Integrated Hyperspectral and Geochemical Study of Sediment-Hosted Disseminated Gold at the Goldstrike District, Utah

The Goldstrike district in southwest Utah is similar to Carlin-type gold deposits in Nevada that are characterized by sediment-hosted disseminated gold. Suitable structural and stratigraphic conditions facilitated precipitation of gold in arsenian pyrite grains from ascending gold-bearing fluids. This study used ground-based hyperspectral imaging to study a core drilled in the Goldstrike district covering the basal Claron Formation and Callville Limestone. Spectral modeling of absorptions at 2340, 2200, and 500 nm allowed the extraction of calcite, clay minerals, and ferric iron abundances and identification of lithology. This study integrated remote sensing and geochemistry data and identified an optimum stratigraphic combination of limestone above and siliciclastic rocks below in the basal Claron Formation, as well as decarbonatization, argillization, and pyrite oxidation in the Callville Limestone, that are related with gold mineralization. This study shows an example of utilizing ground-based hyperspectral imaging in geological characterization, which can be broadly applied in the determination of mining interests and classification of ore grades. The utilization of this new terrestrial remote sensing technique has great potentials in resource exploration and exploitation.


Introduction
The Great Basin of western North America has produced a significant amount of gold, making the United States one of the largest gold producers in the world [1]. Among the gold mines, the most famous ones are of the Carlin-type, carbonate rock-hosted disseminated gold deposits that formed in the Eocene Epoch [1,2]. Since the discoveries of Nevada Carlin-type gold deposits, similar sediment-hosted gold has been searched for around the world [3][4][5][6] and in nearby states [7][8][9]. The Goldstrike district is a gold deposit in southwest Utah similar to Carlin-type deposits [10,11]. The modern production of disseminated gold in Goldstrike was active from 1988 to 1996 producing 209,835 ounces of gold and 197,654 ounces of silver, which ceased because of falling gold price, increasing strip ratios, production costs, and safety concerns [12]. The remaining gold is currently being explored by Pilot Goldstrike Inc., Elko, USA [11].
The exploration and mining of precious metals have been a challenge because of the high expense of drilling, geochemical analyses, and metallurgy tests. On the other hand, hyperspectral imaging as a non-destructive, low cost, and large areal coverage, remote sensing technique, can provide high-resolution mineralogical analyses and is becoming popular in geologic studies [13][14][15][16]. To test the applicability of hyperspectral imaging in mineral exploration, this study combines hyperspectral imaging with fire assay metallurgy and inductively coupled plasma mass spectrometry (ICP-MS) geochemistry data in the study of gold mineralization in a drilled core from the Goldstrike district.

Materials and Methods
This study focused on a core drilled by Pilot Goldstrike Inc. at easting 244,910.51 m, northing 4,141,510.69 m (UTM Zone 12N), on an azimuth of 275° and a dip of −65°. Drilling retrieved core from apparent depths of 24 to 448 ft. (7.3 to 136.6 m), which was 8.5 cm in diameter. The core was split into halves, and one half was again split into two ¼ samples. Half of the core was sent for metallurgical test, one ¼ cut was sent for inductively coupled plasma mass spectrometry (ICP-MS) measurements, and the other ¼ cut was segmented into mostly 5 ft. (

Materials and Methods
This study focused on a core drilled by Pilot Goldstrike Inc. at easting 244,910.51 m, northing 4,141,510.69 m (UTM Zone 12N), on an azimuth of 275 • and a dip of −65 • . Drilling retrieved core from apparent depths of 24 to 448 ft. (7.3 to 136.6 m), which was 8.5 cm in diameter. The core was split into halves, and one half was again split into two 1 4 samples. Half of the core was sent for metallurgical test, one 1 4 cut was sent for inductively coupled plasma mass spectrometry (ICP-MS) measurements, and the other 1 4 cut was segmented into mostly 5 ft. (1.5 m) long sections and imaged by ground-based hyperspectral cameras ( Figure 3A)

Hyperspectral Imaging
Hyperspectral imaging is a remote sensing technique that collects the reflected light spectrum from material surfaces. The reflectance curve contains physical and chemical properties of the material since chemical bonds absorb light at specific wavelengths [21]. Ground-based hyperspectral imaging has been widely used in geologic characterizations [15,[22][23][24][25][26][27][28], in which variations of the sub-centimeter or sub-millimeter scale can be resolved. This study used hyperspectral imaging to identify mineralogy as well as to extract relative abundances of the minerals.
The core samples were placed in fifteen core boxes and scanned with a Specim dual-camera system (Spectral Imaging Ltd., Oulu, Finland). Each scan imaged three core boxes that were placed on a table, and the table was held at around 20° towards the cameras by two people while scanning ( Figure 3B). The hyperspectral camera system consisted of a visible and near-infrared (VNIR) camera over the spectral range of 394-1008 nm at a spectral resolution of 2.8 nm, and a short wave infrared (SWIR) camera over the spectral range of 896-2504 nm at a spectral resolution of 10 nm, and both cameras were push-broom scanners ( Figure 3D). The cameras were mounted with roughly a −10° tilt towards samples on the two arms of a pan and tilt rotating head (FLIR Systems, USA) on top of a tripod, and the rotating head rotated on a horizontal plane so that the push broom cameras swept the core samples. The cameras were about 1.3 to 2.0 m away from the samples, and the spatial resolutions were 1.7 to 2.6 mm for VNIR (with four times of spatial binning) and 1.7 to 2.7 mm for SWIR. Dark current images were taken with lens caps covering lenses, these represented random background noises from the electronics and internal temperature. Due to the inevitable shaking by people holding the inclined table, the hyperspectral imagery is distorted in the across-track direction. The two cameras have different scanning angles on the two sides of the rotating head, thus

Hyperspectral Imaging
Hyperspectral imaging is a remote sensing technique that collects the reflected light spectrum from material surfaces. The reflectance curve contains physical and chemical properties of the material since chemical bonds absorb light at specific wavelengths [21]. Ground-based hyperspectral imaging has been widely used in geologic characterizations [15,[22][23][24][25][26][27][28], in which variations of the sub-centimeter or sub-millimeter scale can be resolved. This study used hyperspectral imaging to identify mineralogy as well as to extract relative abundances of the minerals.
The core samples were placed in fifteen core boxes and scanned with a Specim dual-camera system (Spectral Imaging Ltd., Oulu, Finland). Each scan imaged three core boxes that were placed on a table, and the table was held at around 20 • towards the cameras by two people while scanning ( Figure 3B). The hyperspectral camera system consisted of a visible and near-infrared (VNIR) camera over the spectral range of 394-1008 nm at a spectral resolution of 2.8 nm, and a short wave infrared (SWIR) camera over the spectral range of 896-2504 nm at a spectral resolution of 10 nm, and both cameras were push-broom scanners ( Figure 3D). The cameras were mounted with roughly a −10 • tilt towards samples on the two arms of a pan and tilt rotating head (FLIR Systems, USA) on top of a tripod, and the rotating head rotated on a horizontal plane so that the push broom cameras swept the core samples. The cameras were about 1.3 to 2.0 m away from the samples, and the spatial resolutions were 1.7 to 2.6 mm for VNIR (with four times of spatial binning) and 1.7 to 2.7 mm for SWIR. Dark current images were taken with lens caps covering lenses, these represented random background noises from the electronics and internal temperature. Due to the inevitable shaking by people holding the inclined table, the hyperspectral imagery is distorted in the across-track direction. The two cameras have different scanning angles on the two sides of the rotating head, thus produces different geometries in imagery ( Figure 4). No efforts were made to correct the geometric distortions since geometry was not the focus of the study. spectra of identified mineral absorption features were continuum removed to isolate the non-selective scattering and spectrally inactive mineral effects [31,32]. These absorption features were then modeled with the modified Gaussian model [33] to extract the absorption depth as well as the absorption wavelength. This model is a deconvolution method that models electronic transition bands in reflectance spectra, enables the isolation of absorptions from the continuum and distinct absorption from overlapping wavelengths [33,34]. The absorption depth is a proxy of the mineral abundance [31,35], and the absorption wavelength helps to differentiate similar minerals and to imply mineral chemistry in solid solutions [36,37]. All these image processing steps were performed by Matlab 2016a (MathWorks, Natick, USA) and ENVI 5.5 (Harris Geospatial, Boulder, USA).  The dark current images were subtracted from the image spectra, and the results were converted into at-sensor reflectance using a flat field calibration with a white diffuse reflectance standard. Image stripes were corrected by comparing reflectance values of bad pixels with adjacent pixels. Assuming the noise were spatially related, the images were processed with forward and inverse minimum noise fraction [29] to smooth the spectra and to maximize the signal-to-noise ratio. After smoothing, the five scans were mosaicked together for more straightforward data processing. Regions of interest were manually created on the images for core samples, the other pixels, including core boxes, depth markers, and the white reference panel were masked out ( Figure 4). Minerals were identified by comparing pixel spectra with the U.S. Geological Survey spectral library [30]. The spectra of identified mineral absorption features were continuum removed to isolate the non-selective scattering and spectrally inactive mineral effects [31,32]. These absorption features were then modeled with the modified Gaussian model [33] to extract the absorption depth as well as the absorption wavelength. This model is a deconvolution method that models electronic transition bands in reflectance spectra, enables the isolation of absorptions from the continuum and distinct absorption from overlapping wavelengths [33,34]. The absorption depth is a proxy of the mineral abundance [31,35], and the absorption wavelength helps to differentiate similar minerals and to imply mineral chemistry in solid solutions [36,37]. All these image processing steps were performed by Matlab 2016a (MathWorks, Natick, USA) and ENVI 5.5 (Harris Geospatial, Boulder, USA).
The reflected spectrum hosts several distinctive absorption features, including the overtones of C-O stretch and O-H stretch, combinations of O-H stretch and metal-OH bend vibrations, as well as crystal fields transitions of metal elements, these absorption features enables identification of many minerals [38][39][40]. Common carbonate and phyllosilicate minerals in sedimentary rocks ( Figure 5A) show an Al-OH absorption near 2.2 µm, and a CO 3 2− absorption near 2.34 µm [39,40]. Common ferric iron oxide and hydroxide minerals ( Figure 5B) show absorption features near 0.5 and 0.66 µm [41]. There is another ferric iron absorption near 0.9 µm [41]. However, the bands with long wavelengths (>800 nm) of the VNIR camera had low signal-to-noise ratios, so the absorption feature near 900 nm was not studied. In this study, we examined 50 spectral bands (2109 to 2416 nm) in the SWIR spectrum to look for two absorption bands near 2200 nm and 2340 nm, and examined 118 spectral bands (400 to 736 nm) in the VNIR spectrum to look for the absorption bands near 500 nm and 660 nm. For each core sample section, an average value of absorption depth was calculated for all pixels of the section, and this average absorption depth was compared with geochemical measurements.
Remote Sens. 2019, 11, x FOR PEER REVIEW 8 of 20 respectively. The red lines show the interpreted unconformity between the basal Claron Formation and the underlying Callville Limestone.
The reflected spectrum hosts several distinctive absorption features, including the overtones of C-O stretch and O-H stretch, combinations of O-H stretch and metal-OH bend vibrations, as well as crystal fields transitions of metal elements, these absorption features enables identification of many minerals [38][39][40]. Common carbonate and phyllosilicate minerals in sedimentary rocks ( Figure 5A) show an Al-OH absorption near 2.2 μm, and a CO3 2− absorption near 2.34 μm [39,40]. Common ferric iron oxide and hydroxide minerals ( Figure 5B) show absorption features near 0.5 and 0.66 μm [41]. There is another ferric iron absorption near 0.9 μm [41]. However, the bands with long wavelengths (>800 nm) of the VNIR camera had low signal-to-noise ratios, so the absorption feature near 900 nm was not studied. In this study, we examined 50 spectral bands (2109 to 2416 nm) in the SWIR spectrum to look for two absorption bands near 2200 nm and 2340 nm, and examined 118 spectral bands (400 to 736 nm) in the VNIR spectrum to look for the absorption bands near 500 nm and 660 nm. For each core sample section, an average value of absorption depth was calculated for all pixels of the section, and this average absorption depth was compared with geochemical measurements.

Geochemistry
The dry samples were crushed to 70% less than 2 mm particle size and then riffle-split. A 250 g sub-sample was pulverized to 85% less than 75 μm with ring-mill. Of the pulps 30 g was analyzed for gold by fire assay atomic absorption. After aqua regia digestion, another 1 g sub-sample of the pulps was analyzed by ICP-MS.

Geochemistry
The dry samples were crushed to 70% less than 2 mm particle size and then riffle-split. A 250 g sub-sample was pulverized to 85% less than 75 µm with ring-mill. Of the pulps 30 g was analyzed for gold by fire assay atomic absorption. After aqua regia digestion, another 1 g sub-sample of the pulps was analyzed by ICP-MS.
Multivariate principal component analyses (PCA) [42,43] were performed on the element concentration data and hyperspectral data to show the variability of different elements and affinity of elements. Eigenvectors of element concentrations (vectors showing correlation coefficients) and principal component scores of samples (points showing the linear combinations of eigenvectors) are plotted in bi-plots; the closer the vectors or dots are to each other, the closer affinity or similarity they have.

Hyperspectral Imaging
An average was calculated for all the spectral bands and all the core sample pixels with the same lithology ( Figure 6; for lithology classification see the text in this section); the mean spectral reflectance curves show the major spectral characteristics of core samples. The mean spectral reflectance curves in the SWIR spectrum ( Figure 6A,C) show two strong absorption features at 2340 nm and 2200 nm. The depths of these absorption features were automatically modeled, which represent the relative abundances of calcite and clay minerals, respectively. Possible interference of chlorite with calcite because of chlorite's absorption near 2340 nm was ruled out due to the lack of absorption from chlorite near 2250 nm ( Figure 6A,C). The distinctive duplet absorptions of kaolinite at 2165 nm and 2200 nm were also not observed ( Figure 6A,C). As a result, the clay minerals should be illitic or smectitic. Illite and smectite have similar overlapping Al-OH absorption with each other; this study does not attempt to distinguish between the two species. With the presence of abundant calcite, detection of possible interference of the Mg-OH absorption near 2300 nm was not possible because this weak absorption would be masked out by the strong asymmetric absorption of CO 3 2− represented by two Gaussians at 2340 and 2300 nm in the modified Gaussian model. The Mg-OH feature only interferes with the Gaussian at 2300 nm and does not affect the abundance quantification based on the Gaussian at 2340 nm. All three mean spectral reflectance curves in the VNIR spectrum ( Figure 6B,D) displayed the strong ferric iron absorption near 500 nm without an obvious absorption near 670 nm. The lack of the absorption feature near 670 nm indicated the presence of secondary oxidized ferric iron in ferrihydrite and the lack of primary ferric iron in cementing goethite and hematite [44]. The absorption depth at 500 nm was modeled to represent the relative abundance of the ferric iron. It is shown that the upper sections were mostly deficient in calcite and the lower sections were rich in calcite ( Figure 7A). Compared with the stratigraphic column and the drilling records, these sections were most probably the siliciclastic sediments of the basal Claron Formation on the top and the underlying Callville Limestone on the bottom. The unconformity separating them was probably near the third core column from the left side of Scan3 ( Figure 7A). VNIR spectrum ( Figure 6B,D) displayed the strong ferric iron absorption near 500 nm without an obvious absorption near 670 nm. The lack of the absorption feature near 670 nm indicated the presence of secondary oxidized ferric iron in ferrihydrite and the lack of primary ferric iron in cementing goethite and hematite [44]. The absorption depth at 500 nm was modeled to represent the relative abundance of the ferric iron. It is shown that the upper sections were mostly deficient in calcite and the lower sections were rich in calcite ( Figure 7A). Compared with the stratigraphic column and the drilling records, these sections were most probably the siliciclastic sediments of the basal Claron Formation on the top and Above the unconformity, the calcite abundance in the basal Claron Formation is generally very low, except for a section of high calcite content in the fifth and sixth core column from the left side of Scan2 ( Figure 7A). Calcite abundance is variable in Callville Limestone, including some spots of high calcite concentrations in sections of mostly low calcite content. Clay mineral abundances are significantly higher and display more variability in the basal Claron Formation than in the Callville Limestone ( Figure 7B). Within the siliciclastic sections of the basal Claron Formation, the samples with medium calcite content usually are higher in clay contents. In contrast, within the sections of higher calcite abundance in the Callville Limestone, the clay content is lower. The carbonate-rich section in the basal Claron Formation also shows lower clay content than the siliciclastic sections. Ferric iron abundance is generally higher in the basal Claron Formation than in the Callville Limestone ( Figure 7C). Most of the core samples above the section of high calcite content in Scan 2 ( Figure 7B) display much higher ferric iron content than other samples ( Figure 7C), which is consistent with the red-yellow colors in the VNIR true-color image ( Figure 4A). Some core samples in the Callville Limestone display higher ferric iron content.
Remote Sens. 2019, 11, x FOR PEER REVIEW 10 of 20 the underlying Callville Limestone on the bottom. The unconformity separating them was probably near the third core column from the left side of Scan3 ( Figure 7A). Above the unconformity, the calcite abundance in the basal Claron Formation is generally very low, except for a section of high calcite content in the fifth and sixth core column from the left side of Scan2 ( Figure 7A). Calcite abundance is variable in Callville Limestone, including some spots of high calcite concentrations in sections of mostly low calcite content. Clay mineral abundances are significantly higher and display more variability in the basal Claron Formation than in the Callville Limestone ( Figure 7B). Within the siliciclastic sections of the basal Claron Formation, the samples with medium calcite content usually are higher in clay contents. In contrast, within the sections of higher calcite abundance in the Callville Limestone, the clay content is lower. The carbonate-rich section in the basal Claron Formation also shows lower clay content than the siliciclastic sections. Ferric iron abundance is generally higher in the basal Claron Formation than in the Callville Limestone ( Figure 7C). Most of the core samples above the section of high calcite content in Scan 2 ( Figure 7B) display much higher ferric iron content than other samples ( Figure 7C), which is consistent with the red-yellow colors in the VNIR true-color image ( Figure 4A). Some core samples in the Callville Limestone display higher ferric iron content.

Geochemistry
Fire assay metallurgy measured gold concentrations for the core sections, and fifty-one (51) element concentrations were measured with ICP-MS (see Table S1 in supplemental data). Gold (Au)

Geochemistry
Fire assay metallurgy measured gold concentrations for the core sections, and fifty-one (51) element concentrations were measured with ICP-MS (see Table S1 in supplemental data). Gold (Au) concentrations reported from ICP-MS had fewer significant digits than measurements by the fire assay, so the values from the fire assay were used in the analyses. Concentrations of boron (B), germanium (Ge), indium (In), niobium (Nb), rhenium (Re), tantalum (Ta), and titanium (Ti) were often below the detection limit and were therefore not reported.

Gold Mineralization in Carlin-Type Deposits
Carlin-type gold deposits form several spatial trends spanning the central Great Basin, northern Nevada, and central Nevada, and are characterized by concentrations of very finely micrometer to sub-micrometer sized disseminated arsenian pyrite grains found in structurally controlled, silty, carbonaceous, and calcareous rocks [1]. Regionally, after the flattening of the subducting Farallon plate under the North America plate around 65 Ma [45], extension prevailed in the Great Basin, and the rollback or delamination of the slab renewed magmatism in Eocene and Oligocene that swept the Great Basin southwestward from northern Nevada and Idaho to southern Nevada, southwest Utah, and north Arizona [2,46]. The dehydration of the slab, upwelling of asthenosphere, and the

Gold Mineralization in Carlin-Type Deposits
Carlin-type gold deposits form several spatial trends spanning the central Great Basin, northern Nevada, and central Nevada, and are characterized by concentrations of very finely micrometer to sub-micrometer sized disseminated arsenian pyrite grains found in structurally controlled, silty, carbonaceous, and calcareous rocks [1]. Regionally, after the flattening of the subducting Farallon plate under the North America plate around 65 Ma [45], extension prevailed in the Great Basin, and the rollback or delamination of the slab renewed magmatism in Eocene and Oligocene that swept the Great Basin southwestward from northern Nevada and Idaho to southern Nevada, southwest Utah, and north Arizona [2,46]. The dehydration of the slab, upwelling of asthenosphere, and the extensional environment enabled formation of hydrous, S-and Au-bearing, high-K, calc-alkaline magma with elevated Au/Cu ratio [47,48], which then released CO 2 -, H 2 S-, and Au-bearing aqueous fluid at a significantly deeper depth (about 10 km) than typical porphyry Cu-Au and associated deposits [2,49]. The fluid ascended along high-angle faults, and released vapor with high S/Fe ratios, which was trapped in permeable and reactive rocks by impermeable rocks, and precipitated gold in micro-sized arsenian pyrite grains, forming typical Carlin-type gold deposits. The Goldstrike district in southwest Utah is near the southeastward extension of the 25 million years magmatic front [2] with an active tectonic setting and high angle faults (Figure 1). Therefore gold mineralization may result from similar geologic processes to the Carlin-type deposits.
The host rock alteration of Carlin-type deposits is typically manifested by decarbonatization, argillization, silicification and/or jasperoid, fine-grained disseminated sulfide minerals and oxidation of sulfide minerals, remobilization and/or addition of carbon, and late-stage barite and/or calcite veining, with incipient collapse brecciation that enhances the migration of mineralization fluids [1,11]. Carlin-type deposits are typically stratiform, with mineralization localized with specific favorable stratigraphic units.

Stratigraphic Control
Gold concentrations measured with fire assay metallurgy were used to colorize outlines of scanned core sections and then compared with the mapped calcite and clay abundances (Figure 7) to infer the influences or patterns of mineralogy on gold mineralization. Gold concentrations are not homogeneous throughout whole core sections. Therefore necessary signals may be missed from sampling. However, geochemical analyses cannot sample small enough areas to be comparable to the resolution of hyperspectral imaging. Nevertheless, gold concentration data can show the general variations of the Au-bearing fluid flow patterns.
Gold mineralization is significantly influenced by stratigraphic and structural control in Carlin-type deposits. As shown in Figure 1 the oblique core in this study cut across a fault, which may have facilitated fluid flow. Rocks of low porosity and permeability act as aquitards to prevent fluid from ascending and the Au-bearing fluids react with the permeable rocks below aquitards to precipitate disseminated gold. The calcite-rich section in Scan2, the basal Claron Formation is such an aquitard due to its lower permeability compared with siliciclastic rocks below. Gold concentrations in the calcite-rich section were very low, then very high in the section below that, and decreasing downward ( Figure 9B). Those samples with medium calcite content also showed relatively low gold concentrations. The combination of low permeability carbonate aquitard and high permeability siliciclastic rocks below in Carlin-type deposits were similar to the seal and reservoir rocks in conventional oil and gas industry. Exploration can be focused on such stratigraphic combinations near high-angle faults.

Mineralogical Alterations
Macroscopic rock alterations that are reported to be related to gold mineralization in the Goldstrike district as well as in the Carlin-type deposits include silicification, decarbonatization, argillization, and pyrite oxidation [1,11]. Quartz, chert, and amorphous silica are all not spectrally active in the visible to the short-wave infrared spectrum, only in the thermal infrared, so could not be studied by the hyperspectral cameras used in this study. Decarbonatization stands for the removal of carbonate, so the decline or diminishment of CO 3 2− absorption near 2340 nm represents decarbonatization. Argillization stands for the addition of argillic minerals (most commonly illite and kaolinite), so the increase or appearance of Al-OH absorption near 2200 nm represents argillic alteration. These argillic minerals may come from alterations of felsic minerals or the ore fluids. The basal Claron Formation is sandstone or conglomerate sandstone, which may have argillaceous material, but without a lateral comparison with unaltered strata, it is hard to confirm the argillic alteration in a single core. More cores or outcrops may be helpful. However, the authors only had limited access. On the other hand, Callville Limestone rarely has argillaceous components [50,51], and the detection of Al-OH absorption is interpreted to represent argillic alteration. Pyrite and arsenian pyrite in the strata can be oxidized into ferric iron minerals after mineralization, which is especially common in pyrite-rich silty limestones or limey siltstones [1], so the detection of ferric iron absorption near 500 nm may represent pyrite oxidation.

Mineralogical Alterations
Macroscopic rock alterations that are reported to be related to gold mineralization in the Goldstrike district as well as in the Carlin-type deposits include silicification, decarbonatization, argillization, and pyrite oxidation [1,11]. Quartz, chert, and amorphous silica are all not spectrally active in the visible to the short-wave infrared spectrum, only in the thermal infrared, so could not be studied by the hyperspectral cameras used in this study. Decarbonatization stands for the removal of carbonate, so the decline or diminishment of CO3 2− absorption near 2340 nm represents decarbonatization. Argillization stands for the addition of argillic minerals (most commonly illite and kaolinite), so the increase or appearance of Al-OH absorption near 2200 nm represents argillic alteration. These argillic minerals may come from alterations of felsic minerals or the ore fluids. The basal Claron Formation is sandstone or conglomerate sandstone, which may have argillaceous material, but without a lateral comparison with unaltered strata, it is hard to confirm the argillic alteration in a single core. More cores or outcrops may be helpful. However, the authors only had limited access. On the other hand, Callville Limestone rarely has argillaceous components [50,51], and the detection of Al-OH absorption is interpreted to represent argillic alteration. Pyrite and arsenian pyrite in the strata can be oxidized into ferric iron minerals after mineralization, which is especially common in pyrite-rich silty limestones or limey siltstones [1], so the detection of ferric iron absorption near 500 nm may represent pyrite oxidation.
Much of the scanned Callville Limestone has undergone various degrees of decarbonatization. Patches of high calcite content surrounded by areas of lower calcite content can be seen below the unconformity in Scan3, are prevalent in Scan4, and in some columns in Scan5. Some sections have Much of the scanned Callville Limestone has undergone various degrees of decarbonatization. Patches of high calcite content surrounded by areas of lower calcite content can be seen below the unconformity in Scan3, are prevalent in Scan4, and in some columns in Scan5. Some sections have gone through strong decarbonatization removing all the calcite. Decarbonatization may reflect more active gold mineralization, as the gold concentrations are relatively higher in these sections ( Figure 9B). Some of the scanned Callville Limestone has also undergone various degrees of argillization. Samples of medium clay contents can be seen in Figure 9C, with argillization corresponding to more active gold mineralization and relatively higher gold concentrations ( Figure 9D). Some samples in Callville Limestone showed higher ferric iron ( Figure 9E), which might correlate with pyrite oxidation after mineralization with higher gold concentrations ( Figure 9F). These interpretations were supported by PCA of absorption depths together with element concentrations (Figure 10). Due to the significant differences in lithology and alteration patterns, measurements of seventeen (17) siliciclastic and seventeen carbonate rock sections (including the carbonate aquitard section in the basal Claron Formation) were processed separately. The average pixel value of absorption depths and 12 selected elements were analyzed by PCA since only seventeen sections were available. Analyses in siliciclastic rocks showed that calcite, clay, and ferric iron depths all had minimal affinities with gold ( Figure 10A; See Table S2 in supplemental data). This confirmed that decarbonatization and argillization in siliciclastic rocks could not be easily mapped from a single core and correlate with gold mineralization. However, stratigraphic control by different lithologies played a major role (Figure 9). It was also shown that calcite and clay absorption depths were close to each other, supporting the observation that samples with medium calcite content usually were higher in clay contents. On the other hand, analyses in carbonate rocks showed that both clay and ferric iron depths had close affinities to silver and a little less affinity to gold, while calcite depth showed very limited affinity to gold, silver, and clay depth ( Figure 10B) (See Table S4 in supplemental data). These facts confirmed that decarbonatization, argillization, and pyrite oxidation in carbonate rocks could be mapped from core samples, and correlated with gold mineralization. The identification of these mineralogical alterations could be used as a classifier for ore grades; intense alterations might correlate with higher grades, and weak alterations correlate with lower grades. seventeen sections were available. Analyses in siliciclastic rocks showed that calcite, clay, and ferric iron depths all had minimal affinities with gold ( Figure 10A; See Table S2 in supplemental data). This confirmed that decarbonatization and argillization in siliciclastic rocks could not be easily mapped from a single core and correlate with gold mineralization. However, stratigraphic control by different lithologies played a major role (Figure 9). It was also shown that calcite and clay absorption depths were close to each other, supporting the observation that samples with medium calcite content usually were higher in clay contents. On the other hand, analyses in carbonate rocks showed that both clay and ferric iron depths had close affinities to silver and a little less affinity to gold, while calcite depth showed very limited affinity to gold, silver, and clay depth ( Figure 10B) (See Table S4 in supplemental data). These facts confirmed that decarbonatization, argillization, and pyrite oxidation in carbonate rocks could be mapped from core samples, and correlated with gold mineralization. The identification of these mineralogical alterations could be used as a classifier for ore grades; intense alterations might correlate with higher grades, and weak alterations correlate with lower grades. Besides the macroscopic alterations, some trace elements like As, Hg, and Sb are associated with gold mineralization, and they form sulfide minerals like orpiment, realgar, cinnabar, and Besides the macroscopic alterations, some trace elements like As, Hg, and Sb are associated with gold mineralization, and they form sulfide minerals like orpiment, realgar, cinnabar, and stibnite. These minerals appeared in trace amount, and no investigation was performed on these minerals.

Implications
This study showed an example of using hyperspectral imaging to identify the optimum stratigraphic combination of limestone above and siliciclastic rocks below, as well as mineralogical alterations including decarbonatization, argillization, and pyrite oxidation that are related to gold mineralization. Such mineralogical information may shed light on the geologic mechanism of mineralization, and help in the determination of mining interests with similar stratigraphic and structural conditions, as well as help in the classification of ore grades based on the alteration patterns. Without expenses in chemical consumables and procedure, most of the costs of hyperspectral imaging are for personnel. As a result, hyperspectral imaging may be a cost-effective alternative or complementary method for geochemical methods [52,53]. Besides, the sub-centimeter spatial resolution and detailed mineralogical identification/semi-quantification from hyperspectral imaging are superior to lithologic logging that accompanies exploration drilling. Ground-based hyperspectral imaging, as a new direction in earth sciences, provides high spatial and spectral resolution measurements, fast data collection, sizeable areal coverage, and easy data processing. We envision more studies with hyperspectral imaging in various fields of earth sciences.

Conclusions
Hyperspectral imaging was used to scan a core covering the basal Claron Formation and Callville Limestone in the Goldstrike district, southwest Utah, which is believed to be similar to Carlin-type gold deposits in Nevada with sediment-hosted disseminated gold. This study used spectral modeling to identify and semi-quantify calcite, clay minerals, and ferric iron. Variations in mineralogy are used to identify lithology, as well as decarbonatization, argillization, and pyrite oxidation alterations within the core samples. Compared with metallurgy and ICP-MS geochemical data of the core, this study confirmed the correlation between stratigraphic control as well as mineralogical alterations with gold mineralization. Although the silicification and formation of jasperoids are essential indicators of gold mineralization, silica is not spectrally active in the reflected spectrum and cannot be studied by hyperspectral imaging. This state-of-art technology has excellent potentials in broader applications in the mining industry.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-4292/11/17/1987/s1, Table S1: Element concentrations from fire assay metallurgy and ICP-MS; Table S2: Correlation matrix of absorption depths and element concentrations in siliciclastic rocks; Table S3: Eigenvalues of the correlation matrix of absorption depths and element concentrations in siliciclastic rocks; Table S4: Correlation matrix of absorption depths and element concentrations in carbonate rocks; Table S5: Eigenvalues of the correlation matrix of absorption depths and element concentrations in carbonate rocks.