Multi-scale multi-dimensional microstructure imaging of oil shale pyrolysis using X-ray micro-tomography, automated ultra-high resolution SEM, MAPS Mineralogy and FIB-SEM

(cid:1) Green River oil shale is characterized in 2-D/3-D using complementary techniques. (cid:1) Oil shale pyrolysis can be captured in 3-D using X-ray m CT at the micron- scale. (cid:1) 2-D SEM imaging resolves organic and mineral structures at the micron/nano scale. (cid:1) MAPS mineralogy can quantify the spatial distribution of minerals within oil shale. (cid:1) 3-D FIB-SEM can visualize organic and mineral hosted porosity and connectivity. Multi-scale multi-dimensional shale characterization workﬂow. the the of MAPS Mineralogy (Modular Automated Processing and Focused Scanning Microscopy (FIB-SEM). The organic-rich Eocene Green River (Mahogany zone) oil shale is characterized using a multi-scale multi-dimensional workﬂow both before and after pyrolysis. Observations in 2-D and 3-D and across nm- m m-mm length scales demonstrate both heterogeneity and anisotropy at every scale. Image acquisition and analysis using m CT and SEM reveal a microstructure of alternating kerogen-rich laminations interbedded with layers of ﬁne-grained inorganic minerals. MAPS Mineralogy combined with ultrafast measurements reveal min-eralogic textures dominated by dolomite, calcite, K-feldspar, quartz, pyrite and illitic clays along with http://dx.doi.org/10.1016/j.apenergy.2017.05.039 their spatial distribution, augmenting conventional mineral analysis. From high resolution Backscattered electron (BSE) images, intra-organic, inter-organic-mineral, intra- and inter-mineral pores are observed with varying sizes and geometries. By using FIB milling and SEM imaging sequentially and repetitively, 3-D data sets were reconstructed. By setting 3-D gradient and marker-based watershed transforms, the organic matter, minerals and pore phases (including pore-back artifacts) were segmented and visu- alized and the pore-size distribution was computed. Following pyrolysis, fractures from the mm-to- m m scales were observed with preferential propagation along the kerogen-rich laminations and coalescence leading to an interconnected fracture network. The application of these techniques to worldwide oil shale deposits will allow signiﬁcant insights into estimating mechanical and chemical proprieties of oil shale formations for modeling and designing oil shale pyrolysis processes.


Introduction
Global energy demand is set to rise spectacularly in the coming decades as a result of population growth and economic development [1][2][3]. In many parts of the world, focus on energy reliability and security has brought significant attention to the exploration and production of unconventional hydrocarbon deposits transforming energy markets worldwide [4,5]. Oil shale, an organicrich fine-grained sedimentary rock, represents a tremendous and predominantly untapped natural energy resource with many known deposits across the world [6][7][8]. The Lacustrine Eocene Green River Formation, located in Utah, Colorado and Wyoming ( Fig. 1), contains the largest oil shale deposit in the world with an estimated 4.3 trillion barrels of oil originally in place [8,9]. Previous work has characterized the bulk properties for the Green River oil shale formation including mineralogy (XRD), total organic carbon (TOC), thermal maturity (Rock-Eval Pyrolysis), elemental analysis (CHNOS) and oil yield (Fischer Assay) [10][11][12][13][14][15][16]. Kerogen, which constitutes most of oil shale's organic matter, is a highly cross-linked, insoluble, macromolecular solid material that is the dominant source of carbon for hydrocarbon generation in the subsurface [17][18][19]. Pyrolysis of oil shale, a process which involves heating in the absence of oxygen, breaks down the complex kerogen network structure to produce shale oil and natural gas [20][21][22].
Hydrocarbon recovery from thermally immature oil shale takes place principally through ex-situ (surface retorting) and in-situ (within the subsurface geologic formation) technologies. The source rock is heated above pyrolysis temperatures (300-500°C) at which the immobile organic matter is transformed to mobile oil and gas [23][24][25][26][27].
In the pyrolysis extraction of hydrocarbons, both the chemical and physical properties of oil shale change significantly [28][29][30][31][32][33]. Many studies have been conducted on oil shale pyrolysis to identify the effect of fundamental experimental parameters on shale oil yield and quality. This includes pyrolysis temperature [34][35][36], heating rate [35][36][37][38], residence time for pyrolysis reaction [39,40], composition of pyrolysis atmosphere [41,42], particle size [43][44][45] and mineral matrix effects [46,47]. To further enhance our understanding of oil and gas transport phenomena involved in the pyrolysis of oil shale and ultimate recovery, a comprehensive characterization of the variations in petrophysical properties is essential, in particular, the direct visualization and quantification of the organic matter network, mineralogical phases and the evolution of the pore space during pyrolysis.
The complexity of oil shale is demonstrated both in the compositional heterogeneity of the matrix and in the structure of the pore space. In oil shale, the organic matter is tightly bound within a heterogeneous matrix. The inorganic matrix consists of minerals, including clays, carbonates, feldspars, quartz, and pyrite [6,[47][48][49]. Heterogeneity within oil shale formations exists across multiple length scales due to complex sedimentary and diagenetic processes [23,[50][51][52]. This heterogeneity renders oil shale rocks difficult to characterize petrophysically (organic matter type and distribution, mineral composition, pore volume, pore size distribution, geometry and connectivity) and limits our ability to distribute these properties in oil shale pyrolysis models. Pores within oil shale rocks are orders of magnitude smaller (nanometer scale) than those in conventional carbonate and sandstone samples (micrometer scale). The nano-scale porosity associated with organic matter and how well that organic matter is connected is crucial for the creation of flow channels to allow the generated hydrocarbon fluids to escape. Thus far, characterizing oil shales has proven to be a complex and challenging task, hindered by the lack of tools and techniques to investigate the broad structural and mineralogical heterogeneities over many length scales [53][54][55][56][57][58][59][60][61]. Moreover, the development of pore networks and microfractures within the fine-grained oil shale microstructure during pyrolysis is not well understood and remains largely unknown [62]. Hence, it is vital to comprehensively characterize organic connectivity, mineralogical heterogeneity and the pore space before and after pyrolysis across multiple length scales to determine the microstructural controls on pore space evolution and hydrocarbon flow behavior.
Conventional laboratory porosity measurements such as mercury injection capillary pressure (MICP) and gas adsorption Brunauer, Emmett and Teller (BET) methods provide an indirect approach to investigate pores down to a few nanometers in size [63][64][65][66]. However, these macroscopic averaging methods only provide data related to the connected porosity, do not involve direct observation of individual pores, and are based on simplified models which do not reflect the complexity of the pore space [63,[67][68][69]. Previous experimental studies have used nitrogen adsorption-desorption isotherms to estimate the pore properties of Green River (Western USA) [53], New Albany (Eastern US) [54] and Huadian (China) oil shales [55][56][57][58] before and after pyrolysis. However, the majority of these studies used powdered oil shale samples and an investigative method that lacks direct information about pore geometry, isolated pores and the association of porosity to the surrounding microstructure and mineralogy. Moreover, in immature oil shale, naturally occurring porosities are negligible, although porosity may exist to some degree in formations where fractures, faults, or other structural defects have occurred. Overall, it is believed that a substantial proportion of pores in immature oil shale are isolated or inaccessible to gases and mercury even at high pressures [52].
A wide range of complementary imaging techniques can be used to directly visualize and quantify oil shale structures across multiple scales. With optical thin sections, a useful initial visualization of the microstructure can be acquired to assess the mineralogy and microscopic features in 2-D. However, with traditional optical petrography, the resolution is limited to approximately 0.23 mm (diffraction barrier of visible light) which prevents the fine-grain characteristics within oil shales from being characterized [70]. To visualize and quantify the pores, inorganic mineral grains, organic matter (kerogen) and fine clay structures within oil shale, an imaging technique that resolves features from the optical regime down to the nanometer scale is required. Scanning electron microscopy (SEM) is well-suited to study fine-grained sedimentary rock features from the micrometer to nanometer scales [57,[71][72][73][74][75][76]. For shale samples, mechanical polishing is not suitable, as this technique tends to introduce artifacts and destroy the fine microstructure (e.g. abrasion marks, smearing grain boundaries, embedment of grit into surface of sample) [74,77,78] . With advanced SEM instruments, even a minimal amount of damage can limit our abil-ity to fully resolve or analyze the surface of an oil shale sample. To overcome this difficulty and obtain ultra-flat samples, argon (Ar) ion beam milling can be used resulting in high-quality SEM images that reveal nano-scale microstructures in 2-D with minimal artifacts [74,[78][79][80]. However, this imaging method cannot collect internal information such as 3-D pore, mineral and organic network structures. Some studies have used multiple-point statistics from 2-D images to generate 3-D networks of sandstone and carbonate samples [81][82][83][84], however, this becomes a significant challenge for heterogeneous anisotropic media involving complex microstructures such as in oil shales.
Automated image acquisition for SEM systems has become increasingly common in recent years [85][86][87][88]. In general, this approach captures single images, or groups of connected images, for image analysis purposes. More recently, examples of automated image acquisition and stitching, designed to generate high-resolution large format images have been developed [86,89]. These systems typically produce images that are several gigabytes or larger in size, and can often use a variety of detector types such as backscattered electron (BSE), secondary electron (SE) and cathodoluminescence (CL) [86]. Recently, the use of automated mineralogy in the form of QEMSCAN technology (Quantitative Evaluation of Minerals by Scanning Electron Microscopy) has allowed users to non-destructively quantify the amount, type and distribution of minerals present in a sample [90][91][92][93]. Quantitative mineral identification is performed through particle identification by BSE intensity and X-ray analysis based on energy dispersive spectrometry (EDS) and a mineral database or Species Identification Protocol (SIP). Automated mineralogy determination has provided significant advances in mineralogical mapping at high resolution allowing a detailed evaluation far beyond the resolution of conventional thin section petrography and bulk compositional data obtained from X-ray Diffraction (XRD) and X-ray Fluorescence (XRF) analysis. More recently, MAPS Mineralogy (Modular Automated Processing System), a new Automated SEM-EDS technology, has been developed which does not constrain the number of minerals per pixel, unlike a standard QEMSCAN analysis. Rather, it allows multiple minerals to be reported within a single pixel and avoids the need to manage complex mineral chemistry within the analytical software.
Techniques that allow direct 3-D imaging and quantification are of significant benefit in characterizing the evolution and connectivity of the pore space during oil shale pyrolysis. In recent years, Xray micro-computed tomography (mCT) and Focused Ion Beam Scanning Electron Microscopy (FIB-SEM) have emerged as powerful 3-D imaging tools to capture the internal structures of geological samples, in particular being effective at studying complex pore-scale processes [62,76,[94][95][96][97][98][99][100][101][102][103]. For oil shales, in addition to determining pore space characteristics, a 3-D approach allows us to obtain important information on the spatial distribution of organic matter and constituent inorganic minerals. Quantitative analysis can provide useful insights by characterizing component volume fractions and fundamental pore space geometric attributes, including pore size, shape, tortuosity and connectivity. X-ray mCT offers several advantages: it is non-destructive, provides 3-D imaging, achieves high spatial resolutions at scales down to the micron level, gives good contrast between phases, and is adaptable to many types of experimental procedures. X-ray mCT has been applied to characterize oil shale samples from the United States (Green River) [59,62,104,105], China (Fushun) [60] and Australia (Queensland) [106]. Tiwari et al. [107] characterized pore structures before and after pyrolysis based on 42 mm voxel size scans reporting pores as large as 500 mm after pyrolysis (500°C). More recently, Saif et al. [16] studied the change in pore structure in the Mahogany oil shale with increasing pyrolysis temperature (300-500°C) at 12 mm and 2 mm voxel sizes. The results showed a significant increase in anisotropic porosity associated with pyrolysis between 400 and 500°C with the formation of micron-scale connected pore channels developing principally along the organic-rich lamellar structures. Moreover, dynamic imaging of oil shale pyrolysis using synchrotron X-ray tomography was performed on a Green River (Mahogany Zone) oil shale sample presenting a direct visualization of the temporal evolution of the pore space during pyrolysis [62]. Saif et al. [62] reported micron-  scale disconnected pores at 390°C; with porosity increasing dramatically between 390°C and 400°C where the vast majority of the pore space became connected.
The complex microstructure of oil shales spans multiple orders of magnitude from centimeter scale laminations to nanometer scale pores within the organic material and inorganic minerals. X-ray mCT can resolve features at the micrometer to millimeter scales, but lacks the capability to determine nano-scale structures such as the pore space in immature oil shale samples and the developing pore network during pyrolysis. Recent developments in FIB-SEM systems allow for direct 3-D visualization of a variety of samples at the nanometer-scale resolution [101,[108][109][110][111][112][113]. The system allows the user to mill (with the FIB) and image (with the SEM) in a region of interest (ROI) to obtain a sequence of 2-D cross-sectional images which are then reconstructed to generate a digital 3-D visualization of the analyzed volume. The FIB-SEM technique enables structural features within fine-grained samples to be resolved at the nano-scale including porosity, grain morphologies and organic networks.
In this study, recent developments in digital rock imaging are presented by examining the organic-rich Green River (Mahogany zone) oil shale before and after pyrolysis. The significant challenges related to visualizing and quantifying petrophysical properties of oil shales are addressed by using a workflow which combines com- plementary imaging techniques that cover multiple length scales to comprehensively characterize and correlate distributions of oil shale microstructural and mineralogical properties (Fig. 2). We qualitatively and quantitatively characterize the Mahogany oil shale samples using X-ray mCT, ultra-high resolution automated SEM, MAPS Mineralogy and FIB-SEM. With X-ray mCT, we extend previous work by Saif et al. [62] to image at 0.8 mm voxel size resolving organic matter (kerogen) networks and the pore space in 3-D at the critical ex situ pyrolysis transition temperature range of 380-420°C. We apply MAPS Mineralogy using automated SEM imaging and energy dispersive X-ray spectroscopy (EDS) combined with novel software to generate mineral maps that contain micron-scale textural information as well as pertinent data on the spatial distribution of inorganic minerals in oil shale. Using FIB-SEM, we resolve in 3-D the complex heterogeneous microand nano-structures present in the oil shale samples providing new direct visualizations of kerogen and pore connectivity in unpyrolyzed and pyrolyzed oil shale samples. In addition, estimates of organic volumes, porosity, and pore size distribution have been quantified for both 2-D and 3-D data sets. This multi-scale multi-dimensional workflow provides a valuable approach in integrating microstructural and mineralogical oil shale data with exceptional fidelity.

Samples
Oil shale samples were obtained from an outcrop of the organicrich Mahogany zone of the Green River Formation (Uinta Basin, Utah). The Eocene Green River Formation represents a classic model for lacustrine source rock deposition and is the common reference rock type for Type I kerogen [114,115]. The Mahogany zone (R-7) is the richest oil shale horizon which is a primary target for shale oil production due to its high oil yield which can exceed 250 liters per tonne of rock recovered as liquid fuel by pyrolysis [8]. The composition of the sample used in this study has been   images are recorded in sequence and subsequently stitched to generate one single image. The full surface is also mapped using energy-dispersive (EDS) X-ray analysis with a user-defined spacing. The EDS X-ray spectrum for each analysis point is compared to a database of known mineral species. This process is automated on a pixel by pixel basis to create high resolution mineral maps. (B) MAPS real-time tile acquisition and stitching system with a programmable overlap to generate high resolution images of gigapixel size.
comprehensively characterized by Saif et al. [16]. In this study, the oil shale samples were cut and prepared normal to the bedding for X-ray mCT, ultra-high resolution SEM, MAPS Mineralogy and FIB-SEM, with investigations of both unpyrolyzed and pyrolyzed samples. Samples for each imaging technique were pyrolyzed under vacuum pressure at 0.1 kPa absolute. A heating rate of 10°C/min was used to achieve the reaction temperature of 500°C where the samples were held for one hour. A thermocouple measured sample temperature and a proportional integral differential (PID) controller regulated the temperature to ±1°C. Following pyrolysis at 500°C, the samples were cooled to ambient temperature and returned to ambient pressure conditions.
3. 3-D Micron-scale visualization and quantification of organic matter (Kerogen) and pore structure evolution during oil shale pyrolysis using X-ray Micro-tomography (mCT) A Zeiss Versa XRM-500 X-ray Microscope (Carl Zeiss, CA, USA) was used to image the Green River (Mahogany Zone) oil shale samples. Samples of size 1.0 mm diameter by 5.0 mm length were mounted on a rotary stage, irradiated by a micro-focus polychromatic X-ray source and imaged in transmission onto a detector. The X-rays transmitted through the sample hit the scintillator crystals to give off visible light, which is then focused by the optical objective lens and converted into a digital image by the visible light charge-coupled device (CCD). A series of images (projections) were taken at incrementally spaced angles over 360°as the sample is rotated. The mCT scans of the oil shale samples were taken with the X-ray source operating at 50 keV and acquired at 0.8 mm voxel sizes with 2001 projections collected on a 2000 Â 2000 pixel CCD detector with the pixel size binned to 1000 Â 1000 pixels.  The tomograms were reconstructed into 3-D volumes using proprietary software (Zeiss XMReconstructor, Carl Zeiss X-ray Microscopy Inc., Pleasanton, CA) based on filtered backprojection (FBP) algorithms and image processed using Avizo 9.0 software (FEI, Hillsboro, OR, USA) and MATLAB (MathWorks, Inc., Natick, USA). Following image processing, the total image size for the 0.8 lm scans was 964 Â 993 Â 970 voxels. The images (16-bit unsigned) were represented using unsigned grayscale value integers in the range 0-65,535. The images were processed using a non-local means edge preserving filter [116,117] and segmented using global thresholding to generate a binarized representation of the organic matter network (Fig. 3).
Figs. 3A and B display a 2-D image of a 3-D reconstructed grayscale image acquired at a 0.8 mm voxel size. The grayscale values for each pixel correspond to X-ray attenuation which varies as a function of density and atomic number. The less dense kerogen-rich layers within the oil shale sample are displayed as the darker colored regions, while the denser mineral-rich layers appear lighter. Figs. 3C and D illustrate the segmented organic matter revealing a 99.8% well-connected organic network in the analyzed 3-D oil shale volume.
To study the impact of pyrolysis temperature on the evolution of the pore space, the Green River (Mahogany Zone) oil shale sample was imaged before and after pyrolysis from 20°C to 420°C in 10°C increments. The sample was heated ex situ in an electrically heated furnace (Carbolite ELF 11/14B) at each set temperature point and then allowed to cool to ambient temperature before being placed in a sample holder and imaged using the Zeiss Versa XRM-500 mCT scanner.
Between 20°C and 350°C no change in porosity was identified in the oil shale sample at the 0.8 mm voxel size (Figs. 4A and B). In Fig. 4C, at 380°C, the first micro-fractures are observed parallel to the organic-rich laminations resulting in a sample porosity of 5.6%. A further increase in temperature to 390°C resulted in the propagation of the major fracture along the kerogen-rich bedding and the temporary closure of the minor fracture due to local compressive stresses (Fig. 4D). At 390°C the computed sample porosity was 12.5%. At 400°C (Fig. 4E) and 420°C (Fig. 4F), the growth of multiple major and minor pores and micro-fractures continues, which is accompanied by a volume increase, resulting in a porosity of 21.1% and 23.3% respectively. The development of these pores and fractures within the organic-rich sample is due to the thermallyinduced breakdown of kerogen into hydrocarbon fluids. The increase in fluid pressure during the transformation process causes local stress concentrations within the organic matter and at a critical stress, fractures start to nucleate and propagate. With increasing temperature, growing micro-fractures coalesce leading to an interconnected fracture network within the pyrolyzed oil shale sample.
The results obtained in this study were compared with those obtained in [62], where dynamic imaging of oil shale pyrolysis using a synchrotron X-ray source with a 2 mm voxel size was performed on a Green River (Mahogany Zone) sample with a mixed composition of organic-rich and organic-lean zones (Fig. 5 Oil shale porosity for the Green River (Mahogany Zone) as a function of pyrolysis temperature. A comparison is made between static laboratory based analysis, presented here, and dynamic synchrotron mCT data [62].). The comparison between ex situ (static mCT) and in situ (dynamic synchrotron) results reveal a similar trend in the evolution of porosity with temperature, with a resolvable change in porosity taking place between 380 and 400°C.

2-D Micro-and nano-scale visualization and quantification of oil shale pyrolysis using Ultra-high resolution SEM and MAPS Mineralogy
Ultra-high resolution Scanning Electron Microscopy (SEM) was performed on Green River (Mahogany Zone) oil shale samples to characterize the complex distribution and association of organic, mineral, and pore phases at the nano-to-micro scales. In this study, Mahogany zone samples of about 18 mm Â 14 mm Â 5 mm were initially hand-polished using sandpaper (600, 1200 and 2000 grit) and diamond lapping film (6 mm, 3 mm and 1 mm). The resulting Mahogany oil shale sample surfaces were then attached to 25 mm diameter pin-type stubs using carbon tape and then coated with less than a 100-Å thick elemental carbon film via a vacuum evaporator (Quorum Q150T) to mitigate charge buildup (Fig. 6A). Each sample surface was then argon ion milled using a Fischione 1060 SEM Mill under a 2°incident angle and 360°sample rotation at 6 keV for 22 h. In this process, argon is ionized and accelerated towards the sample surface where the impinging ions sputter material from the surface at a controlled rate (Fig. 6B). The oil shale polished sample surfaces were then examined using a Quanta FEG 650 SEM (FEI) equipped with two Bruker XFlash 6 series detectors for energy dispersive spectroscopy (EDS).
MAPS (Modular Automated Processing System) software was used for automated acquisition of high resolution large-format mosaic images. The surface area of interest was divided into a grid format with 19 Â 16 tiles and a sequence of BSE images were acquired in real-time with a 10% overlap. This resulted in images of 34,733 Â 28,839 pixels for sample 1 (unpyrolyzed) and 35,417 Â 29,216 pixels for sample 2 (pyrolyzed) each with a pixel size 500 nm captured at an energy of 15 keV. The stitched mosaic of images allows for the systematic investigation of fine-grained microstructures from the nm to the mm scales. The full surface was also mapped using automated EDS analysis with an X-ray spacing of 10 mm and acquisition time of 8 ms per pixel (Fig. 7).
Electron beam analysis points return secondary X-ray spectra that are compared with a large database, allowing the assignment of each analysis point to a specific mineral or elemental category. Nanomin software (FEI, Hillsboro, OR, USA) was used to analyse SEM-EDS data, and to map the chemical, mineral composition and textural features in the rock samples at an X-ray spacing of 10 mm and an acquisition time of 8 ms per pixel. Regions of interest were further explored by acquiring ultra-high resolution mosaic BSE images at beam energies of 3-5 keV generating images with 3, 10, 20, 30 and 50 nm pixel sizes. Fig. 8 shows grayscale mosaic BSE images that display compositional contrast in the heterogeneous Green River (Mahogany Zone) oil shale sample. Automated tiling and stitching BSE images into mosaics enable pore-scale resolution over the entire sample surface in one image allowing the simultaneous retrieval of microstructural fabric heterogeneity and nanometer-scale structures without losing correlation. The gigapixel size mosaics expose higher resolution features within a deceased field of view to uncover structural features that constitute the rock fabric ( Fig. 8A-F). In these SEM images, compositional contrast is recorded as BSE intensity which is a function of the mean atomic number of the specimen volume that interacts with the electron beam. Kerogen, which is predominantly composed of relatively low atomic number elements in the form of carbon, hydrogen, and oxygen, translates to darker image pixels. This organic matter is dispersed within a matrix of inorganic minerals such as iron-rich pyrite, calcium-rich dolomite, and silicon-rich quartz, which appear as brighter (light gray to white) image pixels due to higher atomic number elements. The Green River oil shale exhibits a layered structure of alternating kerogen-rich and lean laminations at both the millimeter (Fig. 8A) and micron scales (Figs. 8B-D).
Although the BSE intensity of kerogen is low, void spaces, such as pores and fractures, display even lower BSE intensity, allowing for pore identification and quantification in BSE images of oil shale samples (Figs. 9 and 10). The EDS performed on the oil shale surfaces combined with the mineral identification software package NanoMin confirm that the darker areas are organic kerogen, with the lighter gray inorganic mineral matrix being primarily com-posed of varying amounts of dolomite, calcite, quartz and clays (Fig. 11).
Figs. 9 and 10 show pores present in the oil shale sample which range from a few nanometers to several micrometers in diameter and from simple to complex geometries. These pores form by both depositional and diagenetic processes as well as through multiple stages related to deposition, compaction, cementation, and dissolution. In the Green River (Mahogany Zone) oil shale sample four types of porosity are observed: intra-organic pores, referring to pores bounded by organic matter; inter-organic-mineral pores, referring to pores between organic matter and minerals; intramineral pores, referring to pores fully bounded within mineral grains and inter-mineral pores, referring to pores lying between inorganic minerals. For the Mahogany oil shale sample, we observe that organic matter hosted pores, rather than mineral-hosted pores, were the dominant contributors to total porosity and are predominantly sub-mm in size with typically irregular geometries.
To assess the impact of BSE pixel size on segmented organic, mineral and pore phases, a 50 mm (height) Â 60 mm (width) region of interest (ROI) within the oil shale sample was selected consisting of a range of mineral grains, substantive organic matter and the presence of pores (Figs. 12A-C). BSE images were taken at pixel sizes of 500 nm, 50 nm and 10 nm. For each grayscale image, global threshold segmentation was applied to generate a three-phase seg-mented image constituting organic matter, minerals and pore pores (Figs. 12D-F). At a 500 nm pixel size, the computed organic and mineral relative quantities were 28.1% and 71.9% respectively, with no pores identified at this pixel size. At a 50 nm pixel size, the percentage of resolvable organic matter increased significantly to 40.7%, and mineral relative amounts decreased to 58.8%. Both organic-hosted and mineral-hosted pores were successfully resolved at a 50 nm pixel size with an ROI porosity of 0.5%. At a 10 nm pixel size, the percentage of resolvable organic, mineral and pore phases was computed at 42.2%, 57.1% and 0.7% respectively, indicating a very small change from the relative quantities computed at 50 nm. Figs. 13 and 14 show BSE mosaic images following pyrolysis of the Green River (Mahogany Zone) oil shale at 500°C. The develop-  ment of micro-fracture networks are well defined and occur parallel to the oil shale bedding plane. At high resolutions, we observe an intricate network of well-connected nanopores and fractures surrounding the inorganic mineral grains. Both nano-and microfractures fractures constitute the void spaces generated by the thermal breakdown of the kerogen macromolecular structure that resulted in local overpressure build-up during kerogen maturation and subsequently the preferential propagation of fractures along the kerogen-rich laminations allowing the generated hydrocarbon fluids to escape. In addition to the pore space, at the micron-scale we also observed regions where the organic matter had partially decomposed during pyrolysis as well as regions where kerogen still remains unconverted. This observation illustrates that not all the solid kerogen thermally transforms to fluid hydrocarbons at the same rate or to the same extent which accentuates the complexity in kerogen pyrolysis kinetics.

3-D Nano-scale visualization and quantification of oil shale pyrolysis using focused Ion-beam scanning electron microscopy (FIB-SEM)
The FEI Helios NanoLab 660 DualBeam system was used to obtain 3-D data sets for nano-to micron scale structural character-  ization of the Green River (Mahogany zone) oil shale samples (unpyrolyzed and pyrolyzed). This technology combines automated sequential FIB (Focused Ion Beam) milling and SEM (Scanning Electron Microscope) imaging to create a series of 2-D images (Fig. 15A), which in turn leads to 3-D volume reconstructions. A focused beam of gallium ions (Ga + ) operating at 30 keV and 2.5 nA beam current milled regions of interest in the oil shale samples by sputtering away material via momentum transfer causing the ejection of target atoms and the exposure of a fresh oil shale surface. The SEM imaged the newly milled oil shale crosssectional surface at a 2 keV accelerating voltage and a 0.4 nA beam current under BSE mode to achieve excellent contrast between inorganic minerals, organic matter and the pore space. The ion beam and electron beam were set a 52°angle to each other (Fig. 15B), resulting in parallel cuts and 600 evenly spaced slices. A fiducial reference mark was placed on the oil shale surface next to the region of interest for alignment and registration of the indi-vidual slices (Fig. 15C). Prior to milling a 4 mm strip of platinum was deposited over the site of interest to protect from excessive beam damage and minimize curtaining artifacts [118,119]. Auto Slice and View G3 software (FEI) was used for serial sectioning of the oil shale samples creating 2048 Â 1155 Â 600 voxel volumes with a voxel size of 14.63 Â 18.32 Â 10.0 nm for both sample 1 (unpyrolyzed) and sample 2 (pyrolyzed) (Fig. 16).
Each FIB-SEM data set was then imported to PerGeos software (FEI, Hillsboro, OR, USA) for image processing. A major challenge in processing FIB-SEM shale image data is the presence of poreback or shine-through artifacts arising when structures lying below the milling plane are visible through pores with electrons going behind an imaging plane and hitting the back-side of the pore. The signal reflected from the pore-back leads to difficulties in image segmentation where the pore space should be separated from the solid material. In a typical shale sample, the grayscale of the matrix is similar to the grayscale of the pore backs. The specific FIB-SEM pore-back problem, caused by the electrons reflected, depends on a number of parameters such as the size, the depth, and the orientation of the pore. The traditional approach of global thresholding breaks down, particularly in regions of high porosity where more electrons can be detected from deeper regions in the sample leading to higher grayscale values in the pore space. In this study, the data sets were pre-processed using a Fast Fourier transform (FFT) and a non-local means edge preserving smoothing filter [116,117]. Following this, a three-phase segmentation method was applied to separate organic matter, minerals and the pore space (including pore-back artifacts), relying on 3-D gradient and marker-based watershed transforms (Fig. 17).
From a stack of 2-D images, the kerogen, minerals and pore phases were serially reconstructed to render 3-D FIB-SEM volumes from sample 1 (unpyrolyzed) and sample 2 (pyrolyzed) (Figs. 18A and 19A respectively). Fig. 18B represents the 3-D segmented visualization for the immature Green River (Mahogany Zone) oil shale (sample 1) which reveals that the organic matter network is complex and that connectivity occurs across the 30 mm Â 20 mm Â 6 mm volume. The pore space is isolated with pores ranging in length scale from a several nanometers to a few micrometers and in shape from spherical to elongated geometries. The volume contributions of kerogen, minerals and pores for the reconstructed raw oil shale volume were 51.8%, 47.7% and 0.5% respectively (Figs. 18C-E). Fig. 19B represents the 3-D segmented visualization for the Green River (Mahogany Zone) oil shale following pyrolysis at 500°C (sample 2) revealing a dominant pore space with connecting micro-fractures parallel to the kerogen laminations but limited connection between layers. The FIB-SEM volume also reveals a residual connected kerogen network following pyrolysis. The volume contributions of kerogen, minerals and pores for the reconstructed raw oil shale volume were 39.6%, 42.2% and 18.2% respectively (Figs. 19C-E).
The number of individual pore-body sizes was computed from the FIB-SEM reconstructed volumes. Fig. 20 shows a 3-D rendering of the connectivity of the pore space for both the unpyrolyzed and pyrolyzed samples. For the unpyrolyzed sample, all the pore space appears disconnected, as indicated by the wide range of colors (Fig. 20A). However, for the pyrolyzed sample, the pore space is well-connected laterally across the sample (Fig. 20B). Pore sizes were computed as the radii of spheres of equivalent volume for each pore-body. All the pores have been divided into different size categories each of which is represented by a mean equivalent radius. Histograms of the pore size distribution for the Green River (Mahogany Zone) oil shale for the both the unpyrolyzed and pyrolyzed samples were constructed, along with the cumulative percent of the pore size distribution (Figs. 21 and 22). For the unpyrolyzed oil shale FIB-SEM volume, the histogram indicates that pores having an equivalent radius approximately between 15 nm and 22 nm dominate the distribution (Fig. 21A). Although from Fig. 21A we quantify that relatively smaller pores dominate in number, Fig. 21B shows that pores approximately ranging from 140 nm to 390 nm in equivalent radius have the greatest volumetric contribution, occupying 75.9% of the total pore volume. For the pyrolyzed oil shale, the histogram in Fig. 22A indicates that pores having an equivalent radius of approximately between 9 nm and 18 nm dominate in number while Fig. 22B shows that pores with an approximate equivalent radius between 1.6 mm and 2.0 mm significantly dominate in the volumetric contribution taking 98.7% of the pore space within the analyzed 30 mm Â 20 mm Â 6 mm FIB-SEM volume.
Each oil shale FIB-SEM dataset in this study represents a volume of approximately 3600 mm 3 . It is important to note that the gain in resolution to detect nano-scale features comes at the expense of the volume analyzed. The volumetric contribution and spatial distribution of organic matter, minerals and pores in FIB-SEM oil shale datasets represent a tiny fraction of large oil shale formations. Upscaling petrophysical values computed from FIB-SEM volumes requires an understanding of the representative elementary volume for oil shale, which can vary among different oil shale deposits. In addition, multiple oil shale sampling volumes will be required to achieve statistical significance. Moreover, by correlating FIB-SEM data sets with large-area automated SEM imaging and X-ray mCT which offers lower resolution but larger field of view imaging, a more comprehensive understanding can be acquired of both the oil shale macro-and micro fabric heterogeneity and the association of nano-scale structures within the fabric domains.

Conclusions
(1) The data presented in this study show that combining X-ray mCT with automated ultra-high resolution SEM, MAPS Mineralogy and FIB-SEM provides a powerful and synergistic multi-scale and multi-dimensional workflow. This integrated imaging strategy delivers crucial petrophysical characterization of heterogeneous fabrics in oil shales revealing important microstructural and mineralogical complexities. 2-D and 3-D imaging, allows for the visualization and quantification of pores, organic matter and inorganic mineral phases over mm-mm-nm scales before and after pyrolysis. (2) Oil shale samples from the organic-rich Green River Formation (Mahogany Zone) -the world's largest oil shale deposit -were scanned using X-ray mCT at a 0.8 mm voxel size revealing a well-connected organic network within a heterogeneous inorganic mineral matrix. 3-D X-ray mCT scans fol-lowing incremental ex situ oil shale pyrolysis up to 420°C captured the propagation and coalescence of microfractures, with the first observed fractures nucleating at 380°C and substantive volume expansion occurring at 400°C. (3) Automated tiling and stitching of ultra-high resolution SEM images into mosaics of gigapixel size were captured for unpyrolyzed and pyrolyzed samples over 18 mm Â 14 mm areas. Argon (Ar) ion milling was used to prepare ultra-flat surfaces with minimal artifacts. BSE mosaic images enabled pore-scale resolutions over the entire sample surface in one image allowing the simultaneous retrieval of microstructural fabric heterogeneity and nanometer-scale structures without losing correlation. The stitched mosaic of images allowed for the systematic exploration of fine-grained microstructures within oil shale samples from the nm to mm scales. (4) BSE images revealed that organic matter hosted pores, rather than mineral-hosted pores, were the dominant contributors to total porosity and were predominantly sub-mm in size with irregular polygonal geometries. Following pyrolysis, multiple fractures were observed along with regions of partially decomposed and unconverted kerogen. MAPS Mineralogy indicated that the Mahogany oil shale is calcareous in composition with a complex distribution of dolomite, calcite, K-feldspar, quartz, pyrite and illitic clay platelets within the dispersed organic matter. (5) FIB-SEM volumes of 30 mm Â 20 mm Â 6 mm for unpyrolyzed and pyrolyzed samples were acquired. 3-D gradient and marker-based transforms provided successful three-phase segmentation to separate organic matter, minerals and the pore space (including pore-back artifacts). The results revealed unconnected pores occupying 0.5% of the analyzed volume, with pores between 15 nm and 22 nm in equivalent radius dominating in number. The pyrolyzed oil shale showed well-connected pores across the volume with a total sample porosity of 18.2%. Pores with an equivalent radius between 1.6 mm and 2.0 mm made up the bulk of the pore volume in the pyrolyzed sample. (6) Correlating FIB-SEM data sets with large-area automated SEM imaging and X-ray mCT provides a comprehensive framework to capture oil shale micro-fabric heterogeneity and the association of nano-scale structures within the fabric domains. Multiple 2-D images are needed to qualitatively and quantitatively characterize pores and fine clay minerals at the nano-scale. The grain sizes, identification and distribution of inorganic minerals can be achieved by combining sub-mm BSE imaging with MAPS Mineralogy. Several 3-D image analyses are necessary to substantiate the extent of the organic matter network and pore-space connectivity. Integrating 2-D and 3-D rock fabric analysis provides a promising pathway towards understanding, predicting and modeling hydrocarbon transport and production during oil shale pyrolysis.