Hyperspectral remote sensing of evaporate minerals and associated ...
(This is a sample cover image for this issue. The actual cover is not yet available at this time.)
This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright
Author's personal copy International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Contents lists available at SciVerse ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Hyperspectral remote sensing of evaporate minerals and associated sediments in Lake Magadi area, Kenya Gayantha R.L. Kodikara a,∗ , Tsehaie Woldai b , Frank J.A. van Ruitenbeek b , Zack Kuria c , Freek van der Meer b , Keith D. Shepherd d , G.J. van Hummel e a
Department of Oceanography and Marine Geology, Faculty of Fisheries & Marine Science and Technology, University of Ruhuna, Sri Lanka University of Twente, Faculty of Geo-information Science and Earth Observation (ITC), Hengelosesestraat 99, P.O. Box 6, 7500AA Enschede, The Netherlands c Department of Geology, University of Nairobi, Nairobi, Kenya d World Agroforestry Centre (ICRAF), P.O. Box 30677-00100, Nairobi, Kenya e Institute for Nanotechnology (MESA+), University of Twente, The Netherlands b
a r t i c l e
i n f o
Article history: Received 1 September 2010 Accepted 18 August 2011 Keywords: Hyperion Lake Magadi MTMF Surface mineral mapping Magadiite Kenyaite
a b s t r a c t Pleistocene to present evaporitic lacustrine sediments in Lake Magadi, East African Rift Valley, Kenya were studied and mapped using spectral remote sensing methods. This approach incorporated surface mineral mapping using space-borne hyperspectral Hyperion imagery together with laboratory analysis, including visible, near-infrared diffuse reflectance spectroscopy (VNIR) measurements and X-ray diffraction for selected rock and soil samples of the study area. The spectral signatures of Magadiite and Kenyaite, which have not been previously reported, were established and the spectral signatures of trona, chert series, volcanic tuff and the High Magadi bed were also analyzed. Image processing techniques, MNF (Minimum Noise Fraction) and MTMF (Mixture Tuned Matched Filtering) using a stratified approach (image analysis with and without the lake area), were used to enhance the mapping of evaporates. High Magadi beds, chert series and volcanic tuff were identified from the Hyperion image with an overall mapping accuracy of 84.3%. Even though, the spatial distribution of evaporites and sediments in Lake Magadi area change in response to climate variations, the mineralogy of this area has not been mapped recently. The results of this study shows the usefulness of the hypersspectral remote sensing to map the surface geology of this kind of environment and to locate promising sites for industrial open-pit trona mining in a qualitative and quantitative manner. © 2011 Elsevier B.V. All rights reserved.
1. Introduction Lake Magadi, a dry alkaline saline lake, located in the East African Rift Valley in Kenya (Fig. 1A) contains the most concentrated brines found in the African Rift Valley (Warren, 2006). It contains a large amount of trona (Na2 CO3 ·NaHCO3 ·2H2 O), chert (SiO2 ), Magadiite (NaSi7 O13 (OH)3 ·3H2 O) and Kenyaite (NaSi11 O20.5 (OH)4 ·3H2 O). Little Magadi, or Lake Nasikie Engeda, lies 1.6 km north of Lake Magadi from which it is separated by a narrow horst (Fig. 1C). Little Magadi is perennially saline, while Lake Magadi is intermittently dry. Hydrothermal circulation associated with the numerous active alkaline volcanoes and their feeder faults, supply hot alkaline brines to the many hot springs situated along the edge of the lakes and this system is responsible for the ongoing geochemical evolution of the area (Warren, 2006). The Pleistocene and Holocene history of the Magadi basin sediments and their geochemical evolution is fully described in the work of Baker (1958) and Eugster (1969). The
∗ Corresponding author. Tel.: +94 415674048. E-mail address: gayantha [email protected] (G.R.L. Kodikara). 0303-2434/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jag.2011.08.009
recent geologic history is better understood by studying the spatial distributions of minerals in the area. The interpretation and reconstruction of this geochemical evolution, however, is difficult to establish due to variation in mineralogy (e.g., not easily recognizable in the field) and the large spatial extent that limits accessibility (Warren, 2006). Alternatively, access to relatively inexpensive satellite-borne multi-spectral and hyper-spectral data have created new opportunities for the regional mapping of mineralogy, geological structures and rock types including alteration products (Hewson et al., 2005; Vaughan et al., 2005). Such datasets provide identification of different surface expressions and mapping possibilities for minerals in the hydroxyl, silicate, sulfate, carbonate and iron oxide groups covering large areas and in inaccessible terrains. It has been applied in efflorescent salt crusts in Death Valley, California using airborne hyperspectral and space-borne multispectral imagery to map the different surface saline materials (Crowley, 1993; Crowley and Hook, 1996). The motivation of this research was to evaluate the identification and mapping capability of various evaporites and precipitates in Lake Magadi area using space-borne hyperspectral Hyperion data
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
23
Fig. 1. Location of the Kenya Rift in the East African Rift System (A) and structural sketch of the Kenya rift and location of the Magadi Lake (B) (modified after Turdu et al. (1999)). (C) Geological Map of Lake Magadi (modified after Baker (1958) and Eugster and Jones (1968)).
and their spectral characteristics. In addition to that, the resultant map will be useful for the mining industry to show areas that are favourable for the formation of lacustrine mineral deposits and for planning access for trona mining. 1.1. Geologic setting Lake Magadi (Fig. 1) is located in the deepest depression of the Kenyan Rift in Kenya, surrounded by hills formed by Pleistocene alkali trachyte flows. The valleys of the Lake Magadi and Little Magadi Lake are covered by a series of Lake Beds dating from the middle Pleistocene to present. They consist of vast deposits of trona, bedded cherts and High Magadi beds derived from Magadiite horizons (Fig. 1C). Mineral precipitation, re-solution, re-precipitation and reactions, which are common in this environment, can be understood based on the hydrology and aqueous geochemistry of the area. Atmospheric precipitation, overland flow and percolation of water can react with alkali trachyte terrains and this process can be illustrated by the reaction of albite feldspar with water and aqueous CO2 (Jones et al., 1977): NaAlSi3 O8(s) + 2H2 O(aq) + CO2(aq) = Na+ (aq) + HCO− + 3SiO2(aq) + Ai(OH)3(aq) 3(aq)
After accumulation of these solutes in the lake from rainfall and hydrolysis, the waters are subjected to evaporation, either at the surface or by capillarity processes. This chemistry is ideal for ongoing trona precipitation at higher salinities (Warren, 2006). Trona is the first Na-carbonate mineral to precipitate in equilibrium with concentrating lake brines. + HCO3− + CO2− 2H2 O → Na2 CO3− NaHCO3− 2H2 O(s) 3Na+ (aq) 3(aq) (aq) (Trona)
The composition of the Magadi-trona is straightforward. It consists of trona crystals (sodium sesquicarbonate – Na2 CO3 ·NaHCO3 ·2H2 O), sodium fluoride (NaF), and rarely small amounts of common salts (NaCl) (Baker, 1958). These trona deposits extend for approximately 47 km2 in area with 7–50 m thickness. The Trona is locally known as the “Evaporite Series” which is made up of cm-scale stacked, upward-pointing and growth-aligned trona crystals (Warren, 2006). Precipitation of chert is not straightforward. It can occur through two types of hydrous sodium silicate intermediate products called Magadiite (NaSi7 O13 (OH)3 ·3H2 O) and Kenyaite (NaSi11 O20.5 (OH)4 ·3H2 O). Magadiite occurs in substantial quantities, typically as beds, mounds and nodules. Kenyaite only occurs as nodules in the bedded Magadiite in the High Magadi bed (Warren,
Author's personal copy 24
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 2. Crystallized trona (A001) and its powdered form (B002). One-Euro coin is used for scale.
2006). The precipitation of Magadiite is represented by the following reaction (Eugster, 1969). = NaSi7 O13 (OH)3 .3H2 O(s) + 9H2 O(aq) + H+ 7H4 SiO4(aq) + Na+ (aq) (aq) (Magadiite)
The conversion of Magadiite to chert (SiO2 ) through dehydration and sodium loss may involve Kenyaite (NaSi11 O20.5 (OH)4 ·3H2 O) as an intermediate phase, 22NaSi7 O13 (OH)3− 3H2 O(s) +8H+ → 14NaSi11 O20.5 (OH)4− 3H2 O(s) +8Na+ +33H2 O (aq) (aq) (Magadiite)
(Kenyaite)
NaSi11 O20.5 (OH)4 .3H2 O(s) + H+ → 11SiO2(s) + Na+ + 5 1/2H2 O (aq) (aq) (Kenyaite)
(Chert)
or it may directly produce a silica phase (Eugster, 1969). → 7SiO2(s) + Na+ + 5H2 O NaSi7 O13 (OH)3 .3H2 O(s) + H+ (aq) (aq) (Magadiite)
(chert)
Outcrops of High Magadi beds are located in the alluvial flats and at the foot of the hills surrounding Lake Magadi. The mineralogy of High Magadi is characterized by detrital silicate, saline minerals, calcite, sodium silicates, including Magadiite, Kenyaite, chert and zeolites (Surdam and Eugster, 1976).
Fig. 4. Reflectance spectra of Trona samples. Arrows indicate absorption features of trona.
1.2. Mapping minerals with hyperspectral remote sensing
Fig. 3. Surface evaporates at sample location of P018.
Reflectance spectroscopy offers a rapid, inexpensive and nondestructive tool for determining the mineralogy of rock and soil samples (Gaffey, 1985; Salisbury and D’Aria, 1992; Van der Meer, 2001). In the visible and shortwave infra-red wavelength region (0.4–2.5 m), iron-, hydroxyl-, sulfate-, water and carbonate bearing minerals display diagnostics spectral features (Clarck et al., 1990; Van der Meer, 2001). The absorption features (position, shape, depth and width) are controlled by the particulate crystal structure in which the absorbing species is contained and by the chemical structure of the material (Van der Meer, 2001). Therefore, absorption features are used to identify the type of mineral in reflectance spectroscopy and Imagine spectrometry is used to map those minerals (Kruse, 1988).
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
25
Fig. 5. (A) Crystal casts in chert plates from Southern end of Lake Magadi. (B) Laminated green chert from northern part of the Lake Magadi. (C) Pillow chert near to Magadi town. (D) Chert dykes from North-eastern part of Lake Magadi. (E). Green beds from southern part of Lake Magadi. (F). High Magadi Beds from southern part of Lake Magadi.
Imaging spectrometry which is defined as “the simultaneous acquisition of images in many narrow, contiguous spectral bands” (Goetz et al., 1985) allows extraction of a detailed spectrum for each picture element (pixel) of the image. High spectral resolution reflectance spectra collected by imaging spectrometers allow direct identification of individual materials based upon their reflectance characteristics such as absorption band depth, band position, asymmetry of the absorption features, etc. Mapping of surface reflectance in an imaging spectrometer data set with known diagnostic reflectance spectra of minerals is basically done using spectral matching techniques or sub-pixel classification techniques (Van der Meer, 2006). Two groups of spectral matching techniques have been developed: deterministic empirical measures, including spectral angle mapper (SAM), spectral correlation measures (SCM), Euclidean distance measures (ED), and Mixture Tuned Matched Filtering (MTMF) method (Mason, 2002); and Stochastic measures evaluating the statistical distributions of spectral reflectance values of target end members (Van der Meer, 2006). In this paper the Mixture Tuned Matched Filtering (MTMF) method was used to map surface mineral of the study area. The main advantage of the MTMF method, unlike traditional spectra mixture modeling, is the ability to map a single known target without knowing the other background end member signatures (Kruse, 2003). 1.3. Hyperion satellite sensor imagery Hyperion is the first Earth-orbiting imaging spectrometer, which was launched onboard the Earth Observing 1 (EO-1) satellite on November 21, 2000. Operating across the full solar-reflected spectrum, the nominal spectral coverage ranges from 0.4 m to 2.5 m, with 10 nm sampling and spectral response functions (King et al., 2003). The radiometric range of Hyperion spans from zero to the maximum Lambertian reflected radiance with 12 bits of digitization (King et al., 2003). The Hyperion push broom instrument captures the image frame spectra from an area of 30 m along-track by 7.7 km cross-track (Pearlman et al., 2003). Hyperion has one telescope, slit and two grating spectrometers: one spectrometer with 70 channels in the VNIR wavelength region (0.4–1.0 m) with
silicon detector array and one spectrometer with 172 channels in SWIR wavelength region (0.9–2.5 m) with an HgCdTe detector array (Biggar et al., 2003). The 196 unique channels covering VNIR (band 8–57) and SWIR (band 77–224) regions are calibrated from 242 total channels (Hubbard et al., 2003). 2. Methods The identification and mapping of evaporate minerals and associated sediments in the Lake Magadi area, involved the collection of rocks and soils in the field, their analyses using reflectance spectroscopy and X-ray diffraction, and the processing and interpretation of Hyperion hyperspectral satellite imagery. 2.1. Field rock/ soil sampling Fieldwork was conducted in the Lake Magadi area in July and August 2008. 84 rock and soil samples, representative for the entire study area, were collected from relatively homogeneous bare land areas having spatial extent larger than 60 m × 60 m. In addition, several other samples were collected without considering the homogeneity and the spatial extent of the sample location due to the geological interest of the area. Each sample location was also subjected to land cover and geomorphologic analysis, GPS data collection and field photographs. 2.2. Spectral reflectance analysis The soil and rock samples were dried in the open air. Once dried, all soil samples were crushed and passed through a 2 mm sieve. Reflectance spectra of the 26 soil samples and 58 rock samples were acquired using Multi Purpose Analyzer – Fourier Transformed Spectrometer (MPA – FTIR) and Vertex-70 FTIR instrument spectrometer fitted with a diffuse reflectance accessory over the wavelength range from 350 to 2500 nm. All the rock and soil samples were grouped based on their spectral signatures and the physical properties/appearance of the samples. The position, shape, depth and width of absorption features are used to identify the
Author's personal copy 26
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 7. Reflectance spectra of Kenyaite and Magadiite rock samples.
Fig. 6. Reflectance spectra of siliceous rocks (P006 1 and S032: chert with small cavities filled by carbonate minerals, P019 1: Dyke chert, S008: Quartz chert, P035 and S047: Pillow chert, P017: Laminated chert, P024: Laminated green chert).
mineralogy from field samples as well as to extract mineralogical information from the Hyperion image. 2.3. X-ray diffraction analysis The mineralogical interpretation of the reflectance spectra was verified using X-ray diffraction analyses of 15 samples that were representative of the different categories of rocks and soils. The diffraction patterns of the samples were described in terms of three sets of parameters: (1) the position of the diffraction maxima, (2) the peak intensities, and (3) the intensity distribution as a function of diffraction angle. These three pieces of information were used to identify and quantify the mineralogy of the sample by matching with the ICDD (International Centre for Diffraction Data) PDF (Powder Data File) database (Jenkins, 2000). 2.4. Processing and interpretation of the Hyperion satellite imagery The radiometrically corrected Hyperion image that was acquired on 30th June, 2008 was processed as follows:
a) Fixing of bad pixels and bands. Bad pixels and bands are outliers in mean response and/or variance in response in a given image (Mason, 2002). Their values were replaced by interpolation with its nearest left and right specially row neighbours. If the cell had good neighbours on one side only, the replacement was done by extrapolation. b) Gain and offset correction for columns. The correction was done assuming the response mean and standard deviation of each image-column to be the same. c) Atmospheric correction using Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH). The FLAASH module was performed with and without spectral polishing. The output image that was derived with spectral polishing was used after analyzing reflectance spectra. The parameters used for the FLAASH module are summarized in Table 1. d) EFFORT polishing (Empirical Flat Field Optimal Reflectance Transformation) was performed after the atmospheric correction to reduce systematic linear errors in the data by finding a reasonably large set of calibration pixels for which the “true” spectra can be estimated. It calculates a correction gain and/or offset using least-squares regression between the calibration pixels (true) and actual spectra (Mason, 2002), so the spectra appear more like spectra of real materials. 1000 calibration spectra with 14 polynomial orders were used to correct the data. e) Identification of scene spectral end members. The first step of this process was to perform a minimum noise fraction (MNF) transformation to determine the inherent dimensionality of the image data. The MNF transform uses two principal component transformations. While the first transformation decorrelates and rescales the noise in the data, the second is performed for the noisiest data (ITT ENVI, 2007). Based on MNF results, the lower order MNF bands are usually set aside and the higher order bands are selected for further processing. These were used for selection of pure end members. The pixel purity index (PPI) was used to identify the most spectrally pure or extreme pixels in the imagery. An auto cluster algorithm in the n-dimensional visualizer, a software tool to visualize point clouds in multi-dimensional feature space, was then used to find the end member spectra, Mean spectra of clusters were then extracted to act as end members for mapping minerals, and f) Generation of surface compositional map using Mixture Tuned Matched Filtering (MTMF) and Minimum Noise Fraction (MNF) method (Mason, 2002). MTMF algorithm was applied to MNF
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
27
Fig. 8. (A) Sedimentary deposits at sample location P031. (B) Continuum removed reflectance spectrum of sedimentary rock sample P031 (solid line). Dashed lines show the (Si–OH) absorption feature of chert and the CO3 absorption feature of carbonate material in the matrix. (C) Continuum removed reflectance spectra of different silty-clay sedimentary rock samples.
Table 1 FLAASH input parameters applied on Hyperion radiance image. Parameter
Used value/type
Parameter
Used value/type
Acquired date Acquired time Ground elevation Scene centre latitude Scene centre longitude
2008/06/30 07:23:05 0.738 km −1.994429 36.277371
Atmospheric model Atmospheric zone Aerosol model MODTRAN resolution Output
Tropical Rural Rural 15 cm s−1 10,000
data with the help of image end member spectra derived from n-dimensional space.
3.1. Spectral properties
characteristics of evaporites. A sample containing crystallized trona (B012) was used to verify the mineralogy by X-ray diffraction. The VNIR spectrum of the trona sample (Fig. 4, B012) exhibits absorption features at 1.50 m, 1.74 m, 1.94 m, 2.03 m, 2.22 m, and 2.39 m. These are characteristic for trona and are also present in all evaporate samples in Fig. 4.
3.1.1. Spectral properties of evaporites Six evaporite samples, including trona beds and surface evaporites, were analyzed (see Figs. 2 and 3) to determine spectral
3.1.2. Spectral properties of siliceous rocks Siliceous rocks were analyzed and characterized using reflectance spectra, X-ray diffraction patterns and descriptions of
3. Results
Table 2 Error matrix resulting from surface mineral mapping. Ground truth Classification data
Trona
High Magadi bed
Chert
Total
Error of commission (%)
User accuracy (%)
Trona High Magadi bed Chert Total Error of omission (%) Producer accuracy (%)
5 1 0 6 16.6 83.3
0 7 0 7 0 100
0 4 15 19 21.0 78.9
5 12 15 32
0 41.6 0
100 58.3 100
Overall accuracy = 84.3%
Author's personal copy 28
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 9. The MTMF rule Images 13 (A), 16 (B), 4 (C) and false colour composite of these rule images (R: 4, G: 13, B: 16) (D). Green, blue and pink colours represent the different types/stages of evaporates. Red colours represent the chert beds. (E) Spectral plot of selected field sample spectra and image reflectance spectra of different surface materials that were represented by different colours (A001: Crystallized trona, B012: Powdered form of Trona, P006: Chert). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
hand specimens. They were grouped into the following types of chert: (a) Laminated chert (sample P017, Fig. 5A), (b) Laminated green chert (sample P024, Fig. 5B), (c) Pillow chert (Samples S047 and P035, Fig. 5C), (d) Dyke chert (Samples P019 01, Fig. 5D), (e) Green beds (sample P030, Fig. 5E), (f) Quartz chert and (g) High Magadi beds (sample S037, Fig. 5F) All Samples from all seven groups of siliceous rocks have the same type of reflectance spectra (Fig. 6). Each spectrum contains absorption features at 1.42, 1.91, 2.21 and 2.46 m. All samples contain molecular water as well as hydroxyl bonds due to the 1.9 m molecular water absorption feature and 1.4 m hydroxyl ion absorption feature. In addition to these four (main) common absorption features, samples P019 01, S008 and P006 01 showed another two absorption features at 1.16 m and 1.25 m. The causes of these two absorption features were not studied.
3.1.3. Spectral properties of Magadiite and Kenyaite The mineral Magadiite was identified by X-ray diffraction in one of the samples (P032). The reflectance spectrum of this sample (Fig. 7) contains two absorption features other than the hydroxyl (1.40 m) and water (1.90 m) absorption feature in the shortwave infrared region. A narrow absorption feature at 1.46 m and broad and shallow Si–OH absorption feature near 2.22 m. The mineral Kenyaite was also identified in one of the samples (P036) using X-ray diffraction. The reflectance spectrum of this sample also exhibits a broad Si–OH absorption feature around 2.22 m. In addition to that, it shows another two absorption features at 1.15 m and 1.46 m. The absorption feature at 1.46 m is unique for Kenyaite as well as Magadiite and stronger than that of Magadiite. No spectra that
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
29
Fig. 10. (A) Colour composite of Hyperion MNF bands 2, 4 and 6 after masking out the lake area. (B) Continuum removed image reflectance spectra of different surface materials and reflectance spectra of field samples (P004: volcanic tuff, P030: green beds, P006: Chert), (C) Colour composite of MNF bands 1, 3 and 5, after masking out the lake area. (D) Image reflectance spectra of different surface materials and reference spectra (P031: Sedimentary rocks composed of natrocarbonatite volcanic ash, USGS D & USGS G: Dry and Green vegetation spectra from the USGS reflectance library). (E) Colour composite of MNF bands 4, 5 and 6 after masking out the lake area (Note: black small arrows indicate the position of characteristic absorption features of the image spectra related to the selected field spectra. Black arrows with round head show extracted reference image spectra with corresponding locations). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
were collected from other soil and rock samples show this 1.46 m absorption. The absorption feature at 1.15 m is only present in the Kenyaite spectrum (Fig. 7).
3.1.4. Spectral properties of other sedimentary rocks Sedimentary rock, which is composed of chert pebbles and natro-carbonatite volcanic ash (such as sample P031; Fig. 8A), show mixed absorption features within the 2.1–2.5 m wavelength region (Fig. 8B). A wide absorption feature at around 2.22 m is due to Si–OH groups in chert and the absorption at 2.34 m is related to the carbonate in the matrix (Aines and Rossman, 1984). Spectra of other sedimentary rocks, such as tuffaceous materials that were collected from Lake Magadi area show a common absorption feature at 2.30 m indicating that Mg minerals are more common in igneous rock environment. The spectra of rock samples S021 and P003 show another absorption feature at 2.37 m (may be due to Mg–OH bonds) while rock sample P038 shows an absorption feature at 2.34 m which is related to carbonate. The reflectance spectrum of rock sample S030, a silty clay sedimentary rock, shows
only one absorption feature at 2.30 m which is probably related to the Mg-OH bond of the minerals of the rock. 3.2. Spectral analysis and mapping results A false colour composite of MTMF rule Images 4, 13 and16 (Fig. 9) can be used to discriminate the evaporites series and chert beds in the area according to their general spectral shape and characteristic spectral features. Spectra MF01 and MF02 show spectral shape and features similar to the trona (e.g., sample A001) and evaporates (e.g., sample B012). Spectrum MF03 shows a broad 2.2 m feature related to the presence of chert (e.g., sample P006). Based on their spectral characteristics, the different colours in the false colour composite image can be interpreted as follows: Green and blue represent the different types or stages of evaporites of the area, red colours represent the chert beds and magenta colours depict the combination of volcanic & siliceous rock types (chert series) on the surface. The area covered by the lake shows high spectral variation and this attenuates the spectral dimensionality of the surrounding area. As a result of this, MNF eigen images do not show clear features
Author's personal copy 30
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 11. Intergrated results of the lake Magadi area. (A) Published geology map overlain on hill shaded ASTER DEM, (B) colour composite of Hyperion MTMF bands 2, 6 and 10, (C) colour composite of Hyperion MNF bands 2,4 and 6 (note: interpretation is overlaid as a black line for reference), and (D) geologic interpretation from maps B and C (white: Evaporite series, dark red: high Magadi beds, red: chert series, green: vegetation, grey: volcanic tuff, black: unclassified). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
in the surrounding area. Therefore the Lake area was masked out of the entire image and the remaining area was subjected to the MNF transformation to see more discriminative land cover variation based on the inherent dimensionality of the data. Fig. 10 shows the different MNF image colour composites including MNF band combinations (R:G:B) 2:4:6, 1:3:5, and 4:5:6 with image spectra for different surface covers that have different spectral characteristics. Mean spectra of spectrally contrasting regions in these images were calculated (spectra IS01–IS06 in Fig. 10) from a minimum of 10 pixels each. Image end member spectrum IS01 shows a broad absorption feature near 2.2 m which is similar to that in spectra of the quartzchert series rocks in the area (spectrum P006 in Fig. 10B). Spectrum IS02 shows absorption features in the 2000 nm to 2355 nm wavelength region, which are similar to the absorption features of silica-rich material, containing Si–OH bonds, referred to as “High Magadi bed” in this study (spectrum P030 in Fig. 10B. Spectrum IS03 contains a broad absorption feature at 2.3 m, which is similar to the absorption feature of volcanic tuff of the area (spectrum P004 in Fig. 10B) Spectrum IS04 shows a broad absorption feature between 2.2 m and 2.3 m due to mixture of silica rich materials and volcanic tuffs (spectrum P031 in Fig. 10D). The area represented by spectrum IS05 contains abundant green vegetation which is apparent from the shape of the reflectance spectra that are typical for green vegetation in the 0.4–1.5 m wavelength range (spectrum USGS G in Fig. 10D). Spectrum IS06 shows absorption wings in the 0.4–0.9 m wavelength regions which is characteristic of dry plant materials (spectrum USGS D in Fig. 10D). Absorption wings in this position are generated by intense blue and ultraviolet absorptions in com-
pounds such as lignin, tannins and pectins (Elvidge, 1990). The USGS vegetation library was used to compare green (USGS G) and dry (USGS D) vegetation with image spectra. Spectral end members of the five surface materials (evaporite series, High Magadi beds, chert series, dense vegetation and volcanic tuff) were used to classify selected MTMF and MNF bands using the Maximum Likelihood method. The result of the classification is shown in Fig. 11D. The minerals Magadiite and Kenyaite could not be identified in the imagery due to their limited exposure at the surface and their shallow diagnostic absorption features compared to the noise of the Hyperion image spectra. Kenyaite always occurs as nodules and never forms as a continuous bed. Although Magadiite precipitates as a bed due to its small diagnostic absorption feature, it is difficult to map even from hyperspectral images. 3.3. Accuracy assessment Classifications for all pixels, which were superimposed on the 32 point locations of the ground data, were extracted from the classified raster surface mineral map (Fig. 11D). The corresponding error matrix is shown in Table 2. The overall accuracy of the classification is 84.3%. The producer’s accuracies of trona, High Magadi bed and the chert were reasonably high, and therefore satisfying the primary purpose of this research, which is the mapping of spatial distribution of the evaporites and precipitated minerals in a saline lake environment. A more careful inspection of the error matrix shows that there is a significant confusion between the “High Magadi bed” and “chert” classes. This confusion is acceptable due to two main reasons:
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
◦ Though chert is not always observed in the field with large spatial extent, it is always found with the silica-rich High Magadi bed (spatial association). ◦ The reflectance spectra of chert and High Magadi bed are relatively similar due to presence of SiO2 as the main constituent; hence it is difficult to differentiate them using remote sensing methods (spectral resemblance). 4. Discussion This study demonstrated the possibility of mapping evaporite minerals and associated sediments in Lake Magadi area using space-borne hyperspectral Hyperion data. Spatial distribution of chert and High Magadi beds in the mineral map is superimposed with the published geology map of the area (Fig. 11). Volcanic tuff, which is not recorded in published geological maps known to the authors, can be seen in southern part of the study area in derived mineral map, close to the Lenderut volcano (Fig. 1B). Alluvium terraces in the northern part of Lake Magadi study area, however, were not identified and categorized using remote sensing due to their inhomogeneity of surface minerals (materials). Some of the areas (e.g., contact zones between areas) were left unclassified to facilitate mapping of only the homogeneous surface materials. Even though most of the precipitates and evaporites show characteristic spectral features reflecting the mineralogy of the material, several of them could not be identified because of lack of spectral information related to the mineralogy of some of the evaporites. While the fine-grained silica rich sedimentary rock, chert, exhibits a broad absorption feature at 2.2 m, other siliceous materials such as green beds, diatomite and High Magadi beds exhibit a combination of a broad absorption feature at 2.2 m and a narrow feature at 2.3 m. The main evaporite mineral in the study area, trona, commonly exhibits 6 absorption features at 1.50-, 1.74, 1.94-, 2.03-, 2.22- and 2.39-m. Not only the position and shape of the absorption feature, but also the general shape of the reflectance spectra can be used to identify mineral precipitates and evaporites in the study area. Chert, diatomite, High Magadi beds, green beds and trona were identified using reflectance spectra. Chert was mapped from the Hyperion image based on its 2.2 m broad absorption feature in SWIR region. Diatomite, green beds and High Magadi beds show a relatively similar general spectral shape and absorption features. In addition, diatomite and green bed did not show large spatial extents and identification in the field was solely based on small spots within the High Magadi beds. Diatomite was identified only in one location covering an approximately 2 m × 8 m area. Therefore diatomite, green beds and High Magadi beds were mapped as one unit and referred to as the High Magadi bed. 5. Conclusions Not all the ground mineralogy that was identified by reflectance spectra could be mapped. This is due to factors such as the spectral resolution of the image, the spectral resolution of diagnostic absorption features of the field reflectance spectra, noise in the image, spatial extent of the ground target, spatial resolution of the image, and accuracy of image prepossessing stages. Furthermore, geologic surfaces are often partially covered with non-geologic materials or composed of mixtures of minerals with varying grain sizes and differing degree of compaction and weathering. These factors greatly influence the remote spectral measurements and limit the number of pixels that can be classified and mapped. The overall shape of the laboratory/library spectra was different from the image spectra, probably due to the presence of
31
variable mineral mixtures, grain size variations, residual atmospheric absorption features, desert varnish top of the rock surface, and calibration error of laboratory spectrometer and/or Hyperion instrument. These spectral differences point to the advantage of using image spectra rather than library/laboratory as reference in MTMF processing. The area that had very high spectral variation attenuated the spectral dimensionality of the surrounding low spectral variation area of the Hyperion image. Image processing in two steps, with and without (masking out) the high spectral variation area, could successfully overcome this problem, allowing extraction of information from the entire image. Acknowledgements The authors would like to thank NASA for making the Hyperion image of the area freely available for research, and Boudewijn de Smeth, Chris Hecker for their assistance during the paper preparation. We highly appreciate technical support from the World Agroforestry Centre (ICRAF), in particular from Mr. Elvis Weullow, and from the Department of Geology, University of Nairobi, Kenya. References Aines, R.D., Rossman, G.R., 1984. Water in minerals? A peak in the infrared. Journal of Geophysical Research 89 (B6), 4059–4071. Baker, B.H., 1958. Geology of the Magadi area. Report Geologycal survey of Kenya 42. Biggar, S.F., Thome, K.J., Wisniewski, W., 2003. Vicarious radiometric calibration of EO-1 sensors by reference to high-reflectance ground targets. IEEE Transactions on Geoscience and Remote Sensing 41 (6), 1174–1179. Clarck, R.N., King, T.V.V., Klejwa, M., Swayze, G.A., 1990. High spectral resolution reflectance spectroscopy of minerals. Journal of Geophysical Research 95 (B8), 12,653–12,680. Crowley, J.K., 1993. Mapping playa evaporite minerals with AVIRIS data: a first report from Death Valley, California. Remote Sensing of Environment 44, 337–356. Crowley, J.K., Hook, S.J., 1996. Mapping playa evaporite minerals and associated sediments in Death Valley, California, with multispectral thermal infrared images. Journal of Geophysical Research 101 (B1), 643–660. Elvidge, C.D., 1990. Visible and near infrared reflectance characteristics of dry plant. International Journal of Remote Sensing 11 (10), 1775–1795. Eugster, H.P., 1969. Inorganic bedded cherts from the Magadi area. Kenya: Contributions Mineralogy Petrology 22, 1–31. Eugster, H.P., Jones, B.F., 1968. Gels composed of sodium–aluminium silicate, Lake Magadi Kenya. Science 161, 160–163. Gaffey, S.J., 1985. Reflectance spectroscopy in the visible and near-infrared (0.35–2.55 pm): applications in carbonate petrology. Geology 13, 270–273. Goetz, A.F.H., Vane, G., Solomon, J.E., Rock, B.N., 1985. Imaging spectrometry for earth remote sensing. Science 228 (4704), 1147–1153. Hewson, R.D., Cudahy, T.J., Mizuhiko, S., Ueda, K., Mauger, A.J., 2005. Seamless geological map generation using ASTER in the Broken Hill-Curnamona province of Australia. Remote Sensing of Environment 99 (1–2), 159–172. Hubbard, B.E., Crowley, J.K., Zimbelman, D.R., 2003. Comparative alteration mineral mapping using visible to shortwave infrared (0.4–2.4 m) hyperion, ALI, and ASTER imagery. IEEE Transactions on Geoscience and Remote Sensing 41 (6), 1401–1410. I.T.T.ENVI, 2007. ENVI User’s Guide, v. 4.4. Jenkins, R., 2000. X-ray Techniques: Overview. Encyclopedia of Analytical Chemistry. John Wiley & Sons Ltd. Jones, B.F., Eugster, H.P., Rettig, S.L., 1977. Hydrochemistry of the Lake Magadi basin, Kenya. Geochimica et Cosmochimica Acta 41, 53–72. King, M.D., et al. (Eds.), 2003. EOS Data Products Hand Book, vol. 1. NASA/Goddard Space Flight Center. Kruse, F.A., 1988. Use of airborne imaging spectrometer data to map minerals associated with hydrothermally altered rocks in the northern Grapevine Mountains, Nevada, and California. Remote Sensing of Environment 24 (1), 31–51. Kruse, F.A., 2003. Preliminary Results- Hyperspectral Mapping of Coral Reef Systems Using EO-1 Hyperion. JPL Publication 04-6, Buck Island, U.S. Virgin Islands, pp. 157–173. Mason, P., 2002. MMTG A-list hyperspectral data processing software. In: Manual Version 1.0. CSIRO Division of Exploration and Mining, p. 103. Pearlman, J.S., et al., 2003. Hyperion, a space-based imaging spectrometer. IEEE Transactions on Geoscience and Remote Sensing 41 (6), 1160–1173. Salisbury, J.W., D’Aria, D.M., 1992. Emissivity of terrestial materials in the 8–14 m atmospheric window. Remote Sensing of Environment 42, 83–106. Surdam, R.C., Eugster, H.P., 1976. Mineral reactions in the sedimentary deposits of the Lake Magadi region, Kenya. Geological Society of America Bulletin 87, 1739–1752.
Author's personal copy 32
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Turdu, C.L., et al., 1999. Influence of preexisting oblique discontinuities on the geometry and evolution of extensional fault patterns: evidence from the Kenya rift using SPOT imagery. Geoscience of Rift Systems-Evolution of East Africa: AAPG Studies in Geology 44, 173– 191. Van der Meer, F., 2001. Spectral matching using pixel cross-correlograms for the analysis of LANDSAT TM data. International Journal of Applied Earth Observation and Geoinformation 3 (2), 197–202.
Van der Meer, F., 2006. The effectiveness of spectral similarity measures for the analysis of hyperspectral imagery. International Journal of Applied Earth Observation and Geoinformation 8, 3–17. Vaughan, R.G., Hook, S.J., Calvin, W.M., Taranik, J.V., 2005. Surface mineral mapping at Steamboat Springs, Nevada, USA, with multi-wavelength thermal infrared images. Remote Sensing of Environment 99 (1–2), 140–158. Warren, J.K., 2006. Evaporites: Sediments, Resources and Hydrocarbons. Springer, p. 1032.
This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright
Author's personal copy International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Contents lists available at SciVerse ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Hyperspectral remote sensing of evaporate minerals and associated sediments in Lake Magadi area, Kenya Gayantha R.L. Kodikara a,∗ , Tsehaie Woldai b , Frank J.A. van Ruitenbeek b , Zack Kuria c , Freek van der Meer b , Keith D. Shepherd d , G.J. van Hummel e a
Department of Oceanography and Marine Geology, Faculty of Fisheries & Marine Science and Technology, University of Ruhuna, Sri Lanka University of Twente, Faculty of Geo-information Science and Earth Observation (ITC), Hengelosesestraat 99, P.O. Box 6, 7500AA Enschede, The Netherlands c Department of Geology, University of Nairobi, Nairobi, Kenya d World Agroforestry Centre (ICRAF), P.O. Box 30677-00100, Nairobi, Kenya e Institute for Nanotechnology (MESA+), University of Twente, The Netherlands b
a r t i c l e
i n f o
Article history: Received 1 September 2010 Accepted 18 August 2011 Keywords: Hyperion Lake Magadi MTMF Surface mineral mapping Magadiite Kenyaite
a b s t r a c t Pleistocene to present evaporitic lacustrine sediments in Lake Magadi, East African Rift Valley, Kenya were studied and mapped using spectral remote sensing methods. This approach incorporated surface mineral mapping using space-borne hyperspectral Hyperion imagery together with laboratory analysis, including visible, near-infrared diffuse reflectance spectroscopy (VNIR) measurements and X-ray diffraction for selected rock and soil samples of the study area. The spectral signatures of Magadiite and Kenyaite, which have not been previously reported, were established and the spectral signatures of trona, chert series, volcanic tuff and the High Magadi bed were also analyzed. Image processing techniques, MNF (Minimum Noise Fraction) and MTMF (Mixture Tuned Matched Filtering) using a stratified approach (image analysis with and without the lake area), were used to enhance the mapping of evaporates. High Magadi beds, chert series and volcanic tuff were identified from the Hyperion image with an overall mapping accuracy of 84.3%. Even though, the spatial distribution of evaporites and sediments in Lake Magadi area change in response to climate variations, the mineralogy of this area has not been mapped recently. The results of this study shows the usefulness of the hypersspectral remote sensing to map the surface geology of this kind of environment and to locate promising sites for industrial open-pit trona mining in a qualitative and quantitative manner. © 2011 Elsevier B.V. All rights reserved.
1. Introduction Lake Magadi, a dry alkaline saline lake, located in the East African Rift Valley in Kenya (Fig. 1A) contains the most concentrated brines found in the African Rift Valley (Warren, 2006). It contains a large amount of trona (Na2 CO3 ·NaHCO3 ·2H2 O), chert (SiO2 ), Magadiite (NaSi7 O13 (OH)3 ·3H2 O) and Kenyaite (NaSi11 O20.5 (OH)4 ·3H2 O). Little Magadi, or Lake Nasikie Engeda, lies 1.6 km north of Lake Magadi from which it is separated by a narrow horst (Fig. 1C). Little Magadi is perennially saline, while Lake Magadi is intermittently dry. Hydrothermal circulation associated with the numerous active alkaline volcanoes and their feeder faults, supply hot alkaline brines to the many hot springs situated along the edge of the lakes and this system is responsible for the ongoing geochemical evolution of the area (Warren, 2006). The Pleistocene and Holocene history of the Magadi basin sediments and their geochemical evolution is fully described in the work of Baker (1958) and Eugster (1969). The
∗ Corresponding author. Tel.: +94 415674048. E-mail address: gayantha [email protected] (G.R.L. Kodikara). 0303-2434/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.jag.2011.08.009
recent geologic history is better understood by studying the spatial distributions of minerals in the area. The interpretation and reconstruction of this geochemical evolution, however, is difficult to establish due to variation in mineralogy (e.g., not easily recognizable in the field) and the large spatial extent that limits accessibility (Warren, 2006). Alternatively, access to relatively inexpensive satellite-borne multi-spectral and hyper-spectral data have created new opportunities for the regional mapping of mineralogy, geological structures and rock types including alteration products (Hewson et al., 2005; Vaughan et al., 2005). Such datasets provide identification of different surface expressions and mapping possibilities for minerals in the hydroxyl, silicate, sulfate, carbonate and iron oxide groups covering large areas and in inaccessible terrains. It has been applied in efflorescent salt crusts in Death Valley, California using airborne hyperspectral and space-borne multispectral imagery to map the different surface saline materials (Crowley, 1993; Crowley and Hook, 1996). The motivation of this research was to evaluate the identification and mapping capability of various evaporites and precipitates in Lake Magadi area using space-borne hyperspectral Hyperion data
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
23
Fig. 1. Location of the Kenya Rift in the East African Rift System (A) and structural sketch of the Kenya rift and location of the Magadi Lake (B) (modified after Turdu et al. (1999)). (C) Geological Map of Lake Magadi (modified after Baker (1958) and Eugster and Jones (1968)).
and their spectral characteristics. In addition to that, the resultant map will be useful for the mining industry to show areas that are favourable for the formation of lacustrine mineral deposits and for planning access for trona mining. 1.1. Geologic setting Lake Magadi (Fig. 1) is located in the deepest depression of the Kenyan Rift in Kenya, surrounded by hills formed by Pleistocene alkali trachyte flows. The valleys of the Lake Magadi and Little Magadi Lake are covered by a series of Lake Beds dating from the middle Pleistocene to present. They consist of vast deposits of trona, bedded cherts and High Magadi beds derived from Magadiite horizons (Fig. 1C). Mineral precipitation, re-solution, re-precipitation and reactions, which are common in this environment, can be understood based on the hydrology and aqueous geochemistry of the area. Atmospheric precipitation, overland flow and percolation of water can react with alkali trachyte terrains and this process can be illustrated by the reaction of albite feldspar with water and aqueous CO2 (Jones et al., 1977): NaAlSi3 O8(s) + 2H2 O(aq) + CO2(aq) = Na+ (aq) + HCO− + 3SiO2(aq) + Ai(OH)3(aq) 3(aq)
After accumulation of these solutes in the lake from rainfall and hydrolysis, the waters are subjected to evaporation, either at the surface or by capillarity processes. This chemistry is ideal for ongoing trona precipitation at higher salinities (Warren, 2006). Trona is the first Na-carbonate mineral to precipitate in equilibrium with concentrating lake brines. + HCO3− + CO2− 2H2 O → Na2 CO3− NaHCO3− 2H2 O(s) 3Na+ (aq) 3(aq) (aq) (Trona)
The composition of the Magadi-trona is straightforward. It consists of trona crystals (sodium sesquicarbonate – Na2 CO3 ·NaHCO3 ·2H2 O), sodium fluoride (NaF), and rarely small amounts of common salts (NaCl) (Baker, 1958). These trona deposits extend for approximately 47 km2 in area with 7–50 m thickness. The Trona is locally known as the “Evaporite Series” which is made up of cm-scale stacked, upward-pointing and growth-aligned trona crystals (Warren, 2006). Precipitation of chert is not straightforward. It can occur through two types of hydrous sodium silicate intermediate products called Magadiite (NaSi7 O13 (OH)3 ·3H2 O) and Kenyaite (NaSi11 O20.5 (OH)4 ·3H2 O). Magadiite occurs in substantial quantities, typically as beds, mounds and nodules. Kenyaite only occurs as nodules in the bedded Magadiite in the High Magadi bed (Warren,
Author's personal copy 24
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 2. Crystallized trona (A001) and its powdered form (B002). One-Euro coin is used for scale.
2006). The precipitation of Magadiite is represented by the following reaction (Eugster, 1969). = NaSi7 O13 (OH)3 .3H2 O(s) + 9H2 O(aq) + H+ 7H4 SiO4(aq) + Na+ (aq) (aq) (Magadiite)
The conversion of Magadiite to chert (SiO2 ) through dehydration and sodium loss may involve Kenyaite (NaSi11 O20.5 (OH)4 ·3H2 O) as an intermediate phase, 22NaSi7 O13 (OH)3− 3H2 O(s) +8H+ → 14NaSi11 O20.5 (OH)4− 3H2 O(s) +8Na+ +33H2 O (aq) (aq) (Magadiite)
(Kenyaite)
NaSi11 O20.5 (OH)4 .3H2 O(s) + H+ → 11SiO2(s) + Na+ + 5 1/2H2 O (aq) (aq) (Kenyaite)
(Chert)
or it may directly produce a silica phase (Eugster, 1969). → 7SiO2(s) + Na+ + 5H2 O NaSi7 O13 (OH)3 .3H2 O(s) + H+ (aq) (aq) (Magadiite)
(chert)
Outcrops of High Magadi beds are located in the alluvial flats and at the foot of the hills surrounding Lake Magadi. The mineralogy of High Magadi is characterized by detrital silicate, saline minerals, calcite, sodium silicates, including Magadiite, Kenyaite, chert and zeolites (Surdam and Eugster, 1976).
Fig. 4. Reflectance spectra of Trona samples. Arrows indicate absorption features of trona.
1.2. Mapping minerals with hyperspectral remote sensing
Fig. 3. Surface evaporates at sample location of P018.
Reflectance spectroscopy offers a rapid, inexpensive and nondestructive tool for determining the mineralogy of rock and soil samples (Gaffey, 1985; Salisbury and D’Aria, 1992; Van der Meer, 2001). In the visible and shortwave infra-red wavelength region (0.4–2.5 m), iron-, hydroxyl-, sulfate-, water and carbonate bearing minerals display diagnostics spectral features (Clarck et al., 1990; Van der Meer, 2001). The absorption features (position, shape, depth and width) are controlled by the particulate crystal structure in which the absorbing species is contained and by the chemical structure of the material (Van der Meer, 2001). Therefore, absorption features are used to identify the type of mineral in reflectance spectroscopy and Imagine spectrometry is used to map those minerals (Kruse, 1988).
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
25
Fig. 5. (A) Crystal casts in chert plates from Southern end of Lake Magadi. (B) Laminated green chert from northern part of the Lake Magadi. (C) Pillow chert near to Magadi town. (D) Chert dykes from North-eastern part of Lake Magadi. (E). Green beds from southern part of Lake Magadi. (F). High Magadi Beds from southern part of Lake Magadi.
Imaging spectrometry which is defined as “the simultaneous acquisition of images in many narrow, contiguous spectral bands” (Goetz et al., 1985) allows extraction of a detailed spectrum for each picture element (pixel) of the image. High spectral resolution reflectance spectra collected by imaging spectrometers allow direct identification of individual materials based upon their reflectance characteristics such as absorption band depth, band position, asymmetry of the absorption features, etc. Mapping of surface reflectance in an imaging spectrometer data set with known diagnostic reflectance spectra of minerals is basically done using spectral matching techniques or sub-pixel classification techniques (Van der Meer, 2006). Two groups of spectral matching techniques have been developed: deterministic empirical measures, including spectral angle mapper (SAM), spectral correlation measures (SCM), Euclidean distance measures (ED), and Mixture Tuned Matched Filtering (MTMF) method (Mason, 2002); and Stochastic measures evaluating the statistical distributions of spectral reflectance values of target end members (Van der Meer, 2006). In this paper the Mixture Tuned Matched Filtering (MTMF) method was used to map surface mineral of the study area. The main advantage of the MTMF method, unlike traditional spectra mixture modeling, is the ability to map a single known target without knowing the other background end member signatures (Kruse, 2003). 1.3. Hyperion satellite sensor imagery Hyperion is the first Earth-orbiting imaging spectrometer, which was launched onboard the Earth Observing 1 (EO-1) satellite on November 21, 2000. Operating across the full solar-reflected spectrum, the nominal spectral coverage ranges from 0.4 m to 2.5 m, with 10 nm sampling and spectral response functions (King et al., 2003). The radiometric range of Hyperion spans from zero to the maximum Lambertian reflected radiance with 12 bits of digitization (King et al., 2003). The Hyperion push broom instrument captures the image frame spectra from an area of 30 m along-track by 7.7 km cross-track (Pearlman et al., 2003). Hyperion has one telescope, slit and two grating spectrometers: one spectrometer with 70 channels in the VNIR wavelength region (0.4–1.0 m) with
silicon detector array and one spectrometer with 172 channels in SWIR wavelength region (0.9–2.5 m) with an HgCdTe detector array (Biggar et al., 2003). The 196 unique channels covering VNIR (band 8–57) and SWIR (band 77–224) regions are calibrated from 242 total channels (Hubbard et al., 2003). 2. Methods The identification and mapping of evaporate minerals and associated sediments in the Lake Magadi area, involved the collection of rocks and soils in the field, their analyses using reflectance spectroscopy and X-ray diffraction, and the processing and interpretation of Hyperion hyperspectral satellite imagery. 2.1. Field rock/ soil sampling Fieldwork was conducted in the Lake Magadi area in July and August 2008. 84 rock and soil samples, representative for the entire study area, were collected from relatively homogeneous bare land areas having spatial extent larger than 60 m × 60 m. In addition, several other samples were collected without considering the homogeneity and the spatial extent of the sample location due to the geological interest of the area. Each sample location was also subjected to land cover and geomorphologic analysis, GPS data collection and field photographs. 2.2. Spectral reflectance analysis The soil and rock samples were dried in the open air. Once dried, all soil samples were crushed and passed through a 2 mm sieve. Reflectance spectra of the 26 soil samples and 58 rock samples were acquired using Multi Purpose Analyzer – Fourier Transformed Spectrometer (MPA – FTIR) and Vertex-70 FTIR instrument spectrometer fitted with a diffuse reflectance accessory over the wavelength range from 350 to 2500 nm. All the rock and soil samples were grouped based on their spectral signatures and the physical properties/appearance of the samples. The position, shape, depth and width of absorption features are used to identify the
Author's personal copy 26
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 7. Reflectance spectra of Kenyaite and Magadiite rock samples.
Fig. 6. Reflectance spectra of siliceous rocks (P006 1 and S032: chert with small cavities filled by carbonate minerals, P019 1: Dyke chert, S008: Quartz chert, P035 and S047: Pillow chert, P017: Laminated chert, P024: Laminated green chert).
mineralogy from field samples as well as to extract mineralogical information from the Hyperion image. 2.3. X-ray diffraction analysis The mineralogical interpretation of the reflectance spectra was verified using X-ray diffraction analyses of 15 samples that were representative of the different categories of rocks and soils. The diffraction patterns of the samples were described in terms of three sets of parameters: (1) the position of the diffraction maxima, (2) the peak intensities, and (3) the intensity distribution as a function of diffraction angle. These three pieces of information were used to identify and quantify the mineralogy of the sample by matching with the ICDD (International Centre for Diffraction Data) PDF (Powder Data File) database (Jenkins, 2000). 2.4. Processing and interpretation of the Hyperion satellite imagery The radiometrically corrected Hyperion image that was acquired on 30th June, 2008 was processed as follows:
a) Fixing of bad pixels and bands. Bad pixels and bands are outliers in mean response and/or variance in response in a given image (Mason, 2002). Their values were replaced by interpolation with its nearest left and right specially row neighbours. If the cell had good neighbours on one side only, the replacement was done by extrapolation. b) Gain and offset correction for columns. The correction was done assuming the response mean and standard deviation of each image-column to be the same. c) Atmospheric correction using Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH). The FLAASH module was performed with and without spectral polishing. The output image that was derived with spectral polishing was used after analyzing reflectance spectra. The parameters used for the FLAASH module are summarized in Table 1. d) EFFORT polishing (Empirical Flat Field Optimal Reflectance Transformation) was performed after the atmospheric correction to reduce systematic linear errors in the data by finding a reasonably large set of calibration pixels for which the “true” spectra can be estimated. It calculates a correction gain and/or offset using least-squares regression between the calibration pixels (true) and actual spectra (Mason, 2002), so the spectra appear more like spectra of real materials. 1000 calibration spectra with 14 polynomial orders were used to correct the data. e) Identification of scene spectral end members. The first step of this process was to perform a minimum noise fraction (MNF) transformation to determine the inherent dimensionality of the image data. The MNF transform uses two principal component transformations. While the first transformation decorrelates and rescales the noise in the data, the second is performed for the noisiest data (ITT ENVI, 2007). Based on MNF results, the lower order MNF bands are usually set aside and the higher order bands are selected for further processing. These were used for selection of pure end members. The pixel purity index (PPI) was used to identify the most spectrally pure or extreme pixels in the imagery. An auto cluster algorithm in the n-dimensional visualizer, a software tool to visualize point clouds in multi-dimensional feature space, was then used to find the end member spectra, Mean spectra of clusters were then extracted to act as end members for mapping minerals, and f) Generation of surface compositional map using Mixture Tuned Matched Filtering (MTMF) and Minimum Noise Fraction (MNF) method (Mason, 2002). MTMF algorithm was applied to MNF
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
27
Fig. 8. (A) Sedimentary deposits at sample location P031. (B) Continuum removed reflectance spectrum of sedimentary rock sample P031 (solid line). Dashed lines show the (Si–OH) absorption feature of chert and the CO3 absorption feature of carbonate material in the matrix. (C) Continuum removed reflectance spectra of different silty-clay sedimentary rock samples.
Table 1 FLAASH input parameters applied on Hyperion radiance image. Parameter
Used value/type
Parameter
Used value/type
Acquired date Acquired time Ground elevation Scene centre latitude Scene centre longitude
2008/06/30 07:23:05 0.738 km −1.994429 36.277371
Atmospheric model Atmospheric zone Aerosol model MODTRAN resolution Output
Tropical Rural Rural 15 cm s−1 10,000
data with the help of image end member spectra derived from n-dimensional space.
3.1. Spectral properties
characteristics of evaporites. A sample containing crystallized trona (B012) was used to verify the mineralogy by X-ray diffraction. The VNIR spectrum of the trona sample (Fig. 4, B012) exhibits absorption features at 1.50 m, 1.74 m, 1.94 m, 2.03 m, 2.22 m, and 2.39 m. These are characteristic for trona and are also present in all evaporate samples in Fig. 4.
3.1.1. Spectral properties of evaporites Six evaporite samples, including trona beds and surface evaporites, were analyzed (see Figs. 2 and 3) to determine spectral
3.1.2. Spectral properties of siliceous rocks Siliceous rocks were analyzed and characterized using reflectance spectra, X-ray diffraction patterns and descriptions of
3. Results
Table 2 Error matrix resulting from surface mineral mapping. Ground truth Classification data
Trona
High Magadi bed
Chert
Total
Error of commission (%)
User accuracy (%)
Trona High Magadi bed Chert Total Error of omission (%) Producer accuracy (%)
5 1 0 6 16.6 83.3
0 7 0 7 0 100
0 4 15 19 21.0 78.9
5 12 15 32
0 41.6 0
100 58.3 100
Overall accuracy = 84.3%
Author's personal copy 28
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 9. The MTMF rule Images 13 (A), 16 (B), 4 (C) and false colour composite of these rule images (R: 4, G: 13, B: 16) (D). Green, blue and pink colours represent the different types/stages of evaporates. Red colours represent the chert beds. (E) Spectral plot of selected field sample spectra and image reflectance spectra of different surface materials that were represented by different colours (A001: Crystallized trona, B012: Powdered form of Trona, P006: Chert). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
hand specimens. They were grouped into the following types of chert: (a) Laminated chert (sample P017, Fig. 5A), (b) Laminated green chert (sample P024, Fig. 5B), (c) Pillow chert (Samples S047 and P035, Fig. 5C), (d) Dyke chert (Samples P019 01, Fig. 5D), (e) Green beds (sample P030, Fig. 5E), (f) Quartz chert and (g) High Magadi beds (sample S037, Fig. 5F) All Samples from all seven groups of siliceous rocks have the same type of reflectance spectra (Fig. 6). Each spectrum contains absorption features at 1.42, 1.91, 2.21 and 2.46 m. All samples contain molecular water as well as hydroxyl bonds due to the 1.9 m molecular water absorption feature and 1.4 m hydroxyl ion absorption feature. In addition to these four (main) common absorption features, samples P019 01, S008 and P006 01 showed another two absorption features at 1.16 m and 1.25 m. The causes of these two absorption features were not studied.
3.1.3. Spectral properties of Magadiite and Kenyaite The mineral Magadiite was identified by X-ray diffraction in one of the samples (P032). The reflectance spectrum of this sample (Fig. 7) contains two absorption features other than the hydroxyl (1.40 m) and water (1.90 m) absorption feature in the shortwave infrared region. A narrow absorption feature at 1.46 m and broad and shallow Si–OH absorption feature near 2.22 m. The mineral Kenyaite was also identified in one of the samples (P036) using X-ray diffraction. The reflectance spectrum of this sample also exhibits a broad Si–OH absorption feature around 2.22 m. In addition to that, it shows another two absorption features at 1.15 m and 1.46 m. The absorption feature at 1.46 m is unique for Kenyaite as well as Magadiite and stronger than that of Magadiite. No spectra that
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
29
Fig. 10. (A) Colour composite of Hyperion MNF bands 2, 4 and 6 after masking out the lake area. (B) Continuum removed image reflectance spectra of different surface materials and reflectance spectra of field samples (P004: volcanic tuff, P030: green beds, P006: Chert), (C) Colour composite of MNF bands 1, 3 and 5, after masking out the lake area. (D) Image reflectance spectra of different surface materials and reference spectra (P031: Sedimentary rocks composed of natrocarbonatite volcanic ash, USGS D & USGS G: Dry and Green vegetation spectra from the USGS reflectance library). (E) Colour composite of MNF bands 4, 5 and 6 after masking out the lake area (Note: black small arrows indicate the position of characteristic absorption features of the image spectra related to the selected field spectra. Black arrows with round head show extracted reference image spectra with corresponding locations). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
were collected from other soil and rock samples show this 1.46 m absorption. The absorption feature at 1.15 m is only present in the Kenyaite spectrum (Fig. 7).
3.1.4. Spectral properties of other sedimentary rocks Sedimentary rock, which is composed of chert pebbles and natro-carbonatite volcanic ash (such as sample P031; Fig. 8A), show mixed absorption features within the 2.1–2.5 m wavelength region (Fig. 8B). A wide absorption feature at around 2.22 m is due to Si–OH groups in chert and the absorption at 2.34 m is related to the carbonate in the matrix (Aines and Rossman, 1984). Spectra of other sedimentary rocks, such as tuffaceous materials that were collected from Lake Magadi area show a common absorption feature at 2.30 m indicating that Mg minerals are more common in igneous rock environment. The spectra of rock samples S021 and P003 show another absorption feature at 2.37 m (may be due to Mg–OH bonds) while rock sample P038 shows an absorption feature at 2.34 m which is related to carbonate. The reflectance spectrum of rock sample S030, a silty clay sedimentary rock, shows
only one absorption feature at 2.30 m which is probably related to the Mg-OH bond of the minerals of the rock. 3.2. Spectral analysis and mapping results A false colour composite of MTMF rule Images 4, 13 and16 (Fig. 9) can be used to discriminate the evaporites series and chert beds in the area according to their general spectral shape and characteristic spectral features. Spectra MF01 and MF02 show spectral shape and features similar to the trona (e.g., sample A001) and evaporates (e.g., sample B012). Spectrum MF03 shows a broad 2.2 m feature related to the presence of chert (e.g., sample P006). Based on their spectral characteristics, the different colours in the false colour composite image can be interpreted as follows: Green and blue represent the different types or stages of evaporites of the area, red colours represent the chert beds and magenta colours depict the combination of volcanic & siliceous rock types (chert series) on the surface. The area covered by the lake shows high spectral variation and this attenuates the spectral dimensionality of the surrounding area. As a result of this, MNF eigen images do not show clear features
Author's personal copy 30
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Fig. 11. Intergrated results of the lake Magadi area. (A) Published geology map overlain on hill shaded ASTER DEM, (B) colour composite of Hyperion MTMF bands 2, 6 and 10, (C) colour composite of Hyperion MNF bands 2,4 and 6 (note: interpretation is overlaid as a black line for reference), and (D) geologic interpretation from maps B and C (white: Evaporite series, dark red: high Magadi beds, red: chert series, green: vegetation, grey: volcanic tuff, black: unclassified). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
in the surrounding area. Therefore the Lake area was masked out of the entire image and the remaining area was subjected to the MNF transformation to see more discriminative land cover variation based on the inherent dimensionality of the data. Fig. 10 shows the different MNF image colour composites including MNF band combinations (R:G:B) 2:4:6, 1:3:5, and 4:5:6 with image spectra for different surface covers that have different spectral characteristics. Mean spectra of spectrally contrasting regions in these images were calculated (spectra IS01–IS06 in Fig. 10) from a minimum of 10 pixels each. Image end member spectrum IS01 shows a broad absorption feature near 2.2 m which is similar to that in spectra of the quartzchert series rocks in the area (spectrum P006 in Fig. 10B). Spectrum IS02 shows absorption features in the 2000 nm to 2355 nm wavelength region, which are similar to the absorption features of silica-rich material, containing Si–OH bonds, referred to as “High Magadi bed” in this study (spectrum P030 in Fig. 10B. Spectrum IS03 contains a broad absorption feature at 2.3 m, which is similar to the absorption feature of volcanic tuff of the area (spectrum P004 in Fig. 10B) Spectrum IS04 shows a broad absorption feature between 2.2 m and 2.3 m due to mixture of silica rich materials and volcanic tuffs (spectrum P031 in Fig. 10D). The area represented by spectrum IS05 contains abundant green vegetation which is apparent from the shape of the reflectance spectra that are typical for green vegetation in the 0.4–1.5 m wavelength range (spectrum USGS G in Fig. 10D). Spectrum IS06 shows absorption wings in the 0.4–0.9 m wavelength regions which is characteristic of dry plant materials (spectrum USGS D in Fig. 10D). Absorption wings in this position are generated by intense blue and ultraviolet absorptions in com-
pounds such as lignin, tannins and pectins (Elvidge, 1990). The USGS vegetation library was used to compare green (USGS G) and dry (USGS D) vegetation with image spectra. Spectral end members of the five surface materials (evaporite series, High Magadi beds, chert series, dense vegetation and volcanic tuff) were used to classify selected MTMF and MNF bands using the Maximum Likelihood method. The result of the classification is shown in Fig. 11D. The minerals Magadiite and Kenyaite could not be identified in the imagery due to their limited exposure at the surface and their shallow diagnostic absorption features compared to the noise of the Hyperion image spectra. Kenyaite always occurs as nodules and never forms as a continuous bed. Although Magadiite precipitates as a bed due to its small diagnostic absorption feature, it is difficult to map even from hyperspectral images. 3.3. Accuracy assessment Classifications for all pixels, which were superimposed on the 32 point locations of the ground data, were extracted from the classified raster surface mineral map (Fig. 11D). The corresponding error matrix is shown in Table 2. The overall accuracy of the classification is 84.3%. The producer’s accuracies of trona, High Magadi bed and the chert were reasonably high, and therefore satisfying the primary purpose of this research, which is the mapping of spatial distribution of the evaporites and precipitated minerals in a saline lake environment. A more careful inspection of the error matrix shows that there is a significant confusion between the “High Magadi bed” and “chert” classes. This confusion is acceptable due to two main reasons:
Author's personal copy G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
◦ Though chert is not always observed in the field with large spatial extent, it is always found with the silica-rich High Magadi bed (spatial association). ◦ The reflectance spectra of chert and High Magadi bed are relatively similar due to presence of SiO2 as the main constituent; hence it is difficult to differentiate them using remote sensing methods (spectral resemblance). 4. Discussion This study demonstrated the possibility of mapping evaporite minerals and associated sediments in Lake Magadi area using space-borne hyperspectral Hyperion data. Spatial distribution of chert and High Magadi beds in the mineral map is superimposed with the published geology map of the area (Fig. 11). Volcanic tuff, which is not recorded in published geological maps known to the authors, can be seen in southern part of the study area in derived mineral map, close to the Lenderut volcano (Fig. 1B). Alluvium terraces in the northern part of Lake Magadi study area, however, were not identified and categorized using remote sensing due to their inhomogeneity of surface minerals (materials). Some of the areas (e.g., contact zones between areas) were left unclassified to facilitate mapping of only the homogeneous surface materials. Even though most of the precipitates and evaporites show characteristic spectral features reflecting the mineralogy of the material, several of them could not be identified because of lack of spectral information related to the mineralogy of some of the evaporites. While the fine-grained silica rich sedimentary rock, chert, exhibits a broad absorption feature at 2.2 m, other siliceous materials such as green beds, diatomite and High Magadi beds exhibit a combination of a broad absorption feature at 2.2 m and a narrow feature at 2.3 m. The main evaporite mineral in the study area, trona, commonly exhibits 6 absorption features at 1.50-, 1.74, 1.94-, 2.03-, 2.22- and 2.39-m. Not only the position and shape of the absorption feature, but also the general shape of the reflectance spectra can be used to identify mineral precipitates and evaporites in the study area. Chert, diatomite, High Magadi beds, green beds and trona were identified using reflectance spectra. Chert was mapped from the Hyperion image based on its 2.2 m broad absorption feature in SWIR region. Diatomite, green beds and High Magadi beds show a relatively similar general spectral shape and absorption features. In addition, diatomite and green bed did not show large spatial extents and identification in the field was solely based on small spots within the High Magadi beds. Diatomite was identified only in one location covering an approximately 2 m × 8 m area. Therefore diatomite, green beds and High Magadi beds were mapped as one unit and referred to as the High Magadi bed. 5. Conclusions Not all the ground mineralogy that was identified by reflectance spectra could be mapped. This is due to factors such as the spectral resolution of the image, the spectral resolution of diagnostic absorption features of the field reflectance spectra, noise in the image, spatial extent of the ground target, spatial resolution of the image, and accuracy of image prepossessing stages. Furthermore, geologic surfaces are often partially covered with non-geologic materials or composed of mixtures of minerals with varying grain sizes and differing degree of compaction and weathering. These factors greatly influence the remote spectral measurements and limit the number of pixels that can be classified and mapped. The overall shape of the laboratory/library spectra was different from the image spectra, probably due to the presence of
31
variable mineral mixtures, grain size variations, residual atmospheric absorption features, desert varnish top of the rock surface, and calibration error of laboratory spectrometer and/or Hyperion instrument. These spectral differences point to the advantage of using image spectra rather than library/laboratory as reference in MTMF processing. The area that had very high spectral variation attenuated the spectral dimensionality of the surrounding low spectral variation area of the Hyperion image. Image processing in two steps, with and without (masking out) the high spectral variation area, could successfully overcome this problem, allowing extraction of information from the entire image. Acknowledgements The authors would like to thank NASA for making the Hyperion image of the area freely available for research, and Boudewijn de Smeth, Chris Hecker for their assistance during the paper preparation. We highly appreciate technical support from the World Agroforestry Centre (ICRAF), in particular from Mr. Elvis Weullow, and from the Department of Geology, University of Nairobi, Kenya. References Aines, R.D., Rossman, G.R., 1984. Water in minerals? A peak in the infrared. Journal of Geophysical Research 89 (B6), 4059–4071. Baker, B.H., 1958. Geology of the Magadi area. Report Geologycal survey of Kenya 42. Biggar, S.F., Thome, K.J., Wisniewski, W., 2003. Vicarious radiometric calibration of EO-1 sensors by reference to high-reflectance ground targets. IEEE Transactions on Geoscience and Remote Sensing 41 (6), 1174–1179. Clarck, R.N., King, T.V.V., Klejwa, M., Swayze, G.A., 1990. High spectral resolution reflectance spectroscopy of minerals. Journal of Geophysical Research 95 (B8), 12,653–12,680. Crowley, J.K., 1993. Mapping playa evaporite minerals with AVIRIS data: a first report from Death Valley, California. Remote Sensing of Environment 44, 337–356. Crowley, J.K., Hook, S.J., 1996. Mapping playa evaporite minerals and associated sediments in Death Valley, California, with multispectral thermal infrared images. Journal of Geophysical Research 101 (B1), 643–660. Elvidge, C.D., 1990. Visible and near infrared reflectance characteristics of dry plant. International Journal of Remote Sensing 11 (10), 1775–1795. Eugster, H.P., 1969. Inorganic bedded cherts from the Magadi area. Kenya: Contributions Mineralogy Petrology 22, 1–31. Eugster, H.P., Jones, B.F., 1968. Gels composed of sodium–aluminium silicate, Lake Magadi Kenya. Science 161, 160–163. Gaffey, S.J., 1985. Reflectance spectroscopy in the visible and near-infrared (0.35–2.55 pm): applications in carbonate petrology. Geology 13, 270–273. Goetz, A.F.H., Vane, G., Solomon, J.E., Rock, B.N., 1985. Imaging spectrometry for earth remote sensing. Science 228 (4704), 1147–1153. Hewson, R.D., Cudahy, T.J., Mizuhiko, S., Ueda, K., Mauger, A.J., 2005. Seamless geological map generation using ASTER in the Broken Hill-Curnamona province of Australia. Remote Sensing of Environment 99 (1–2), 159–172. Hubbard, B.E., Crowley, J.K., Zimbelman, D.R., 2003. Comparative alteration mineral mapping using visible to shortwave infrared (0.4–2.4 m) hyperion, ALI, and ASTER imagery. IEEE Transactions on Geoscience and Remote Sensing 41 (6), 1401–1410. I.T.T.ENVI, 2007. ENVI User’s Guide, v. 4.4. Jenkins, R., 2000. X-ray Techniques: Overview. Encyclopedia of Analytical Chemistry. John Wiley & Sons Ltd. Jones, B.F., Eugster, H.P., Rettig, S.L., 1977. Hydrochemistry of the Lake Magadi basin, Kenya. Geochimica et Cosmochimica Acta 41, 53–72. King, M.D., et al. (Eds.), 2003. EOS Data Products Hand Book, vol. 1. NASA/Goddard Space Flight Center. Kruse, F.A., 1988. Use of airborne imaging spectrometer data to map minerals associated with hydrothermally altered rocks in the northern Grapevine Mountains, Nevada, and California. Remote Sensing of Environment 24 (1), 31–51. Kruse, F.A., 2003. Preliminary Results- Hyperspectral Mapping of Coral Reef Systems Using EO-1 Hyperion. JPL Publication 04-6, Buck Island, U.S. Virgin Islands, pp. 157–173. Mason, P., 2002. MMTG A-list hyperspectral data processing software. In: Manual Version 1.0. CSIRO Division of Exploration and Mining, p. 103. Pearlman, J.S., et al., 2003. Hyperion, a space-based imaging spectrometer. IEEE Transactions on Geoscience and Remote Sensing 41 (6), 1160–1173. Salisbury, J.W., D’Aria, D.M., 1992. Emissivity of terrestial materials in the 8–14 m atmospheric window. Remote Sensing of Environment 42, 83–106. Surdam, R.C., Eugster, H.P., 1976. Mineral reactions in the sedimentary deposits of the Lake Magadi region, Kenya. Geological Society of America Bulletin 87, 1739–1752.
Author's personal copy 32
G.R.L. Kodikara et al. / International Journal of Applied Earth Observation and Geoinformation 14 (2011) 22–32
Turdu, C.L., et al., 1999. Influence of preexisting oblique discontinuities on the geometry and evolution of extensional fault patterns: evidence from the Kenya rift using SPOT imagery. Geoscience of Rift Systems-Evolution of East Africa: AAPG Studies in Geology 44, 173– 191. Van der Meer, F., 2001. Spectral matching using pixel cross-correlograms for the analysis of LANDSAT TM data. International Journal of Applied Earth Observation and Geoinformation 3 (2), 197–202.
Van der Meer, F., 2006. The effectiveness of spectral similarity measures for the analysis of hyperspectral imagery. International Journal of Applied Earth Observation and Geoinformation 8, 3–17. Vaughan, R.G., Hook, S.J., Calvin, W.M., Taranik, J.V., 2005. Surface mineral mapping at Steamboat Springs, Nevada, USA, with multi-wavelength thermal infrared images. Remote Sensing of Environment 99 (1–2), 140–158. Warren, J.K., 2006. Evaporites: Sediments, Resources and Hydrocarbons. Springer, p. 1032.
