Ore mineral discrimination using hyperspectral remote sensinga field-based spectral analysis U.A.B.Rajasimman Balasubramanian, J.Saravanavel & S.

Gunasekaran

Arabian Journal of Geosciences ISSN 1866-7511 Arab J Geosci DOI 10.1007/s12517-012-0721-4

1 23

Your article is protected by copyright and all rights are held exclusively by Saudi Society for Geosciences. This e-offprint is for personal use only and shall not be self-archived in electronic repositories. If you wish to selfarchive your work, please use the accepted authors version for posting to your own website or your institutions repository. You may further deposit the accepted authors version on a funders repository at a funders request, provided it is not made publicly available until 12 months after publication.

1 23

Author's personal copy

Arab J Geosci DOI 10.1007/s12517-012-0721-4

ORIGINAL PAPER

Ore mineral discrimination using hyperspectral remote sensinga field-based spectral analysis
U. A. B. Rajasimman Balasubramanian & J. Saravanavel & S. Gunasekaran
Received: 1 March 2012 / Accepted: 16 October 2012 # Saudi Society for Geosciences 2012

Abstract Advanced Spaceborne Thermal Emission and Reflection Radiometer data of Salem District and field-based spectral observations using SVC HR 1024 spectral radiometer is used to make a clear discrimination of ore mineral deposits in parts of Salem District of Tamil Nadu. Spectral analyses, one of the most advanced techniques, are used to discriminate the magnesite deposits in the central northern part of Salem District. Different spectral processes were used in ore discrimination, which include the following: (1) atmospheric correction (FLAASH), (2) minimum noise fraction and (3) pixel purity index preparation which helps in discrimination by matching these purest pixels with field spectral observations. Spectral angle mapper method is used to produce score between 0 and 1, where the value of 1 makes a perfect match showing the exact ore deposit in the study area. Using these techniques, we were able to find two ore deposits in the study area, i.e. magnesite and lateritic bauxite, recording different scores related to their abundance. Keywords ASTER image . Spectral analysis . SAM . Ore deposit

Introduction Hyperspectral imagery has been particularly very effective in mapping the lithological variations and the alteration minerals.
U. A. B. R. Balasubramanian (*) : J. Saravanavel : S. Gunasekaran Centre for Remote Sensing, Bharathidasan University, Khajamalai Campus, Tiruchirappalli 620 023, India e-mail: rajasimmanuab@hotmail.com J. Saravanavel e-mail: rajasimmanuab@hotmail.com S. Gunasekaran e-mail: rajasimmanuab@hotmail.com

Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images can reveal successfully the mineralogical variations and alterations induced by both physical and chemical factors which represent the opening of mineral explorations that could make a direct appeal of growth in economic conditions. Spectral mapping methods such as spectral angle mapper technique were adopted by many researchers for lithological discriminations, vegetation analysis, etc. in remote sensing studies. These algorithms are implemented based on the comparative analysis of the spectral responses of the purest materials that are observed in the field with the pixel spectrum of the purest pixels. Reflectance spectra of the samples are obtained using a portable spectral radiometer SVC HR 1024 that covers the wide range of spectrum from 0.35 to 2.5 m, i.e. covers the spectrum ranges of visible light, visible and near-infrared (VNIR) and short-wave infrared (SWIR). These ranges of spectrum can be easily applied for analysis over ASTER image, since the ranges covered in ASTER are merely close to them as cited in Table 1. The obtained spectrum in the field shows absorption features in the visible to short-wave infrared wavelength region of the EM spectrum, caused primarily by electronic transitions and vibrational transitions. The purpose of the study is to employ spectral angle mapper technique on the ASTER data using field-observed spectral reflectance data to discriminate the mineral ore deposits in the study area. Analysis focuses on the spatial distribution of the main deposits of mineral ore like magnesite, lateritic bauxite, etc. Geological review The study area of the Salem District is endowed with economically exploitable deposits magnesite, bauxite and minerals such as quartz, etc. The area is made up of high-grade supra-crustals of Archaean age, comprising khondalite group, charnockites group and satyamangalam group and younger intrusive alkaline syenitecarbonatite complex, ultramafic, basic and acid rocks. The khondalite group occurs

Author's personal copy

Arab J Geosci Table 1 Spectral ranges covered in ASTER multispectral images Subsystem Band Spectral range 0.520.60 0.630.69 0.780.86 0.780.86 1.601.70 2.1452.185 2.1852.225 2.2352.285 2.2952.365 2.3602.430 8.1258.475 8.4758.825 8.9259.275 10.2510.95 10.9511.65 Spatial resolution 15 m Quantization levels 8 bits

SVC HR1024a spectral radiometerwas used to acquire field sample reflectance in the study area. The observed sites of the field samples are tabulated in Table 2. Selection of minerals Out of 27 minerals present in Salem District, four samples are selected for analysis according to the availability of area of interest. The main aim of this study is to make a clear discrimination of ore minerals like magnesite along with minerals like quartz, orthoclase and anorthosite; this was done in such a way that the mineral has varied physical properties(Table 3) within the study area. Field spectral response curves The field spectral signatures of the minerals are shown in Fig. 1. After collecting the spectral curves for every individual mineral, they are then correlated with the spectral library of USGS for the same minerals. By this correlation, the following characters are studied: (a) Spectral character of Quartz Figure 2a shows the USGS spectral curve for quartz; the spectral signature shows the absorption at two wavelengths: the first one is at 1,440 nm and final absorption is at 2,180 nm. Even though the field spectral observation (b) shows a fluctuation of absorption and reflection at 1,800 and 2,500 nm, they show similar absorption characteristics at both 1,480 and 2,000 nm. Thus, the spectral signatures are similar in both the USGS and field absorption. (b) Spectral character of Orthoclase Figure 3a, b displays that the spectral signatures show absorption points at 1,440, 1,920 and 2,150 nm and flow in a similar pattern of curve. Thus, the spectral signatures are similar in both the USGS and SVC spectral signature curves. (c) Spectral character of Anorthite Figure 4a, b displays that the spectral signature of anorthite shows absorption points at 1,415, 1,945 and 2,337 nm. Though the spectral flow is not so similar to USGS curve pattern, the field spectra consist of
Table 2 Observed sites of field sample Sl. no. 1 2 3 4 5 Sample Magnesite Magnesite Quartz Orthoclase Anorthite Location Magnesite mine Kanjamalai Kanjamalai West Sankagiri Sithampundi Latitude 11.72267 11.62315 11.61882 11.495452 11.3333 Longitude 78.16755 78.03911 78.02157 77.87821 78.02

VNIR

SWIR

1 2 3N 3B 4 5 6 7 8 9 10 11 12 13 14

30 m

8 bits

TIR

90 m

12 bits

as dismembered parallel lenses of varying width and length in the Attur valley. It comprises garnetiferous silimanite gneiss as interbands within charnockites. The charnockites in the Shevaroy Hills are altered to bauxite and laterite. The most prominent deposit of magnesite is located in Chalk Hills of Salem over an area of 17 km2 and estimated to be 44 million tons. This is used by Burn & Co, TANMAG, TATA Refractories, Dalmia and Ramakrishna magnesite for the manufacture of dead burnt magnesite and calcined magnesite. In Shevaroy Hills of Yercaud taluk, six bauxite occurrences are known in Semmaduvu, Manjakuttai and Puliyur villages. The bauxite is being exploited by MALCO for their aluminium plant at Mettur Dam.

Materials and methods The ASTER data used in this study are level 1B data; it is pregeoreferenced to geographical latitude/longitude with WGS1984 datum. The SWIR spectrum data of the ASTER are then resampled to VNIR spectrum data so that all the 9band data which are taken for the analysis could have the same 1515m pixel size. Atmospheric correction (FLAASH) was applied on the ASTER data and vegetation index (NDVI) was calculated to mask the vegetation canopy in the study area. The study mainly carried out into two levels. Of which, the first category deals with the comparative study of spectral reflectance obtained using the spectral radiometer with the USGS and JHU spectral library for various rocks and minerals. The second category is to validate the spectral signatures over ASTER image analysis under various digital image processing techniques.

Author's personal copy

Arab J Geosci Table 3 Minerals and their physical properties Mineral Color Cleavage Fracture Transparency Hardness Sp. gravity Occurrence Chemical composition SiO2 KAlSi3O8 MgCo3 CaAl2Si2O8

Quartz Orthoclase Magnesite Anorthite

Colorless Pink White Grey

Absent Two set Perfect rhombohedral Two set

Concoidal Concoidal Concoidal Concoidal to Uneven

Transparent Sub transparent to translucent Transparent to Opaque Sub transparent to translucent

7 6.06.5 3.54.5 6.06.5

2.65 2.56 34.8 2.56

Acid igneous rock Alkali and acid igneous rocks Irregular veins and fracture zones Alkali and acid igneous rocks

absorption and reflection points same as those of the USGS. Hence, it can be concluded from the spectral signatures that they are similar. (d) Spectral character of Magnesite Similarly, Fig. 5a, b displays that the spectral

signatures of magnesite show absorptions at 1,480, 1,990 and 2,300 nm. Here, the magnesite was collected from the magnesite mine. Plot b shows a clear portrait pattern of spectral response curve as that of the USGS spectral library. Although the spectral response curve

Fig. 1 Field spectral observation map

Author's personal copy

Arab J Geosci

Fig. 2 Spectral properties of quartz observed in both USGS library and field spectra collection

of the magnesite observed in Kanjamalai Hills, which shows undefined hikes and drops in-between the absorption and reflection wavelengths, may be caused by the weathering properties present in the field sample, it may also be due to manual interruptions or to the influence of shadow effect while making the observation. Thus, the spectral signatures collected from the magnesite mine are considered as the true grade of magnesite and it is taken for the analysis.

varied in their mode only at the presence of weathering character or the presence of soil particles over it. Spectral analysis with ASTER image The ASTER data L1B are used for this process. It is atmospherically corrected using the FLAASH module by entering the parameters like scene and sensor information, sensor altitude, ground elevation, scene centre location (latitude and longitude), flight time GMT, flight date, atmospheric model, aerosol model, etc.; these corresponding details are collected from the metadata of the ASTER.HDF file. After the data are atmospherically corrected, the DN value of the data now shows the original reflectance value of the terrain features. Then, the atmospherically corrected data (Fig. 6a) are to be made vegetation-free data so that further analysis will not be interrupted by the vegetation value of the study area. For this process, the image is

Observations From the comparative analysis of the spectral response curves of the selected minerals (Clark 1999), it is observed that the spectral pattern is followed in their response flow by acquiring the reflection and absorption pattern and it gets

Fig. 3 Spectral properties of orthoclase observed in both USGS library and field spectra collection

Author's personal copy

Arab J Geosci

Fig. 4 Spectral properties of anorthite observed in both USGS library and field spectra collection

calibrated with NDVI module to identify the vegetation indices of the study area, and by studying the vegetation statistics of the terrain, the minimum and maximum values are noted (minimum, 4.0149344; maximum 0.891533). Here, the maximum value indicates very thick vegetation present in the area. Particularly in the study area, Shevaroy Hills is covered by thick forest vegetation. And some of the area is covered by agricultural plantation. The conversion and change in the spectral characteristics are clearly observed in Fig. 6. To remove the vegetation canopy in the study area, masking technique is used were a mask is created with the low abundance value which is calculated using the formula NDVI NIR RED=NIR RED: From the resultant output, the minimum (4.014493) and maximum (0.591533) values are noted and saved as masking layer which is then applied over the atmospherically

corrected image that makes the thick canopy of vegetation cover present in the study area get masked and a new image is created with minimum vegetation influence for every individual VNIR and SWIR bands data. This masked image (Fig. 6b) is more prompt for analysing the lithology by spectral analyses techniques. The newly saved masked image is processed with minimum noise fractionforward rotation statistics to identify the spectral anomalies in the image which acts as the indicative of alterations. MNF rotation transforms to determine the inherent dimensionality of image data, to segregate noise in the data and to reduce the computational requirements for subsequent processing (Boardman and Kruse 1994). The MNF transform as modified from Green et al. and implemented in ENVI is a linear transformation that consists of the following separate principal components analysis rotations:

Fig. 5 Spectral properties of magnesite observed in both USGS library and field spectra collection

Author's personal copy

Arab J Geosci

Fig. 6 a The atmospherically corrected image and b the vegetation DNs masked output

Fig. 7 Minimum noise fraction forward rotation output of ASTER bands

Author's personal copy

Arab J Geosci

&

&

The first rotation uses the principal components of the noise covariance matrix to decorrelate and rescale the noise in the data (a process known as noise whitening), resulting in transformed data in which the noise has unit variance and no band-to-band correlations. The second rotation uses the principal components derived from the original image data after they have been noise-whitened by the first rotation and rescaled by the noise standard deviation. Since further spectral processing will occur, the inherent dimensionality of the data is determined by examining the final eigenvalues and the associated images.

The output band of MNF rotation (Fig. 7) is processed to get pure pixels since the dimensionality is calculated from the variance level acquitted from bands 1, 2, 3, 4, 5, 7 and 8, and their noise is reduced to further process for the identification of their standard deviation reflectance value, i.e. 2.800, which was done to delineate the purer pixels from more mixed pixels.
Fig. 8 Spectral angle mapper class image for the selected ROIs

From the delineated end members, it is further categorised as region of interest that is prepared for the minerals identified by the maximum score values. For this categorisation of region of interest, these end members are subjected to ndimension visualisations where it can be correlated, through spectral analysis, with the spectral library that is prepared using the calibrated spectral signatures of minerals collected from the field using spectral radiometer. By correlating the end members with the spectral data acquired in field observation, the pure pixels show maximum of 1.023 for the spectral signature of magnesite but the other minerals did not make their scores up to their level in the pure pixel ranges of the study area; hence, there is no categorisation of the minerals like quartz, orthoclase and anorthite. So the end members are correlated with the USGS inbuilt spectral library for any suspects of categorisation. Here, the end members had shown maximum scores for lateritic bauxite and charnockite, so they are also categorised in to region of interest based on their maximum scores for the corresponding mineral spectra in

Author's personal copy

Arab J Geosci

the library. For these categorised end members, region of interest is applied over the VNIR and SWIR bands of ASTER data in spectral angle mapping technique. The angle of spectral identification is set to a default value of 0.2. On applying the spectral angle mapper (SAM) process over the ASTER 9 band data (Crosta et al. 2003; Crsta and De Souza Filho 2000; Kruse and Perry 2006; Gersman et al. 2007; Das 2002), the SAM rule images for the identified minerals and rocks (region of interest (ROI)) (Lnio Soares et al. 2005). After defining the default threshold values for the corresponding ROIs, it prepares the SAM class image that shows the actual position of occurrence of the targeted minerals as shown in Fig. 8; the resultant image shows a clear distinction for every ROI (Tangestani 2006). Since the canopy was thick over the mountain region, there are no pure pixels selected in those areas, so the resultant output of SAM does not show much variation in its processing for the area and it remains the same as that which appeared after vegetation was masked. Hence, those areas are lithologically marked as unknown.

the change of values of the minerals between USGS spectra and device spectra may be the alteration representation/ influence of sediments or foreign matters over it/the gradation change of weathering processes. The lab analysis of every mineral under lab condition may give the evidences of cause of the change in values of the spectra.

References
Boardman JW, Kruse FA (1994) Automated spectral analysis: A geological example using AVIRIS data, northern Grapevine Mountains, Nevada. In: Proceedings, Tenth Thematic Conference, Geologic Remote Sensing, 912 May 1994, San Antonio, Texas, p. I-407I-418 Clark RN (1999) Spectroscopy of rocks and minerals and principles of spectroscopy. In: Rencz AN (ed) Remote sensing for the earth sciences: manual of remote sensing, vol 3. Wiley, New York, pp 358 Crosta AP, De Souza CR, Azevedo F, Brodie C (2003) Targeting key alteration minerals in epithermal deposits in Patagonia, Argentina, using ASTER imagery and principal component analysis. Int J Remote Sens 24(21):42334240 Crsta AP, De Souza Filho CR (2000) Hyperspectral remote sensing for mineral mapping: a case-study at Alto Paraso De Goas, Central Brazil. Revista Brasileira de Geocincias 30(3):551554 Das IC (2002) Spectral signatures and spectral mixture modeling as a tool for targeting aluminous laterite and bauxite ore deposits. GISdevelopment.net, Koraput, India Gersman R, Ben-Dor E, Beyth M, Avigad D, Abraha M, Kibreab A (2007) Hyperspectral remote sensing as a tool for geological explorationexamples from the Northern Danakil Depression, Eritrea. Proceedings of the 5th EARSeL workshop on imaging spectroscopy. Bruges, Belgium, April 2325 2007, pp: 114. Kruse FA, Perry SL (2006) Regional mineral mapping by extending hyperspectral signatures using multispectral data. IEEEAC paper #1078, version 4, updated November 24, 2006. pp: 115. Lnio Soares G, Almeida Filho R, caro V (2005) Use of ASTER short-wave infrared bands for the spectral discrimination of hydrothermally altered-materials: evaluation in a tropical savannah environment. Anais XII Simpsio Brasileiro de Sensoriamento Remoto, Goinia, Brasil, 1621 April 2005, INPE, p. 17831787. Tangestani MH (2006) A comparative approach on TIR and VNIRSWIR datasets of ASTER instrument for lithological mapping in Neyriz Ophiolite Zone, SW Iran. Proceedings of Map Asia 2006, 29 Aug.1 Sept., 2006, Bangkok, Thailand.

Result and conclusion From the analysis of ASTER using the field observation data, it is concluded that the instrument used for field analysis is more similar as the ASTER sensor. Since the SAM analysis was attempted with nine-band data of ASTER using the field observation which consists of the same wavelength signature, it is proved that the field observation using the device can be directly applied for ASTER analysis without resampling them as what is done for USGS spectra. The USGS spectra cover a wide range of wavelength from 0.35 to 14.00 m, so in the case of spectral analysis with ASTER, it is required to resample them to ASTER wavelengths to acquire the best result. Moreover, with the spectral analysis over thick canopy, it was impossible to discriminate the lithology beneath it but the overall influence of vegetation mask can be reduced to make a perfect result. The calibration of the instrument data with USGS had been made up to an acceptance level only;