Spectral signatures and spectral mixture modeling as a tool for targeting aluminous laterite and bauxite ore deposits, Koraput, India Material and method: Image acquisition: Landsat TM data of 3rd march, 2000 (mid dry season) was acquired for this study. All bands were used in this study except band number 6. All the image processing and spectral processing were done in ENVI 3.2 and ILWIS 2.2. Image analysis:
Digital images from space borne imaging sensors can be visualized either as single band images or as additive color composites using the three primary colors; Red, Green and Blue (RGB). The RGB displays are used extensively in digital processing to display normal color, false color and hybrid color composites. Landsat TM images are acquired over three different range of Electro magnetic spectrum (EMS) such as visible (band 1,2 and 3), near infrared (band 4) and mid infrared (band 5 and 7). For the purpose of geological feature extraction Landsat TM band 7 (spectral range: 2.08 to 2.35 mm) has already proved to be most suitable for lithological discrimination. TM band 4 (spectral range: 0.76 to 0.90 mm) is the only band representing near infrared region and gives good information about the vegetation cover. Visible bands mostly give similar signatures for many of the earth objects. But band 3 (spectral range: 0.63 to 0.69 mm) is proved to be useful for soil-boundary and geological- boundary delineations. Band 3 also exhibits more contrast than bands 1 and 2 because of the reduced effect of atmospheric attenuation. Therefore a colour composite of bands 7(red), 4(green), 3(blue) was prepared to delineate the laterite and bauxite cappings as well as the extent of vegetation and soil cover in this region (figure 2). In this figure; all the red patches on the hilltops mostly correspond to aluminous laterite and bauxite deposits where as green patches are vegetation cover and pink colour corresponds to bare soil even though specific separation of endmembers are not possible in this process. 
Digital data acquired from satellites are provided to the user in the form of quantified and calibrated values (QCALs) for individual picture elements (pixels). These post-calibration QCAL values are in units of digital numbers, which have a full range of 8 bits. Conversion from calibrated QCAL values of the raw image data to spectral radiance, Ll, was done using the following equation (Markham and Barker 1986). 
Where QCAL = Calibrated and quantified scaled radiance values in digital numbers Lmin(l) = Spectral radiance value at QCAL = 0 Lmax(l) = Spectral radiance value at QCAL = QCALmax QCALmax = Range of rescaled radiance values in DN Ll = Spectral radiance Values for Lmax and Lmin vary for each of the Landsat satellites and for different period of their use. Often the L max and L min and Ll values published for a given sensor are expressed in units of milliwatts per square centimeter per steradian per micrometer (mW cm-2 ster-1 mm-1). Subsequently, spectral radiance was correlated for solar irradiance by converting the values obtained from the above equation to effective at satellite reflectance or planetary TM albedo, rp, by 
Where rp = Effective at satellite planetary reflectance composed the combined surface and atmospheric reflectance of the earth (unitless). Ll = Spectral radiance at sensor (in mW cm-2 ster-1mm-1). d = Earth-Sun distance in astronomical units. Esun(l) = Mean solar exoatmospheric irradiances (in mW cm-2 ster-1mm-1). qs = Solar zenith angle (in degrees), and p = 3.1416. The updated mean solar exoatmospheric irradiances, Esun (l), from Markham and Barker (1985) are given in table 1. 
The Earth-Sun distance, d, was approximated as 
Where D is the day number of the year (Van der Meer 1996). In the present study one pre-defined algorithm in ENVI (Environment for image analyst) was run to convert DN values into raw radiance values. Necessary parameters for calibration are given in table 2. 
A further nominal calibration to reflectance using a standard atmospheric model was not done because the necessary input data and software facilities for such model were not available. Alternately, a haze correction formula by means of dark-object subtraction with band minimum was applied, thus disregarding atmospheric absorption and assuming that atmospheric scattering is an additive component that has the effect of adding a constant value to each pixel in a spectral band. This calibration was applied uniformly to each of the TM bands, thus assuming a constant atmosphere across the image. Finally spectral processing of Landsat TM dataset was carried out by converting the digital data from quantized and calibrated values (QCALs) to reflectance values. All the reflectance values of six TM bands were used interactively to locate pure pixels within the dataset. The pure pixels are derived through Pixel Purity Index (PPI) which is a means of finding the most ‘spectrally pure’ (extreme) pixels in multispectral and hyperspectral images (Boardman et al. 1995). The pixel purity index is computed by repeatedly projecting n-dimensional scatterplots onto a random unit vector. The PPI is typically run on a Principal Component transformation image excluding noise bands. The extreme pixels in each projection are recorded and the total number of times each pixel is marked as extreme is noted. A ‘pixel purity image’ is created in which the DN of each pixel corresponds to the number of times that pixel was recorded as extreme. These spectrally pure pixels derived from the multispectral images are plotted in an n-dimensional Scatterplot. The n-Dimensional scatterplot allows for interactive rotation of data in n-D space, selection of groups of pixels into different classes (Boardman and Kruse 1994). The Scatterplot of the pure pixels of TM bands enabled to locate, identify, and cluster the purest pixels and most extreme spectral responses in the data set (figure 3). Here in this study this procedure was followed to isolate different groups of pixels representing different materials. Several groups of pixels such as redsoil, vegetation, laterite/bauxite, Khondalite and mud water (red mud slurry coming out of alumina plant) were isolated and mostly projected at the corners of the scatter plot or completely isolated in the interactive scatterplot (figure 3). It was observed during the field visit that forest concentration is found mostly in the hilly parts. Sal (Shorea robusta) forests are abundant in the eastern and southern part of the study area. Other areas are either cultivated or covered by scrubs. Soil is mostly red in colour due to high iron content in the soil. The hills containing bauxite ores are covered with hard and compact laterite cappings with little or no soil cover on the top and are generally devoid of vegetation cover. The sequence of bauxite deposit in this area is as follows: weathered khondalite or lithomerge at the bottom of the sequence, above that bauxite ore and on the top of the bauxite horizon thick laterite cover of about 5 to 6 meters are present. Soil cover of about ½ to 1 meter is found in patches and at some places laterite is exposed on the surface. Out of the five classes (endmembers) separated in the above process, mud water, vegetation and red soil could be considered ‘pure’ in the sense that they would only comprise single surface material. But khondalite and laterite/bauxite cannot be considered ‘pure’ in the similar manner. On the contrary, we selected those areas that show a representative mix of different ground materials (e.g., green grass, dry grass, soil and lithology) that are characteristic for the laterite/bauxite cappings and khondalite.  Figure 3: n-D Scatterplot of the pure pixels. The pixels market in the corners correspond to their position in the TM image  Fig 3: Predictive mineral potential map outlying high potentiel areas Spectral processing method applied here is Mixture Tuned Matched Filtering (MTMF) which performs a partial unmixing - finding the abundances of user defined materials (endmembers). All the endmembers in the image need not to be known for applying this technique. This technique maximizes the response of the known endmember and suppresses the response of the composite unknown background, thus "matching" the known signature. The matched filtering results appear as gray-scale images with value ranging from 0.0 to 1.0 (Zero to One). These images provide a means of estimating relative degree of match to the reference spectrum and approximate sub-pixel abundance, where 1.0 is a perfect match with the reference spectrum and 0.0 is no match situation. It provides a rapid means of detecting specific materials based on matches to library or image endmember spectra and does not require knowledge of all the endmembers within an image scene (Boardman et al., 1995). Here three classes (endmembers) such as laterite/bauxite, vegetation and red soil were considered for the spectral processing of the image. Spectra of these three groups of pixels were plotted in figure 4. Khondalite and mud water endmembers were not considered for the spectral processing, not because their occurrence is very less at pixel resolution but because laterite/bauxite, vegetation and red soil describe about 95% of the variance in the image. |