown in Figure 10. The R2 coefficient for sericite was Triplets of spectral ratio images displayed in colors of
nearly 0.35, and for kaolinite was nearly 0.52 in the SOM red, green, and blue (as a ratio color composite) were used
applied to the 15 ratio images. Moreover, sericite showed to find training areas. When these 3 colors are combined,
640
TAYEBI et al. / Turkish J Earth Sci
Table 4. Confusion matrix of SOM for the classification image of 15 ratio images input.
Ground truth
Table. 5. Confusion matrix of SOM for the classification image of 5 ratio images input.
Ground truth
R 2 = 0.3477
0.8 0.8
0.6 0.6
0.4 0.4
2
0.2
R2 =0.3477 0.2
R =0.5224
0 0
0 0.2 0.4 0.6 0.8 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Actual Fraction (sericite) Actual Fraction (kaolinite)
1.2 y = 0.8391x - 0.0159 1.2
y = 0.9462x + 0.0671
1
c R 2 = 0.6413 d
1 R 2 = 0.6841
Estimated Fraction
0.8 0.8
0.6 0.6
0.4 0.4
0.2 2 2
R =0.6413 0.2 R =0.6841
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Actual Fraction (sericite) Actual Fraction (kaolinite)
Figure 10. Scatter plots of accuracy assessment results for SOM applied to the 15 (a and b) and 36 (c and d) ratio image data sets.
641
TAYEBI et al. / Turkish J Earth Sci
they produce different colors in the visible spectrum. The applied to 36, 15, and 5 ratio image data sets showed good
resulting color ratio image generally enhances specific identification of phyllic and argillic units. We achieved
minerals of interest in different colors. This mineral may overall accuracies of 87% (kappa 0.73) for SOM applied to
appear in shades of blue, green, and red or various color the 36, 86% (kappa 0.70) for 15, and 85% (kappa 0.69) for 5
combinations of the 3, depending on which ratios are ratio image data sets. The results also showed that the SOM
employed to yield the color ratio image. More pointedly, applied to 15 and 36 data sets generally enhanced similar
we can use a ratio code method to determine which areas spatially. However, the output results showed some
spectral ratios are very high and very low in values for a differences in the fractions of minerals within each pixel.
given mineral, compared to other minerals and rocks for The R2 coefficient for sericite was 0.35 and for kaolinite was
which we have an archived spectral library of mineral and 0.52 in the SOM applied to the 15 ratio images. Sericite
rock reflectances, and we can create a spectral ratio image showed 0.64 and kaolinite 0.68 of R2 coefficients in the
in which we assign the red display color to the highest SOM applied to the 36 ratio images.
spectral ratio for a mineral that we wish to highlight,
It is concluded that coded spectral ratio imaging is
and assign the blue and green display colors to 2 spectral
efficient for selecting training areas for a given mineral or
ratios that are very low for that mineral. The resulting
rock, especially when we do not have access to the area and
target mineral is displayed as red, and almost all and
we only have access to the archived spectral library spectra.
sometimes all other surface areas are displayed as other
Meanwhile, the SOM neural network model applied to
colors, except red. Red in the scene, for that ratio-code-
selected color ratio composite, is the mineral we seek to 36 ratio image data sets can also be effective for mapping
find (the target mineral). Then it is possible to input dark- alteration minerals, yielding estimates of pixel fractions
object–corrected spectral ratio imaging data sets to the covered by each target mineral of interest. However, more
SOM model that are rid of the topographic effects to map study areas need to be tested in the future, and the search
and estimate abundance fractions of minerals of interest. for a better method than the ratio code method for finding
While this method can identify minerals of interest, more the most characteristic spectral parameters of target
study areas need to be tested in the future. minerals should be sought.
6. Conclusion Acknowledgments
The spatial distribution of altered units at the Masahim The authors thankfully acknowledge the constructive
volcano was determined using ratio image/ratio codes suggestions received from the TJES’s editor and 2
and a SOM neural network model. The result of the SOM anonymous reviewers.
References
Ahmadipour H (1994). Dynamism of Volcanic Activities in Clark RN, Swayze GA, Gallagher A, King TVV, Calvin WM (1993).
Mosahim volcano (Shahr-e-Babak). MSc, Department of The U.S. geological survey digital spectral library: Version 1:
Geology, Shahid Bahonar Univerity, Kerman, Iran. 0.2 to 3.0 µm. USGS Open File Report 93–592, U.S. Geological
Survey, p. 1340.
Bagan H, Wang QX, Watanabe M, Yang YH, Jianwen M (2005).
Land cover classification from MODIS EVI times-series Cohen J (1960). A coefficient of agreement for nominal scales. Educ
data using SOM neural network. Int J Remote Sens 26: Psychol Meas 20: 37– 44.
4999–5012. Crosta AP, Souza Filho CR, Azevedo F, Brodie C (2003). Targeting
Bedini E (2009). Mapping lithology of the Sarfartoq carbonatite key alteration minerals in epithermal deposits in Patagonia,
complex, southern West Greenland, using HyMap imaging Argentina, using ASTER imagery and principal component
spectrometer data. Remote Sens Environ 113: 1208–1219. analysis. Int J Remote Sens 24: 4233–4240.
Bedini E (2012). Mapping alteration minerals at Malmbjerg Crowley JK, Zimbelman DR (1997). Mapping hydrothermally altered
molybdenum deposit, central East Greenland, by Kohonen rocks on Mount Rainier, Washington, with Airborne Visible/
self-organizing maps and matched filter analysis of HyMap Infrared Imaging Spectrometer (AVIRIS) data. Geology 25:
data. Int J Remote Sens 33: 939–961. 559–562.
Bian ZQ, Zhang XG (2000). Pattern Recognition, 2nd ed, Beijing, Crowley JK, Hubbard BE, Mars JC (2003). Analysis of potential
China: Tsinghua University Press. debris flow source areas on Mount Shasta, California, by
using airborne and satellite remote sensing data. Remote Sens
Chen X, Warner TA, Campagna DJ (2007). Integrating visible,
Environ 87: 345–358.
near-infrared and short-wave infrared hyperspectral and
multispectral thermal imagery for geological mapping at
Cuprite, Nevada. Remote Sens Environ 110: 344–356.
642
TAYEBI et al. / Turkish J Earth Sci
Cudahy T, Hewson R (2002). ASTER geological case histories: Hunt GR, Salisbury JW (1970) Visible and near-infrared spectra of
porphyry-skarnepithermal, iron oxide Cu-Au and Broken hill minerals and rocks, I. Silicate minerals. Mod Geol 1: 283–300.
Pb-Zn-Ag. Annual General Meeting of the Geological Remote Ji CY (2000). Land-use classification of remotely sensed data
Sensing Group ‘ASTER Unveiled’, Burlington House, Piccadilly, using Kohonen self organizing feature map neural network.
London, UK. Photogram Eng Remote Sensing 66: 1451–1460.
Dimitrijevic M (1973) Geology of Kerman region. (Iran Geological Jianwen M, Bagan H (2005). Land-use classification using ASTER
Survey Report No. Yu/52.) Institute for Geological and Mining data and self organized neutral networks. Int J Appl Earth
Exploration and Institution of Nuclear and Other Mineral Raw Obs 7: 183–188.
Materials, Beograd Yugoslavia, p334.
Joyce K, Samsonov S, Manville V, Jorgens R, Graettinger A, Cronin
DuBose P, Klimasauskas C (1989). Introduction to Neural Networks S (2009). Remote sensing data types and techniques for lahar
with Examples and Applications.Pittsburgh, PA, USA: path detection: a case study at Mt Ruapehu, New Zealand.
NeuralWare Inc. Remote Sens Environ 113: 1778–1786.
Ehsani AH, Quiel F (2008). Geomorphometric feature analysis using Kangas JA, Kohonen TK, Laaksonem JT (1990). Variants of Self-
morphometric parameterization and artificial neural networks. Organizing Maps. IEEE T Neural Networ 1: 93–99.
Geomorphology 99: 1–12.
Kerle N, Froger JL, Oppenheimer C, De Vries BV (2003). Remote
Emaruchi B, Jin Y, Sauchyn D, Fuller G (1997). Estimation of runoff sensing of the 1998 mudflow at Casita volcano, Nicaragua. Int
index using Landsat (TM) data and an artificial neural network J Remote Sens 24: 4791−4816.
system. Can Water Resour J 22: 155–166.
Kohonen T (1995). Self-Organizing Maps. Berlin, Germany:
Fujisada H (1995). Design and performance of ASTER instrument. Springer.
In: Fujisads H, Sweeting MN, editors. Proceedings of SPIE
Kruse FA (1988). Use of Airborne Imaging Spectrometer data to
(International Society for Optical Engineering), 25–28
map minerals associated with hydrothermally altered rocks in
September 1995, Paris, France, 2583, pp.16–25.
the northern Grapevine Mountains, Nevada and California.
Geaga JV, Le TC, Sailer CT (1992). Neural net spectral pattern Remote Sens Environ 24: 31–51.
recognition. In: Proc. SPIE, 1766. SPIE, Bellingham, WA: pp.
Lee S, Lathrop RG (2006). Subpixel analysis of Landsat ETM+ using
609–618.
Self-Organizing Map (SOM) neural networks for urban land
Gomez C, Delacourt C, Allemand P, Ledru P, Wackerle R (2005). cover characterization. IEEE T Geosci Remote 44: 1642–1654.
Using ASTER remote sensing data set for geological mapping,
Li Z, Eastman JR (2006). The nature and classification of unlabelled
in Namibia. Phys Chem Earth 30: 97–108.
neurons in the use of Kohonen’s self-organizing map for
Grebby S, Cunningham D, Naden J, Tansey K (2010). Lithological supervised classification. Trans GIS 10: 599–613.
mapping of the Troodos ophiolite, Cyprus, using airborne
Mars JC, Rowan LC (2006). Regional mapping of phyllic- and
LiDAR topographic data. Remote Sens Environ 114: 713–724.
argillic-altered rocks in the Zagros magmatic arc, Iran, using
Gupta RP (2003). Remote Sensing Geology. Berlin, Germany: Advanced Spaceborne Thermal Emission and Reflection
Springer-Verlag. Radiometer (ASTER) data and logical operator algorithms.
Hassanzadeh J (1993). Metallogenic and tectonomagmatic events Geosphere 2: 161–186.
in the SE sector of the Cenozoic active continental margin Mars JC, Rowan LC (2010). Spectral assessment of new ASTER
of central Iran (Shahr e Babak area, Keman Province). PhD, SWIR surface reflectance data products for spectroscopic
University of California, Los Angeles, Los Angeles. mapping of rocks and minerals. Remote Sens Environ 114:
Hellman MJ, Ramsey MS (2004). Analysis of hot springs and 2011–2025.
associated deposits in Yellowstone National Park using ASTER Mather PM, Tso B, Koch M (1998). An evaluation of Landsat
and AVIRIS remote sensing. J Volcanol Geotherm Res 135: TM spectral data and SAR-derived textural information for
195–219. lithological discrimination in the Red Sea Hills, Sudan. Int. J.
Hosseinjani M, Tangestani MH (2011). Mapping alteration minerals Remote Sens 19: 587–604.
using sub-pixel unmixing of ASTER data in the Sarduiyeh area, Perry JL, Vincent RK (2009). ASTER Brightness and Ratio Codes
SE Kerman, Iran. Int J Digital Earth 4: 487–504. for Minerals: Application to Lithologic Mapping In West-
Hu X, Weng Q (2009). Estimating impervious surfaces from medium Central Powder River Basin, Wyoming. Rev Econ Geol 16:
spatial resolution imagery using the self-organizing map and 143–168.
multi-layer perceptron neural networks. Remote Sens Environ Rajendran S, Thirunavukkarasu A, Balamurugan G, Shankar K
113: 2089–2102. (2011). Discrimination of iron ore deposits of granulite
Hubbard BE, Sheridan MF, Carrasco-Núñez G, Díaz-Castellón R, terrain of Southern Peninsular India using ASTER data. J
Rodríguez SR (2007). Comparative lahar hazard mapping at Asian Earth Sci 41: 99–106.
Volcan Citlaltépetl, Mexico using SRTM, ASTER and DTED-1 Research Systems Inc (2003). ENVI Tutorial, ENVI Software
digital topographic data. J Volcanol Geotherm Res 160: 99–124. Package Version 4.0.
643
TAYEBI et al. / Turkish J Earth Sci
Rowan L, Wetlaufer PH, Goetz AFH, Billingsley FC, Stewart Terunuma T, Nishida K, Amada T, Mizuyama T, Sato I, Urai M
JH (1974). Discrimination of rock types and detection of (2005). Detection of traces of pyroclastic flows and lahars
hydrothermally altered areas in south-central Nevada by use with satellite synthetic aperture radars. Int J Remote Sens 26:
of computer-enhanced ERTS images. US Geological Survey 1927−1942.
Professional, Paper, 883. Tso B, Mather PM (2001). Classification Methods for Remotely
Rowan LC, Mars JC (2003). Lithologic mapping in the Mountain Sensed Data. New York, NY, USA: Taylor and Francis.
Pass, California, area using Advanced Spaceborne Thermal Villmann T, Merenyi E, Hammer B (2003). Neural maps in remote
Emission and Reflectance Radiometer (ASTER) data. Remote sensing image analysis. Neural Networks 16: 389–403.
Sens Environ 84: 350–366.
Vincent RK (1972). An ERTS multispectral scanner experiment
Rowan LC, Hook SJ, Abrams MJ, Mars JC (2003). Mapping for mapping iron compounds. International Symposium on
hydrothermally altered rocks at Cuprite, Nevada using the Remote Sensing of Environment, 8th, Ann Arbor, University
Advanced Spaceborne Thermal Emissivity and Reflection of Michigan, Proceedings, 2, pp. 1239–1247.
Radiometer ASTER. A new satellite-imaging system. Econ
Geol 98: 1019–1027. Vincent RK (1997). Fundamentals of Geological and Environmental
Remote Sensing, , Upper Saddle River, NJ, USA, Prentice Hall.
Schowengerdt RA (2007). Remote Sensing: Models and Methods for
Image Processing. New York, NY, USA, Academic Press. Yamaguchi Y, Naito C (2003). Spectral indices for lithologic
discrimination and mapping by using the ASTER SWIR bands.
Souza Filho CR, Tápia CH, Crósta AP, Xavier RP (2003). Infrared Int J Remote Sens 24: 4311–4323.
spectroscopy and ASTER imagery analysis of hydrothermal
alteration zones at the Quellaveco Porphyry-Cooper Deposit, Zhang X, Pamer M, Duke N (2007). Lithologic and mineral
Southern Peru. Proceedings of the American Society of information extraction for gold exploration using ASTER
Photogrammetry and Remote Sensing (ASPRS), Annual data in the south Chocolate Mountains (California). Isprs J
Conference, Technology: Converging at the Top of the World, Photogramm 62: 271–282.
pp.1−12. Zhuang X, Engel BA, Baumgardner MF, Swain PH (1991). Improving
Tangestani MH, Mazhari M, Agar B, Moore F (2008). Evaluating classification of crop residues using digital land ownership data
Advanced Spaceborne Thermal Emission and Reflection and LANDSAT TM Imagery. Photogram Eng Remote Sensing
Radiometer (ASTER) data for alteration zone enhancement in 57: 1487–1492.
a semiarid area, northern Shahr-e-Babak, SE Iran. Int J Remote
Sens 29: 2833–2850.
644