Principal components analysis for borate mapping

This article was downloaded by: [Dokuz Eylul University ]On: 06 November 2014, At: 21:37Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

International Journal of RemoteSensingPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/tres20

Principal components analysis forborate mappingH. Kargi aa Department of Geological Engineering , Pamukkale University ,Kinikli, 20020 Denizli, TurkeyPublished online: 09 Jul 2007.

To cite this article: H. Kargi (2007) Principal components analysis for borate mapping, InternationalJournal of Remote Sensing, 28:8, 1805-1817, DOI: 10.1080/01431160600905003

To link to this article: http://dx.doi.org/10.1080/01431160600905003

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/loi/tres20

http://www.tandfonline.com/action/showCitFormats?doi=10.1080/01431160600905003

http://dx.doi.org/10.1080/01431160600905003

http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/page/terms-and-conditions

Principal components analysis for borate mapping

H. KARGI*

Department of Geological Engineering, Pamukkale University, Kinikli, 20020 Denizli,

Turkey

(Received 7 March 2006; in final form 12 July 2006 )

Principal components analysis (PCA) of remotely sensed satellite image data is a

widely used method in mineral exploration. Generally, the method is used for

iron oxide and hydroxyl mapping. In this study, however, the PCA method is

adopted for borate exploration. This paper demonstrates how PCA of Landsat

TM data can be used to map borate minerals. The method has been applied to

the sub-scene of Bigadic and tested on the borate field in Kirka, Turkey.

Anomalous pixels for borate minerals in PC6 images have coincided with known

borate deposits. Whether borate minerals are mapped into a PC image depends

on the appearance of opposite signs in eigenvector loadings for TM4 and TM7 in

one or more PCs. Borate coverage in an image is important to emphasize the

appearance of opposite signs in eigenvector loadings for TM4 and TM7 in more

than one PC.

1. Introduction

Remote sensing technology can play a significant role in mineral exploration. Using

remote sensing technology, researchers/scientists can cover large amounts of landsurface to choose areas that require further detailed studies. Satellite remote sensing

images have been used widely and successfully for mineral exploration in arid and

semi-arid environments since the launch of second generation Landsats in 1982 that

carry a multispectral scanner called the Thematic Mapper (see Sabins 1999). This

application relies mostly on the capability of the sensor to register spectral

signatures of minerals such as iron oxide and hydroxyl minerals associated with

mineralization (Kaufman 1988, Knepper and Simpson 1992, Bennett et al. 1993,

Goosens and Kroonenberg 1994, Abdelsalam et al. 2000) and other geologicalfeatures, such as lithology and faults, related to ore deposits (Rowan and Bowers

1995, Kusky and Ramadan 2002).

A few works on the exploration of borate deposits by satellite remote sensing have

also appeared in the literature (Sabins and Miller 1994, Sabins 1999, Posadas et al.

2005). High reflectance of borate minerals at NIR and high absorption at SWIR

(JPL 2006, USGS 2006) suggest that they can be mapped using the 4th and 7th TM

bands of Landsat. In fact, 4/7-TM band ratioing has been used by Sabins (1999) to

map boron concentration at Salar de Uyuni, southwest Bolivia. He has shown thatthe highest ratio values coincided with the contours of maximum boron

concentration in an embayment at the south margin of the Salar. He has also

proposed that additional anomalous pixels elsewhere around the margin of the Salar

indicated potential borate reserves that were not actually detected by the sample

*Email: [email protected]

International Journal of Remote Sensing

Vol. 28, No. 8, 20 April 2007, 1805–1817

International Journal of Remote SensingISSN 0143-1161 print/ISSN 1366-5901 online # 2007 Taylor & Francis

http://www.tandf.co.uk/journalsDOI: 10.1080/01431160600905003

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

program. A binarized TM 4/7 ratio image was used by Posadas et al. (2005) to

describe multifractal characterization of ulexite distribution in the Salar.

Turkey possesses the majority of the world’s borate deposits. So far, no satellite

remote sensing related works have been carried out on these deposits. This study is

an attempt to develop an alternative technique that can better recognize boron

anomalies on Landsat TM images than TM 4/7 ratioing, which in this study has

failed to identify some Turkish borate deposits. First, TM 4/7-band ratioing has

been tested on a Landsat TM sub-scene that covered the Bigadic borate ore field in

Turkey. Second, principal components analysis (PCA) on six bands, excluding the

TM thermal infrared band, has been applied to the same sub-scene to examine its

ability to correctly locate borate deposits. The aim of applying PCA here is neither

to identify evaporate mineral end members, as done by Bryant (1996), nor to map

ore-associated clay minerals, as done in most studies mentioned above, but to

directly map borate minerals using higher order PC components. In a typical PCA

on six TM bands, as characterized by eigenvectors, the PC1 image maps albedo, the

PC2 or PC3 image maps vegetation or visible-infrared difference, the PC4 or PC5

image maps iron oxides or clay minerals, and the PC6 image maps either iron oxides

or clay minerals (Loughlin 1991). The idea that borate minerals may be mapped into

one of the PC images, most likely into a PC6 image, is based on the extensive

occurrence of borate pits in the Bigadic region and the presence of a significant

proportion of anomalous pixels in terms of boron concentration in the sub-scene

studied.

To test the results, both band ratioing and PCA have also been applied to the

Kirka region, which hosts another important borate deposit in Turkey. In addition,

PCA was applied to several non-borate deposit hosting areas for comparison. The

possibility of applying PCA to Landsat TM data for use in borate exploration and

selection of image size are discussed in the text.

The Landsat TM image reproduced to study the Bigadic area (from p180r33) is a

8576489 pixels sub-scene (figure 1). The image was acquired on 5 June 1987 and

orthorectified by the EARTHSAT Company (NIK Insaat Ticaret Limited

Campany, personal communication). The accuracy of the rectification was

rechecked by ground control points. A Landsat TM image of Kirka (hosting other

borate deposits) was reproduced from p179r33. The acquisition date of both

p179r33 and p179r34 (non-borate hosting area) was 1 August 1987. No correction

for atmospheric scattering has been done on the images presented in this paper, as it

is unnecessary for single date images (Song et al. 2001).

2. Geology and borate deposits of the Bigadic region

The geology of the Bigadic region is described in detail by Helvaci and Alaca (1991)

and is summarized below.

Palaeozoic metamorphic rocks (composed mainly of schist and marble) and

Mesozoic limestone, ophiolite, and associated sedimentary rocks are the basement

rocks in the study area. Neogene volcano-sedimentary rocks, formed in a playa

basin and hosting borate deposits, unconformably lie on the basement. Helvaci and

Alaca (1991) divided this volcano-sedimentary rock into seven units (figure 2). These

are listed in chronological order: (1) lower volcanic unit, consisting of basic to acidic

volcanic rocks and their pyroclastics; (2) lower limestone unit that includes layered

limestone, dolomitic limestone, and interbedded marl and shale; (3) lower tuff unit,

deposited in both continental and lake environments; (4) lower borate-bearing unit

1806 H. Kargi

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

Figure 1. Locations of study areas in southwestern Turkey.

Figure 2. Geological map of the Bigadic region (after Helvaci and Alaca 1991).

Principal components analysis for borate mapping 1807

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

composed of limestone, marl, shale, tuff, and boron ores. The thickness of theborate ore zone in this sequence varies from 0.2 m to 76 m. The zone consists of

borate minerals (mainly colemanite and ulexite) and accompanying clay, gypsum,

and calcite minerals; (5) upper tuff unit that separates lower and upper boron-

bearing units and includes a significant amount of zeolite minerals; (6) upper borate-

bearing unit having somewhat similar stratigraphy to the lower borate-bearing unit.

The thickness of the ore zone in this unit varies from 0.1 m to 80 m; and (7) the

basalt unit, which cuts all older rocks. Both basement rocks and Neogene volcano-

sedimentary rocks are unconformably overlaid by quaternary sedimentary rocksand alluviums. Ore deposits and associated rocks are both faulted and folded by

post-tectonic activities.

At the time of the satellite pass, the pits mined for borate were the Yenikoy and

Tulu pits, where ores were recovered from the lower borate-bearing unit, and the O.

gunevi, A. gunevi, Kireclik, Acep, Avsar, and Simav pits, where ores were recovered

from the upper borate-bearing unit. Among these, the Acep and Tulu pits were the

only open-pit mines (figure 2).

3. Band ratioing

Figure 3 shows reflectance spectra of colemanite and ulexite, which are both

common in the Bigadic borate deposits. These borate minerals have distinctiveabsorption features reflectance minima at wavelengths within the band pass of TM

band 7, and have higher reflectance values within TM band 4. Ratio images can

emphasize and quantify these spectral differences. A ratio image for the Bigadic

region has been prepared by dividing the value of TM band 4 by that of TM band 7

and rescaling values to 255 (figure 4).

Several thresholding techniques have developed to separate anomalies from the

background in grey level images (see Sahoo et al. 1988). No matter what technique is

applied, it just changes thresholds, and the highest ratios in the histogram mustbelong to anomalous pixels. Therefore, the threshold has been chosen simply as the

mean + 3 std (standard deviation), 4 std, 5 std, and 6 std, and the ratio image was

binarized to have the anomalies mapped for each threshold level in order to

Figure 3. Reflectance spectra of common minerals in the borate-bearing units of the Bigadicregion. Note that all minerals have similar high reflectance at NIR, but borate minerals havelower reflectance at the wavelengths within the band pass of TM band 7 (data from USGS2006).

1808 H. Kargi

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

compare to actual location of the borate ore pits (figure 5(a–d)). Almost none of the

anomalous pixels of the ratio image coincided with the Bigadic borate ore pits when

the threshold was set on mean + 6 std. Shifting the threshold from mean + 6 std to

mean + 5 std, 4 std, and 3 std did not emphasize the borate deposits as additional

anomalous pixels fell elsewhere rather than coinciding with borate pits. The problem

with the ratio image was due mainly to the presence of water (and high humidity)

sites in the image, which absorbed short-wave infrared lights and caused these pixels

to have high ratios. The undesirable effects of water on band ratioing were also

observed in the other sub-scenes studied in the following sections. Water sites

Figure 4. TM 4/7 ratio image for the Bigadic sub-scene. Squares show locations of boratepits.

(a)

(b) (d)

(c)

Figure 5. Anomaly maps for the TM 4/7 ratio image of Bigadic at thresholds mean + 6 std(a), 5 std (b), 4 std (c), and 3 std (d).


Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

probably should be masked when band ratioing applies. No masking has been

carried out in this study.

4. Principal components analysis

The principal component transformation is a multivariate statistical technique that

selects uncorrelated linear combinations (eigenvector loadings) of variables in such a

way that each successively extracted linear combination, or principal component,

has a smaller variance (Singh and Harrison 1985). This is a well-known method for

alteration mapping (Crosta and Moore 1989, Loughlin 1991, Ranjbar et al. 2004).

Image eigenvectors and eigenvalues obtained from the principal components

transformation, using the covariance matrix, on all six reflective bands of TM of the

Bigadic sub-scene are presented in table 1. The transformation was carried out on

unstretched data. In this transformation, the first principal component (PC1) is

composed of a positive weighting of all total bands. As indicated by the eigenvalues,

PC1 accounts for 83.30% of the total variance for the data. Overall scene brightness,

or albedo, is responsible for the strong correlation between multispectral image

channels (Loughlin 1991). PCA has effectively mapped albedo into PC1 of the

transformation. Eigenvector loadings for PC2 indicate that PC2 is dominated by

vegetation, which is highly reflective in TM4; the positive loading of TM4 in this PC

(0.972) also indicates that strongly vegetated pixels will be bright in this PC image.

The percentage of variance mapped into this ‘vegetation’ PC is only 10.22, which is

not a measure of vegetation abundance in the Bigadic area, where most pixels will

contain some vegetation. Eigenvector loadings for PC3 in table 1 indicate that PC3

describes the difference between the visible channels (TM1, 2, and 3) and the

infrared (IR) channels (TM4, 5, and 7). Having mapped albedo to PC1 and

vegetation and visible to IR differences to PC2 and PC3, respectively, the remaining

three PCs can be expected to contain information due to the varying spectral

response of iron oxides and hydroxyl-bearing minerals (Loughlin 1991). By looking

for moderate or large eigenvector loadings for TM5 and TM7 in PCs where these

loadings are also opposite in sign, we can predict that hydroxyl-bearing minerals will

be distinguished by dark pixels in PC4 of table 1. Iron oxides are mapped as bright

pixels in PC5 due to the fact that the contribution in this PC is positive from TM3

and negative from TM1. The percentage of variance mapped into PC6 is only 0.05.

Eigenvector loadings in this PC are both negative for TM1 and TM3 (20.172 and

20.396, respectively) and positive for TM5 and TM7 (0.004 and 0.033, respectively).

The disappearance of opposite signs in eigenvector loadings for TM1–TM3 and

TM5–TM7 pairs suggest that neither iron oxide nor hydroxyl minerals are mapped

into this PC. On the other hand, signs of eigenvector loadings for TM4 and TM7 are

Table 1. Principal components analysis on six TM bands, Bigadic sub-scene.

BandTM1 TM2

TM3 TM4TM5 TM7 Eigenvalues

(%)Eigenvector matrix

PC1 0.313 0.219 0.392 0.121 0.685 0.466 83.80PC2 0.025 0.044 20.052 0.972 20.014 20.225 10.22PC3 20.566 20.335 20.503 0.047 0.536 0.160 5.13PC4 0.248 0.039 20.057 20.192 0.492 20.810 0.57PC5 20.700 0.162 0.656 20.005 0.046 20.224 0.23PC6 20.172 0.900 20.396 20.050 0.004 0.033 0.05

1810 H. Kargi

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

opposite, 20.050 and 0.033, respectively, which is in accordance with the reflection

spectra of colemanite and ulexite. If PC6 maps borate minerals, borate-rich fields

(open pit areas and tailings) must appear as the darkest pixels in the PC6 image

(figure 6). Anomaly maps similar to those in the band ratioing have been prepared

by binarizing the PC6 image after inverting the colours and putting thresholds at

mean + 6 std, 5 std, 4 std, and 3 std. Figure 7 shows that anomalous pixels at mean

+ 6 std threshold level coincided with most of all known borate pits. Shifting the

threshold from mean + 6 std to mean + 5 std, 4 std, and 3 std better emphasized

the borate pits. This result shows that PC analysis may also be used for the

Figure 6. PC6 image for the Bigadic sub-scene. Squares show locations of borate pits. Thedarkest pixels correspond to borate-rich areas.

(a)

(b) (d)

(c)

Figure 7. Anomaly maps for the PC6 image of Bigadic at thresholds mean + 6 std (a), 5 std(b), 4 std (c), and 3 std (d).


Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

determination of high boron concentration areas, and may be a powerful tool for

borate exploration.

5. Discussion

Although PCA has effectively differentiated borate-rich areas and mapped them

into the PC6 image, there might be some debate on whether reflectance spectra of

borate minerals or some other land cover features were responsible for this result. It

is therefore necessary to clarify the following points before proposing an adopted

PCA technique for use in borate exploration:

1. Is boron anomaly or absence of vegetation mapped into the PC6 image?

2. Can a similar result be obtained from other borate mining areas?

3. Is the proportion of boron anomaly areas in an image important?

Some approaches to answering these questions are discussed in the following

sections, along with their results.

5.1 Is boron anomaly or absence of vegetation mapped into the PC6 image?

Almost no vegetation cover is present in the mining areas. It seems to be a real

problem whether, in fact, boron anomaly or the absence of vegetation are

responsible for the anomalous pixels in the PC6 image because vegetation also has a

high reflectance in TM4 and a lower reflectance in TM7. If the positive loading for

TM4 (0.972) and negative loading for TM7 (20.225) in PC2 represent vegetation,

similarly it may be expected that negative loading for TM4 (20.050) and positiveloading for TM7 (0.033) in PC6 resulted from lack of vegetation in the mining areas.

If this is the case, absence of vegetation rather than boron anomaly is mapped into

PC6 and the same result may be obtained from a borate-free field. Several PCA

must be carried out on randomly selected images in order to clarify this point. For

this purpose, the image of p179r34 was chosen. Vegetation cover (mainly forest

cover) and climate in the area are somewhat similar to those of the Bigadic region.

Widespread occurrence in the area of marl (clay + limestone), extensive travertine

and marble quarries are comparable to the clay formation and mining pits (in termsof vegetation cover) of the Bigadic region. No borate occurrences have been

recorded in the area. PCA has been carried out on several sub-scenes extracted from

the p179r34 image by excluding the thermal infrared band. Signs of eigenvector

loadings for TM4 and TM7, especially in PC6s but also other PCs, were examined.

None of the eigenvector loadings for TM4 and TM7 and most of the eigenvector

loadings for TM5 and TM7 in PC6s are opposite in sign, which show that PC6

images generally map hydroxyl minerals. This experiment supports the fact that the

opposite sign of eigenvector loading for TM4 and TM7 in PC6 of the Bigadic regionis neither accidental nor due to lack of vegetation in the mining areas but is actually

due to a high concentration of boron.

5.2 Can a similar result be obtained from other borate mining areas?

Another important boron deposit in Turkey is located in the Kirka region. There is

only one large open-pit mining field in the region. The deposit was formed bysedimentation of borate minerals interbedded with clay minerals in a Neogene playa

basin. Common borate minerals are borax and ulexite. Borax (and also tincaltonite,

which forms by dehydration of borax) has a similar reflectance spectra to those of

1812 H. Kargi

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

ulexite and colemanite, with a high reflectance at TM band 4 and a lower reflectance

at TM band 7 (JPL 2006, USGS 2006). When both spectra of borax (or tincaltonite)

and ulexite are considered, it is expected that borate minerals in the Kirka region are

also mapped into a PC image. PCA of six TM bands of Landsat showed that

eigenvector loadings for TM4 and TM7 were opposite in sign in two PCs: PC2 and

PC3 (table 2). It is obvious that vegetation is mapped into PC3. In PC analysis,

visible to IR difference is generally mapped into either PC2 or PC3. For this PC

analysis, it should be mapped into PC2; this is, in fact, the case. However, positive

signs of eigenvector loadings for TM4 with the same sign as eigenvector loadings for

visible light bands, and opposite signs of eigenvector loadings for TM4 and TM7 in

PC2 suggest that borate minerals are also mapped into PC2 as bright pixels

(figure 8(a)). Mapping of borate minerals into PC2 instead of PC6 is due to the fact

that the area covered by borate minerals in the image is relatively large. The effect of

the proportion of borate coverage in an image is discussed in detail in the following

section. The reader should note that the presence of water sites in an image may

complicate interpretation of eigenvector matrix, but it is easy to distinguish between

the colour of clear water in the PC image and that of borate minerals (figure 8(a)).

Pixels having the highest digital numbers in the PC2 image fall into the borate pit

and the margins of polluted water sites at which borate precipitates from the water

(figure 8(b)), whereas pixels having the highest digital numbers in the 4/7 image fall

into water and high humidity sites (figure 8(c)).

5.3 Is the proportion of boron anomaly areas in an image important?

In order to effectively use the method in borate exploration, another topic for

discussion is how the proportion of borate coverage in an image affects the

eigenvector matrix and what the image size will be. This is particularly important for

detecting small outcrops having borate anomalies. For this purpose, the left side of

the Bigadic sub-scene, having no borate deposits, was successively excluded in PC

analyses (figure 9(a)) in order to increase step-by-step the proportion of areas

covered by borate minerals. The magnitudes and signs of eigenvector loadings for

TM4 and TM7 in PCs were examined. At first glance, it could be expected that the

magnitudes of eigenvector loadings for TM4 and TM7 in PC6 would increase

gradually as the proportion of borate coverage in the image increases until, after a

certain point, borate minerals are characterized in the first order PCs. Figure 9

shows that the results are somewhat as expected in general, but slightly different in

detail. In all PCAs carried out on the Bigadic sub-scenes (25 sub-scenes), signs of

eigenvectors are positives for TM4 and negatives for TM7 in PC2s (figure 9(a)).

Similar opposite signs are observed in PC6s as signs of eigenvectors in all PCAs are

Table 2. Principal components analysis on six TM bands, Kirka sub-scene.

BandTM1 TM2

TM3 TM4TM5 TM7 Eigenvalues

(%)Eigenvector matrix

PC1 0.373 0.253 0.416 0.260 0.636 0.388 91.55PC2 0.502 0.390 0.430 0.037 20.567 20.295 5.09PC3 0.162 0.030 0.098 20.926 0.009 0.324 2.63PC4 20.487 0.014 0.598 20.202 0.292 20.526 0.41PC5 20.519 0.091 0.376 0.162 20.420 0.614 0.23PC6 0.274 20.879 0.3632 0.073 20.107 0.047 0.06


Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

negatives for TM4 and positives for TM7 in PC6s (figure 9(c)). The most significant

change, as the proportion of borate coverage increases in the sub-scene, happens in

the eigenvector loadings for TM4 and TM7 in PC3s. Figure 9(b) shows that the sign

of eigenvector loading for TM7 is positive in PC3 for a while, then shifts to be

negative, whereas for TM4 it is always positive. The sign shift in the eigenvector

loading for TM4 in PC3 is due to the increased amount of borate coverage in the

sub-scene. It should be noted that eigenvector loadings for TM1s, TM2s, and TM3s

are all negatives and those for TM4s, TM5s, and TM7s are positives in PC3s until a

critical point (sign shift) that shows visible–IR difference is mapped only into PC3.

After the critical point, PC3 not only maps the visible–IR difference but also borate

minerals and vegetation. It may seem to be complex to interpret the change in

eigenvectors as representing the change in the proportion of area covered by borate

minerals in a sub-scene, but in fact the result is obvious and its interpretation is

simple in term of borate exploration: boron anomaly in a Landsat TM image is

(a)

(c)(b)

Figure 8. A PC2 image for the Kirka sub-scene showing borate-rich areas and water sites (a)and anomaly maps for the PC2 (b) and 4/7 (c) images showing approximately 250 mostanomalous pixels of each image. Both borate-rich areas and water sites appear as bright pixelsin the PC2 image (a), but the brightest pixels in the PC2 image correspond to borate-richareas, which are typically the pit and margins of the water sites just near the pit (b), whereasthose in the 4/7 image correspond to water and high humidity sites (c).

1814 H. Kargi

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

characterized by opposite sign eigenvector loadings for TM4 and TM7 in two or

more PCs, depending on the proportion of boron anomaly area in the image. This

actually is a very useful result for mineral exploration in that one can divide an

image into small sub-scenes and carry out PCA on each sub-scene to detect possible

small outcrops having boron anomalies. If the signs of eigenvector loadings for TM4

and TM7 are opposite in two or more PCs, coordinates of anomalous pixels in the

sub-scene are marked as a target area for a ground study.

(a)

(b)

(c)

Figure 9. Eigenvector loadings for TM4 and TM7 in PC2s (a), PC3s (b), PC6s (c), obtainedby successively excluding left sides of the Bigadic sub-scene in PCAs.


Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

6. Conclusions

PCA on Landsat TM data in mineral exploration has been used previously by

researchers for alteration mapping (Crosta and Moore 1989, Loughlin 1991,

Ranjbar et al. 2004). Variances in the last three principal components (PC4, PC5,

and PC6) were proposed to result from iron oxide and hydroxyl cover on the surface

(Loughlin 1991, Ranjbar et al. 2004). The PCA method was adopted here for

mapping borate minerals. The results reported in this paper show that borate

minerals might be mapped into principal component images if borate coverage in an

image was high enough. Usefulness of the method in borate exploration depends on

the selection of image size and concentration of borate in the soil. The image size

should be as small as possible in order to increase the possibility of obtaining

opposite signs for eigenvector loadings for TM4 and TM7 in two or more PCs.

Acknowledgements

The author would like to thank E. Cloutis and an anonymous reviewer who

substantially improved the style, organization, and content of this paper.

ReferencesABDELSALAM, M.G., STERN, R.J. and BERHANE, W.G., 2000, Mapping gossans in arid

regions with Landsat TM and SIR-C/X-SAR imagery: the Beddaho alteration zone in

Northern Eritrea. Journal of African Earth Sciences, 30, pp. 903–916.

BENNETT, S.A., ATKINSON, W.W. and KRUSE, F.A., 1993, Use of thematic mapper imagery to

identify mineralization in the Santa Teresa District, Sonora, Mexico. International

Geology Review, 35, pp. 1009–1029.

BRYANT, R.G., 1996, Validated linear mixture modeling of Landsat TM data for mapping

evaporite minerals on a playa surface: methods and applications. International Journal

of Remote Sensing, 17, pp. 315–330.

CROSTA, A.P. and MOORE, J., 1989, Enhancement of Landsat Thematic Mapper imagery for

residual soil mapping in SW Minais Gerais State, Brazil: a prospecting case history in

Greenstone belt terrain. In Proceedings of the 7th Thematic Conference on Remote

Sensing for Exploration Geology, 2–6 October 1989, Calgary, Alberta, Canada (Ann

Arbor, MI: Environmental Research Institute of Michigan), pp. 1173–1187.

GOOSENS, M.A. and KROONENBERG, S.B., 1994, Spectral discrimination of contact

metamorphic zones and its potential for mineral exploration, province of

Salamanca, Spain. Remote Sensing of Environment, 47, pp. 331–344.

HELVACI, C. and ALACA, O., 1991, Bigadic borat yataklari ve cevresinin jeolojisi ve

mineralojisi. MTA Dergisi, 113, pp. 61–92.

JPL, 2006, Spectral library. Available online at: http://speclib.jpl.nasa.gov (accessed 7 March

2006).

KAUFMANN, H., 1988, Mineral exploration along the Aqaba–Levant structure by use of TM

data; concepts, processing, and results. International Journal of Remote Sensing, 9, pp.

1639–1658.

KNEPPER, D.H. and SIMPSON, S.L., 1992, Remote sensing in geology and mineral resources of

the Altiplano and Cordillera Occidental, Bolivia. U.S. Geological Survey Bulletin,

1975, pp. 47–55.

KUSKY, T.M. and RAMADAN, T.M., 2002, Structural controls on Neoproterozoic

mineralization in the South Eastern Desert, Egypt: an integrated field, Landsat

TM, and SIR-C/X SAR approach. Journal of African Earth Sciences, 35,

pp. 107–121.

LOUGHLIN, W.P., 1991, Principal components analysis for alteration mapping.

Photogrammetric Engineering and Remote Sensing, 57, pp. 1163–1169.

1816 H. Kargi

Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

POSADAS, A.N.D., QUIROZ, R., ZOROGASTU, P.E. and LEO N-VELARDE, C., 2005, Multifractal

characterization of the spatial distribution of ulexite in a Bolivian salt flat.

International Journal of Remote Sensing, 26, pp. 615–627.

RANJBAR, H., HONARMAND, M. and MOEZIFAR, Z., 2004, Application of the Crosta

technique for porphyry copper alteration mapping, using ETM + data in the southern

part of the Iranian volcanic sedimentary belt. Journal of Asian Earth Sciences, 24, pp.

237–243.

ROWAN, L.C. and BOWERS, T.L., 1995, Analysis of linear features mapped in Landsat

thematic mapper and side-looking airborne radar images of the Reno, Nevada 18u by

28u quadrangle, Nevada and California — implications for mineral resource studies.

Photogrammetric Engineering and Remote Sensing, 61, pp. 749–759.

SABINS, F.F., 1999, Remote sensing for mineral exploration. Ore Geology Reviews, 14, pp.

157–183.

SABINS, F.F. and MILLER, R.M., 1994, Resource assessment—Salar Uyuni and vicinity. In

Proceedings of Tenth Thematic Conference on Geologic Remote Sensing, 9–12 May

1994, San Antonio, Texas (Ann Arbor, MI, Environmental Research Institute of

Michigan), pp. I92–I103.

SAHOO, P.K., SOLTANI, S., WONG, A.K.C. and CHEN, Y.C., 1988, A survey of thresholding

techniques. Computer Vision, Graphics, and Image Processing, 41, pp. 233–260.

SINGH, A. and HARRISON, A., 1985, Standardized principal components. International Journal

of Remote Sensing, 6, pp. 883–896.

SONG, C., WOODCOCK, C.E., SETO, K.C., LENNEY, M.P. and MACoMBER, S.A., 2001,

Classification and change detection using Landsat TM data: when and how to correct

atmospheric effects? Remote Sensing of the Environment, 75, pp. 230–244.

USGS, 2006, USGS Spectroscopy Lab. Available online at: http://speclab.cr.usgs.gov

(accessed 7 March 2006).


Dow

nloa

ded

by [

Dok

uz E

ylul

Uni

vers

ity ]

at 2

1:37

06

Nov

embe

r 20

14

Documents

Principal components analysis for borate mapping