Mineralogical mapping using EO-1 Hyperion data for iron mine identification

. The Tamera open pit mine is located in the Nefza mining district (north-west of Tunisia) where the reddish soil reveals its strong potential for heavy Fe mineralization, along with other metals (Pb, Zn, and Mn). This study explored Iron oxide mineralization in a particular nappe zone using multiscale field visible and near infrared/short waves infrared and EO-1 Hyperion hyperspectral data. Thus, spectral features fittings were conducted on Hyperion data for mineral mapping using scene-extracted endmembers. A first minimum noise fraction algorithm was used to reduce the hyperspectral data ’ s dimensionality, followed by the Pixel Purity Index and n -dimensional visualization to extract the pure pixels (endmembers) from the cluster of pixels in the data cube. Furthermore, 56 soil samples were collected from the test site and spectrally measured using Analytical Spectral Devices FieldSpecHiRes spectrometer, which covers the spectral range between 350 and 2500 nm. All samples were analyzed in the laboratory with x-ray diffraction. Endmember extracted spectra were compared with field measurements before EO-1 Hyperion processing. Mapping results revealed the occurrence of a mineralogy composition consisting particularly of iron oxide minerals (hematite and goethite) and bearing minerals, such as clays (kaolinite), within all samples. © The Authors. Published by SPIE under a Creative Commons Attribution 4.0 International License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI. [DOI: 10 .1117/1.JRS.16.024514]


Introduction
Tunisian iron minerals mainly include Fe oligist (hematite), goethite, and stilpnosiderite. These three minerals make up almost all exploitable minerals and are accompanied by traces of magnetite, lepidocrocite, and various impurities varying with the deposits. 1 In Tunisia, two types of ores can be distinguished, an oolithic sedimentary deposit (Jebel Ank) and epithermal deposits, with the following aspects among them: 1. Northern deposits are considered to be the result of the substitution of Pontian sediments by iron oxides, and they appear to be related to underlying Blende-Pyrite-Galena deposits. 2. Central Tunisia's deposits are mixed, vein, and substitute deposits in limestones generally of aptian age. 3. Gossans can also be distinguished from varied deposits. The first type can be related to Harraba, I'Hameima, Slata, Jerissa, Nebeur, the Bellouta, the Mrhila, and the Chambi. The second type of epithermal deposits belongs to the mines of Ras Rajel, Moktael Hadid BouLanague, Tamera-Ganara-El Harech, and Douahria, which are located in the Nefza district and are considered to be a major contribution to the iron supply in the north of Tunisia.

Study Site
The study area (37°3.318'N, 9°6.43'E) is situated in the north-west of the Oued Belief volcanosedimentary complex (Belt), 120 km NW of the city of Tunis, 30 km to the east of the town of Tabarka, and 10 km north of Nefza province (Fig. 1). The Nefza region is characterized by high vegetation and a tropical microclimate environment with an average rainfall of about 100 mm/year. In the middle of the century, this province was known as a mining district center because of its tremendous economic potential of ore deposit and different Pb-Zn and Fe mineralization indexes. It is part of the widespread northern Tunisian "Nappe zone," 2,3 composed of thrust sheets. The small post-nappe basins, ranging from the Messinian to Pliocene in age and generally located near magmatic rocks, host Fe and Pb-Zn ores. The Pb-Zn sulfide ore relating to these post-nappe basins is observed at Sidi Driss and consists of galena and sphalerite associated with As-bearing marcasite, pyrite, barite, and celestite.
From the outside to the center of Oued Belief's complex structure (Fig. 3), we can see a primary edge breccia creating a belt, mainly East North East-West South West, along the axis of the elliptical shape (6 km by 3 km). 3 The secondary Breccia can take the form of large plages aligned along accidents with a subparallel direction to that of the edge breccia, for example, the west sector of Boukhchiba, or of scattered blocks preferentially located upstream of N-S trending discontinuities, for example, the block near the granodiorite. The ferruginous breccia of the margin and those of the interior, overall, have little morphological differences: they are all extremely cemented sheaths. They produce projections over the entire structure and are missing in the rivercut low parts 2,46,47 (Fig. 3). They are distinguished by the size of their elements, which varies depending on their placement. The ferruginous bodies of Boukhchiba, Raget Ouled Mariem, and Kef Salem are encased in large lenses brecciated at the base, whereas the summit horizons, surface alteration, and pedogenesis have formed banks and layers intercalated in ferruginous silts,   causing concentrations of centimetric to decametric iron. Fractures as well as syn-deposit fractures are filled with hydrothermal clays. 1,3,45,46 3 Materials and Methods

Field Sampling and Laboratory Analysis
A sampling campaign was conducted between March and April 2017 in Nefza to define the diverse lithological facies at the level of iron mining sites: the site of Tamera, the Pb-Zn mine of Sidi Driss, and the abandoned site of Boukhchiba. Climatic conditions were excellent during the sampling, with clear weather locally supplemented by some cloudy arrivals that became more varied to quite dense on several days.
A layer of 0.5 to 1.5 cm was scraped from the surface for each of the 65 samples collected as follows: 20 surface samples for the "Tamera" iron mine site, 20 surface samples for the Sidi Driss Pb-Zn mining site, and 16 surface samples for the "Boukhchiba" iron mining site. All selected samples were then reduced to a powder size of more than 2 mm in an agate mortar and analyzed by XRD for the identification of minerals on a PANalyticalX'Pert Pro diffractometer and the estimation of their respective abundance (Fig. 4).

Hyperspectral Field Data
In situ spectroradiometric hyperspectral measurements were acquired using the Analytical Spectral Devices (ASD) FieldSpecHiRes spectrometer, which operates in a full range between 350 and 2500 nm to provide a prior knowledge of geology. Aiming for maximum solar lighting, all measurements were conducted between 10 and 2 μm. Latitude and longitude coordinates of all measured samples were recorded using a portable global positioning system (Fig. 5).

EO-1 Hyperion Imagery Data
In the framework, simultaneous to the field camping period, we used hyperspectral Hyperion data for iron ore mapping in the Nefza district. "Environment for Visualizing Images" (ENVI) 5.1 software package was used to process the Hyperion, Level 1T dataset scenes that took place on March 25, 2003 (Table 1). The choice of the scene used is justified by its good quality with a lower density of cloud cover as well as the seasonal coincidence between its acquisition date and the field campaign to maintain the same meteorological conditions. Moreover, the scene of the March 25, 2003, is the only one that covers the entire study area. The hyperspectral remote sensing satellite, EO-1 Hyperion, was launched in November 2000 with two spectrometers in VNIR and SWIR covering the wavelength range of 400 to 2500 nm with 242 bands of approximately 10 nm spectral bandwidth and 30-m spatial resolution. 51 The VNIR (VIS + NIR) detector covers the range of 400 to 900 nm in 40 channels and the SWIR detector covers a range of 1000 to 2500 nm in 172 channels. 52

Methodology
The proposed methodology consists of two main parts, namely, the preprocessing of the datasets and the main processing. The flowchart of the methodological steps followed in this study is presented in Fig. 6.

Preprocessing
Hyperion image atmospheric and geometric corrections. Pre-processing steps included the removal of overlapping and inactive bands (1 to 7, 58 to 78, and 225 to 242). 53 Only 153 bands were used for radiometric calibration after suppressing the bad bands. The surface reflectance was derived after that from the radiance data. Both atmospheric and geometric corrections were processed using the ENVI 5.1 software to remove noise and acquire surface reflectance. The fast line of sight atmospheric analysis of hyperspectral cubes atmospheric correction algorithm was used to correct the absorption and scattering effects. Finally, after the low signals and noisy channels were removed, a resulting Hyperion image with 153 bands was used for this study.  Hyperion image enhancement. The band rationing was widely used as a simple enhancement technique in remote sensing analysis. It has been used to emphasize the anomaly of target alteration minerals that have absorption features in certain wavelengths, such as iron and clay indexes, for instance, which have absorptions around 550, 660, 760, 800, 860, 1600, 1700, 2167, and 2203 nm wavelength regions, respectively. [54][55][56] Geological mapping has benefited from the band ratio technique in several studies, 57 particularly when using hyperspectral imagery data. 39 In this study, the band ratio based on Hyperion data successfully delimited iron ore zones in Tamera, which contain ferric and ferrous oxide minerals as well as bearing minerals.
Both indices were calculated for semi-automatic mapping of ferric iron oxides and clays in the iron ore deposits in the three selected areas, using the equations E Q -T A R G E T ; t e m p : i n t r a l i n k -; e 0 0 1 ; 1 1 6 ; 1 9 0 Ferric iron oxide ¼ ðb23∕b13Þ; (1) E Q -T A R G E T ; t e m p : i n t r a l i n k -; e 0 0 2 ; 1 1 6 ; 1 4 7 Clays ¼ ðb134 × b138∕b1482Þ; where ρ red and ρ NIR refer to red band surface reflectance and near infrared band surface reflectance, respectively. The NDVI was calculated, and the respective mask was applied to exclude pixels with dense vegetation from further processing and thus to emphasize the geological interpretation.

Main processing
MNF transformation. With regard to image quality, the minimum noise fraction (MNF) transform used in this study is capable of supplying excellent high to low order to the bands. MNF specifies how geographical information can be used to interpret the signal and noise covariance structure. 58 The MNF procedure was implemented to decrease the dimensionality of Hyperion data by suppressing noise. It is also useful for displaying variance within the data set. 59 In the first eigen value image, there is a bright gradient with a significant smile. To reduce spatial noise, the MNF was employed to separate uncorrelated noises. 59 In general, the first few MNF bands communicate the most significant information, whereas the subsequent bands become increasingly noisy. Noise is often present in MNF bands with computed eigen values below one. 60 It was decided to save just 10 of the MNF bands and turn them into reflectance data. The resulting reflectance dataset contained 153 bands, but only with the spectral information of the chosen MNF bands.
Endmember extraction. For more than two decades, many methods for endmember bundle extraction from hyperspectral images have been proposed, 61,62 with many of them extracting endmembers from each subregion of the original image and then clustering the extracted endmembers to generate endmember bundles, which means that each feature's endmembers are represented by a collection including numerous spectra. 63 When the subregion contains no pure pixels, however, the experiment methods may extract mixed pixels as possible endmembers. The researchers used Pixel Purity Index (PPI) 64 to assess the purity of candidate endmembers to overcome these concerns. In this paper, the PPI algorithm is used to extract the pure pixels of oxide and clay minerals.
Spectral endmembers were extracted from Hyperion imagery using a combination of the MNF, PPI, and n-dimensional visualizer technique. 65,66 The PPI is calculated by projecting n-D scatter plots onto a random unit vector many times. In each projection, ENVI maintains track of the extreme pixels (those at the ends of the unit vector), as well as how many times each pixel has been designated as extreme in all of the projections (ENVI user guide 2003). During the region of interest's (ROI) selection, which is geologically known, 67,68 the highest pixel values are used to compute the n-dimensional visualization. To extract the pure pixels and determine their spectra, the n-dimensional visualization is applied to MNF images. The extracted spectra were characterized using the spectral analysis procedure available at ENVI.
Spectral feature fittings. In this study, we used the spectral feature fittings (SFF) method based on the comparison of spectra with a mineral reference spectrum and the estimation of the root mean square error (RMSE). A low RMSE and a high scale factor value indicate that the minerals are closely matched. The RMSE image quantity and the condition scale will be equal. 44,[69][70][71] The continuum removal was applied before creating scaled images, which is given by E Q -T A R G E T ; t e m p : i n t r a l i n k -; e 0 0 5 ; 1 1 6 ; 4 8 5 Scr ¼ S∕C; where Scr is the continuum-removed spectra, S is the original spectra, and C is the continuum curve.
Steps in SFF include creating a scale image and an RMS error image, or RMSE. After subtracting the continuum, which removed one spectrum at a time from the image, scale images were created. Using the scale image, each pixel's absorption feature can be predicted and measured. The scale factor, which is calculated by comparing bands to bands using the least-square fit procedure, measures the abundance of spectral features. A higher scale factor value correlates with a higher mineral absorption rate. 72 According to the brightness of the scale picture, the image spectrum was more closely aligned with the reference spectra of minerals in the pixel than what was previously believed. A low RMSE and a high scale factor value indicate that minerals are closely matched. RMSE image quantity and condition scale were thus equal.

Mineralogical Composition
The XRD results show high levels of hematite and goethite within Tamera samples. Other minerals were mainly identified through the mineral composition such as quartz and clays minerals. Feldspath, sulphate minerals (barite and gypsum), and sulfide minerals such as galena were, however, identified as accessory minerals. Jarosite was found within only nine samples, indicating that this iron sulfate mineral is -significantly non-existent [ Fig. 8(a)]. The mineralogical results of the Sidi Driss samples show high levels of quartz due to the lithological nature of Sidi Driss series. Moreover, except the presence of quartz, XRD results revealed the existence of a similar mineralogical composition containing mainly clay minerals, hematite, goethite, and sulphate minerals (barite and gypsum) [ Fig. 8(b)]. Specifically, the Boukhchiba iron mine, shows the highest concentration values of clay minerals (kaolinite (d ¼ 7.185 A)], which was estimated to be around 50% [ Fig. 8(b)].

Fe Mineralization Surface Spectra
The presence of oxides and hydroxides was revealed through all of the measured spectra in the three sites. Both important absorption features were represented by approximately 0.520 and 0.890 μm, which were reciprocally proportional and related to the presence of ferric iron (Fe 3þ ). 73 The presence of jarosite was not revealed through all soil sample spectra. 39 For total clays, absorption peaks were expected to approximately be within the SWIR region at 2.206 μm, due to the combination of OH and OH-Al bending. 25,39,73 The ASD spectra matched the laboratory results (Fig. 9).
The existence of ferric iron and clay minerals was verified by comparing measured and resampled spectra of selected samples with those extracted from the Hyperion subset image (Fig. 10).
Furthermore, a subtle displacement in the unique absorption properties of these minerals was detected when comparing these processed ASD and Hyperion extracted spectra with U.S. Geological Survey oxide (hematite, goethite,) and clay (kaolinite) spectra at around 0.49, 0.56, 0.88, and 2.203 μm, respectively (Fig. 10). Thus, the presence of hematite and goethite minerals can explain the detected absorption feature at around 0.52 μm in both ASD recorded spectra and the extracted ones from the Hyperion image. XRD results confirmed that oxide and clay minerals are present in all selected samples.

Characterization of Iron Oxide and Clay Minerals
The Hyperion image was utilized to construct the band ratio technique ferric and ferrous iron index, which was then used for delineating zones showing the subsurface existence of iron ore. 40,54-56,74,75 NDVI's and NDWI's goal is to hide heavy vegetation and water regions and turn the pixels with high vegetation and water values in black color, which means there is no value. A water mask also was generated according to a visual inspection and field truth. In particular, Boukhchiba is a mine bonded to a dense forest in the northern area, so when we replicate the mask of vegetation, the northern zone turned black. The two black holes located in Sidi Driss are small lakes. The spatial distribution of clay mineral and iron oxide classes is determined and given in Table 2. Based on the same threshold level for iron oxide and clay ratio bands, it was discovered that the iron oxide ratio band has the highest value of the total district area with 1.707% in the image sequence. Most of the oxides and iron minerals are geographically distributed in the Tamera mine, whereas clay minerals are concentrated in the Boukhchiba mine.

Iron Oxide and Clay Mineral Mapping
The SFF algorithm was conducted on the subset of the Hyperion data using extracted endmember spectra. The scene extracted spectra were compared with their corresponding measured Fig. 9 ASD soil spectra taken from the three sites with their corresponding mean abundance (%) of iron and clay minerals: (a) reflectance curves of "Tamera" soil samples, (b) reflectance curves of "Sidi Driss" soil samples, and (c) reflectance curves of "Boukhchiba" soil samples. spectra before processing, thus serving as ground truth (Fig. 11). Furthermore, a subtle displacement in the unique absorption properties of these minerals was detected when comparing these properties, which were measured by ASD, and the endmembers spectra extracted from Hyperion oxide (hematite, goethite,) and clay (kaolinite) spectra at around 0.49, 0.56, 0.88, and 2.203 μm, respectively (Fig. 11). Thus, the presence of oxide minerals can explain the detected absorption feature at around 0.52 and 0.89 μm in both ASD recorded spectra and the extracted ones from the Hyperion image. Moreover, the clay minerals class extracted from the Hyperion and measured by ASD shows a strong absorption peak at around 2.203 μm.
Iron oxide (hematite and goethite) and clay (kaolinite and illite) mineral distributions are showed in the resulting maps after SFF method-based Hyperion processing (Figs. 12 and 13). As a result of the Hyperion image processing, iron oxide (hematite and goethite) was identified  Fig. 11 Comparison between the extracted spectra from the image and those measured by the ASD for oxides and clays. in the three studied areas with high concentration in the Tamera region, where such minerals are dominant. However, the Boukhchiba and Sidi Driss sites show low and very low abundances of iron oxide minerals, respectively (Fig. 12). Iron oxide amounts of 28.52%, 8.61%, and 14.19% were estimated for the Tamera, Sidi Driss, and Boukhchiba mines, respectively. The primary edge breccia Raget Ouled Mariem, located to the North of Boukhchiba, and Kef Salem, situated in its central-eastern part, have shown considerable values of iron oxide at around 15.6% and 16.36%, respectively. Some other pixels, not belonging to zones with dense vegetation and that are attributed to interior secondary breccias, revealed moderate values of iron oxide at around 9.22%. These results converge with those determined by the mineralogical analysis through XRD. The significant amounts of iron oxides, which were revealed by the resulting map and considered to be economically profitable, suggest that Boukhchiba site could be re-exploited.
Moreover, band ratio results revealed similar geographical distribution of iron oxide mineral contents, particularly high in the Tamera site, moderate in the Boukhchiba site, and low in the Sidi Driss site [ Fig. 7(a)].
According to the resulting map, clay minerals (kaolinite and illite) were detected with a higher grade in the Boukhchiba site. A lower grade was, however, revealed in the Tamera and Sidi Driss sites. The determined amounts of clays ranged from 6.95% to 57.14% in Boukhchiba and less than 56.59% in Tamera and Sidi Driss. Clay estimated contents are close to those measured by XRD analysis, with mean values at 31.19% compared with 16.38% in Boukhchiba, 5.83% compared with 15.1% for Sidi Driss, and 17.47% compared with 15.63% for Tamera. A convergence between clay maps and band ratio results were found, thus confirming their significance.

Validation using kriging maps
One of the most often used methods for quantifying classification accuracy is the kriging method. 76,77 A normal krige model is applied to calculate, or rather estimate, the global estimation variances, as well as to identify the kriging weights in the local estimation. 78 In this context, we used the XRD results to generate kriging maps for Iron oxide and clay maps validation. The ordinary kriging model was estimated by the variance of iron oxide minerals in the three sites (Fig. 14). The highlighted region of high iron oxide grades was the same for both the normal kriging method and the SFF classification findings. High oxide amounts were located particularly in the eastern, western, and central areas of Tamera, with over 20%, which overlaid with the greatest value pixels of the SFF classification map, with an amount estimated at about 48.41% [ Fig. 14(a)]. They were, however, revealed in the north of Sidi Driss with maximum values at around 31.51% in the SFF classification map [ Fig. 14(b)]. The highest pixels of iron oxide grade were located specifically in the central area of the Boukhchiba mine with a maximum value at around 35.48% in the SFF classification map [ Fig. 14(c)].
The variability of clay mineral contents in the three sites was also evaluated using the standard kriging model. The spatial distribution showed agreement between the normal kriging map and the SFF classification results. Pixels showing higher amounts were located particularly in the central and northern areas of Tamera's reddish Mio-Pliocene formation with estimated amounts of more than 20%, which was overlaid with the greatest value pixels of clay results from SFF classification with an amount at around 56.59% [ Fig. 15(a)]. Furthermore, the SFF classification map of the Sidi Driss showed moderate to low values of clay contents, with a maximum of 15.96%, which converges with the kriging map results [ Fig. 15(b)]. According to resulting kriging and SFF maps, Boukhchiba seems to be the site with the highest grade of clay minerals, with a     Fig. 15(c)]. The convergence between the kriging and SFF classification results confirmed the usefulness of the proposed approach for mapping Iron oxide and clay minerals in an attempt to delineate the spatial distribution of Fe ores in the three studied sites. An accurate map of the mineralization will be of utmost interest for selecting the judicious procedure for extracting Fe ores to minimize the impacts on environment.

Validation using test samples
A total of 20 soil samples were randomly selected for the Tamera and Sidi Driss areas, and 16 samples were selected for the Boukhchiba mine. The calculated coefficient of correlation between measured iron oxide abundances by XRD analysis and their corresponding values, estimated from SFF-based Hyperion processing, revealed a relative important convergence between measured and estimated abundances in the Tamera and Boukhchiba mines (R 2 ¼ 0.644 and R 2 ¼ 0.619, respectively). There was less convergence between measured and estimated oxide abundances in the Sidi Driss mine, as revealed by the R 2 of 0.543. For clay minerals, a strong correlation was revealed by an R 2 coefficient equal to 0.715, particularly in the Tamera mine site (Fig. 16). Less important R 2 coefficients were calculated for the Sidi Driss and Boukhchiba mine sites, equal to 0.516 and to 0.527, respectively. These correlation values indicate that Hyperion processing results were highly converged with those measured by XRD laboratory analysis for iron oxides and clays in ore deposits. This study has shown that the SFF-based methodology can be used successfully for iron ore mapping and grade assessment using Hyperion data.

Conclusion
This study aimed to explore Iron oxide mineralization in a particular nappe zone using multiscale VNIR/SWIR and EO-1 Hyperion image data. Thus, a methodological approach was conducted based on the SFF classification and reflectance spectroscopy. A total of 65 soil samples were selected from the three studied sites, and in situ reflectance spectra were taken from the same point locations. The mineralogical composition of each sample was identified through the XRD analysis, which was used for the validation of the resulting mineral maps. In particular, Iron oxide and clay maps were generated based on the SFF classification method and using Hyperion scene-extracted spectra. All extracted spectra were compared with the ASD measured spectra before the classification step.
In situ reflectance measurement revealed the presence of oxide and hydroxide minerals with absorption features of approximately 520 and 890 nm. Moreover, clays were depicted with a principal absorption peak in the SWIR region at around 2200 nm. These results converged with XRD results, which confirmed the presence of oxides (hematite and goethite) and clays (kaolinite and illite).
The generated SFF maps have shown a dominance of oxides in the Tamera mine and lower grades in the Boukhchiba and Sidi Driss mine sites, with estimated contents of around 48.41%, 35.48%, and 31.51%, respectively. According to the results, clays were identified with high content in the Boukhchiba mine with 57.14%. Less important contents were determined in the Tamera and Sidi Driss mines considering the revealed amounts at about 56.59% and 37.92%, respectively. These results perfectly highlighted the Iron oxide and clay areas, which were also enhanced based on band ratio method. Both oxide and clay maps, generated from the SFF classification, were validated using XRD analysis. A first validation was conducted using the generated kriging maps, using XRD results, to validate maps spatially. A significant distribution similarity was revealed between the kriging map and SFF map for the Tamera mine site. The second validation step was conducted through the comparison between the estimated contents with those measured by XRD analysis of each test sample. In particular, the oxide map that was generated for Tamera mine showed the best correlation coefficient, up to 0.6. The convergence between XRD and SFF classification results confirmed the usefulness of the proposed approach for mapping Iron oxide and clay minerals in an attempt to delineate the spatial distribution of Fe ores in the three studied sites. Moreover, other significant amounts of iron oxides, which were revealed by the resulting map, particularly in the northern and central-eastern parts of the abandoned Boukhchiba site as well as some zones far from the three studied sites, are considered to be economically profitable and thus could be re-exploited. An accurate map of the mineralization will be of utmost interest for selecting the judicious procedure for extracting Fe ores to minimize the impacts on environment. This study has shown that the SFF-based methodology can be successfully used for iron ore mapping and grade assessment using Hyperion data.