Journal of Earth Science  2019, Vol. 30 Issue (2): 407-421   PDF    
Automated Image Analysis of Mud and Mudrock Microstructure and Characteristics of Hemipelagic Sediments:IODP Expedition 339
Bankole Shereef A. 1,2, Buckman Jim 1, Stow Dorrik 1, Lever Helen 1     
1. Institute of Petroleum Engineering, Heriot-Watt University, Edinburgh EH14 4AS, UK;
2. Department of Chemical and Geological Sciences, Al-Hikmah University, P. M. B 1601, Nigeria
ABSTRACT: The microstructural analysis of muds and mudrocks requires very high-resolution measurement. Recent advances in electron microscopy have contributed significantly to the improved characterisation of mudrock microstructures and their consequent petrophysical properties. However, imaging through electron microscopy is limited to small areas of coverage such that upscaling of these properties is a great challenge. In this paper, we develop a new methodology for multiple large-area imaging using scanning electron microscopy through automated acquisition and stitching from polished thin-sections and ion-milled samples. The process is fast, efficient and minimises user-input and bias. It can provide reliable, quantifiable data on sediment grain size, grain orientation, pore size and porosity. Limitations include the time involved for individual runs and manual segmentation, the large amount of computer memory required, and instrument resolution at the nano-scale. This method is applied to selected samples of Quaternary muddy sediments from the Iberian margin at IODP Site 1385. The section comprises finegrained (very fine clayey silts), mixed-composition, biogenic-terrigenous hemipelagites, with a pronounced but non-regular colour cyclicity. There is a multi-tiered and diverse trace fossil assemblage of the deep-water Zoophycos ichnofacies. The sediment microstructures show small-scale heterogeneity in all properties, and an overall random fabric with secondary preferred grain-alignment. These results on the fabric differ, in part, from previous studies of hemipelagic muds. Further work is underway on their comparison with other deep-water sediment facies.
KEY WORDS: mudrocks    microstructure    microporosity    grain-orientation    hemipelagites    trace fossils    

Microstructure is an important feature that affects many physical properties such as porosity, pore connectivity and permeability of mudrocks. The renewed interest in microstructural study has a link to the growing interest in the development of shale gas, carbon storage and radioactive disposal (Hemes et al., 2013; Curtis et al., 2012). Due to the small grain size of particles in mudrocks, their characterisation is challenging and conventional equipment such as optical microscopy cannot quantify their microstructure. The use of high-resolution instrument such as scanning electron microscopy (SEM) is well suited for studying the complex mudrock microstructure, but there is a trade-off between resolution and coverage area of the SEM.

The present study utilised scanning electron microscopy and its aims are twofold: (1) to present an improved methodology for investigating the microstructure of mudrocks; and (2) to apply this methodology to better understand the characterisation of hemiplegic sediments on the Iberian continental margin.

The microstructural methodology presented here includes analysis of grain size, grain orientation and arrangement, porosity and pore-size distribution, and mineral composition. Such studies are extremely challenging because the very small grain size and even smaller pore sizes are at the resolution limit of most conventional techniques (Camp et al., 2013). However, mudrock microstructure has a significant effects on understanding the petrophysical properties, geotechnical characteristics and diagenesis of fine-grained sediments (Janssen et al., 2012; Josh et al., 2012). These are important properties that control hydrocarbon storage and migration in shale reservoirs, primary migration from shale source rocks, sealing integrity of cap rocks, and hydraulic fracking properties of shale reservoirs. They are also significant for storage of nuclear waste in fine grained sediments (Houben et al., 2013; Keller et al., 2013). Furthermore, the preferred orientation of clay particles causes seismic elastic anisotropy (Wenk et al., 2014, 2008; Lonardelli et al., 2007; Wenk and Houtte, 2004), and also has implications for the deformation history of mudrocks (DeVasto et al., 2012).

Electron microscopy, especially scanning electron microscopy, is the principal method to directly investigate the microstructure of mudrocks at high-resolution (micrometer to nanometer scales), both qualitatively and quantitatively. There are two key problems associated with electron microscopy: (1) the acquisition of images is limited to a very small sample area, which might not be representative of a larger sample size (Saraji and Piri, 2015; Hemes et al., 2013) and (2) mudrocks are noted for their high heterogeneity at a variety of scales (Aplin and Macquaker, 2011; Macquaker and Howell, 1999). Recently, methods that involve the automated collection and stitching of thousands of image tiles at high-resolution using scanning electron microscopy have been presented (Bankole et al., 2016; Buckman, 2014; Lemmens and Richards, 2013). However, the images produced through this process may run to gigabytes of memory and are therefore difficult to handle manually. Hence, we present here a workflow involving an automated method of handling such large data sets, which is crucial for efficient time management in interpreting petrophysical properties of mudrocks.

The methods presented herein are applied to a uniform mid-slope series of hemipelagic sediments that were retrieved during Expedition 339 of the International Ocean Discovery Program (IODP) at Site 1385 on the Iberian continental margin off SW Portugal. This is known as the 'Shackleton Site' in honour of Sir Nick Shackleton, whose pioneering work on earlier cores from this location has been pivotal in the understanding of millennial-scale climatic variation over the past glacial cycle. Following careful shipboard and shore-based study of the cores, a total of five representative samples were selected for this study from the bioturbated, calcareous and unconsolidated mud rich intervals, interpreted as hemipelagites by the shipboard scientists. The age of the sediments retrieved from the Iberian margin site is Quaternary (Hodell et al., 2013; Stow et al., 2013). Further details on the samples used are presented in Table 1.

Table 1 Brief description of samples analysed for this study

Hemipelagites are fine-grained sediments, typically muds and mudrocks, which comprise mixtures of terrigenous and biogenic material, of which the terrigenous component is silt-rich. They are deposited by a combination of vertical settling and slow lateral advection (Stow and Tabrez, 1998). Hemipelagites are one of the principal marine sediment types covering large tracts of continental margins worldwide, and forming the 'background' facies of many deep-water successions (Pickering and Hiscott, 2015; Stow et al., 2001; Stow, 1985). Many black-shale source rocks and organic-rich shale-gas reservoirs are largely of hemipelagic origin, although other processes may also be involved (Stow et al., 2001). However, the detailed sedimentary characteristics of hemipelagites is still quite rare (see summary in Stow and Tabrez, 1998). Hence, we present here both lithological and microstructural characteristics for the hemipelagites in our study area.

1 MATERIALS AND METHODS 1.1 Core Description and Sampling

Detailed visual core description was carried out on board of the Joides Resolution during IODP Expedition 339 by the Expedition scientists. This was augmented by petrographic analysis of smear slides, selected X-ray diffraction analysis of powdered bulk samples, and geochemical analysis of total carbonate content (organic and inorganic). Physical properties measurements on whole cores included sediment colour and reflectance spectrometry, magnetic susceptibility, natural gamma radiation, grain density and sediment strength. A post-cruise review of the cored section at the core repository in Bremen, Germany, and a separate ichnological study were also carried out on Site 1385, and the principal results presented herein.

Five samples were selected for microstructural analyses. These were oven dried slowly at a temperature of 60 ℃ until the weight of the sample became constant regardless of further drying. The dried samples were vacuum impregnated prior to the preparation of well-polished thin-sections and ion milled samples.

1.2 SEM High-Resolution Montages and Automated Imaging

The procedure for acquisition of large sale images as employed here is similar to the method presented by Lemmens and Richards (2013) for which further details can be found elsewhere (Bankole et al., 2016; Buckman, 2014). A field emission Quanta 650 SEM equipped with a backscattered detector was used to acquire the images. The operational settings used were: low vacuum mode (0.83 Torr), 15 kV, 4.5 spot size and 10 mm working distance.

A step-wise scanning procedure was adopted (Fig. 1) that involves: (1) a low-resolution SEM image of the whole polished thin-section, in order to obtain information on the broad-scale distribution of different mineral phases and sample cracks; (2) a high-resolution SEM image at 45 nm per pixel, for analysis of grain orientation and arrangement; and (3) a super high-resolution SEM image at 3.2 nm per pixel, in order to investigate porosity and pore size distribution. Randomly selected areas were imaged at both high-resolution and super high-resolution for grain orientation and porosity (Table 2).

Figure 1. A step-wise method of SEM imaging of polished thin section. (a) Low resolution image of the whole polished thin-section with resolution of 1 μm per pixel; (b) high-resolution SEM image at 45 nm per pixel; (c) super high-resolution SEM image at about 5 nm per pixel.
Table 2 Number of SEM images and EDX acquired at randomly selected areas (subsets) per sample
1.3 Image Analysis

Image analysis was performed with Fiji software (version 1.51 for windows, 64 bits), an adaptation of image J. Fiji is an open source software developed by the National Institute of Health (NIH), United States of America. Fiji hosts a library of algorithms for practically handling image analysis. The software was initially developed as a platform to iteratively handle and analyse biological images (Schindelin et al., 2012). The usage of the software is not limited to biology but can be applied across several disciplines.

The SEM images were processed through contrast and brightness enhancement and the application of a median filter with radius value of 4 pixels. There are several segmentation methods reported in the literatures (Zaitoun and Aqel, 2015; Pal and Pal, 1993), the simplest is by manual thresholding which is based on visual judgement. In this study, manual segmentation was employed for analysis of pores, such that the features of interest (pores) were rendered to the foreground and the scale of the image resolution was set based on the horizontal field of view. The scale was set in nanometres per pixel. To fully automate the workflow, a subset of images that is adjudged representative of several images is first analysed in order to set certain parameters, such as contrast and brightness, and to threshold in grayscale.

Data on pore size were obtained via the particle analysis function with Fiji. Information produced on pores in this way include Feret diameter, pore area, circularity, aspect ratio and orientation angles. All the steps involved were turned into script via macro recording and then run on a folder containing several images through batch processing. The macro recording produces a script of all the steps involved. The recorded macro is like script writing but it has the advantage that it requires no prior knowledge about scripting or any programming language. Further information on operations of Fiji can be found elsewhere (Schindelin et al., 2015, 2012; Schneider et al., 2012). Due to the intricacy involved in segmenting clay platelets and silt-size particles, Trainable Weka segmentation was employed. Trainable Weka segmentation (TWS) is a form of pattern recognition by the system through supervised machine learning segmentation. This process requires manual annotation of features of interest to train the classifier. The image is segmented based on the selected classifier. The segmentation may require several iterations until the user is satisfied with the result. Trainable Weka segmentation was applied and the classifier was trained to identify the following classes: quartz, feldspar, muscovite, calcite, clay platelets, undifferentiated grains and pores (Figs. 2b, 2e). In the final binarised image, grains were rendered to the foreground while undifferentiated grains and pores were rendered to the background (Figs. 2c, 2f).

Figure 2. Sample of SEM images segmented through Trainable Weka segmentation (a), (d) processed SEM image (b), (e), output of Trainable Weka segmentation (c), (f) binarised images. The horizontal field of view (HFOV) of the images is 300 μm. Note the outputs of the Trainable Weka contain three classes; pores, grains and undifferentiated grains. In the binarised image, the grains were rendered to foreground. The undifferentiated grains are aggregates of grains which cannot be separated into individual grains during thresholding.

The setting employed with Trainable Weka involves the selection of minimum but reasonable training features that is appropriate for the segmentation. The features selected are Sobel filter, Gaussian blur and median. Once the classified image is satisfactory, the classifier was saved and later applied to several other images in order to reduce the running time.

1.4 Grain Size

Grain-size measurements were made from the processed images using Fiji software. Six randomly selected areas (or subsets) were imaged at high-resolution (45 nm per pixel) for each of the five polished thin sections. The dimension of each area is approximately 650 µm by 400 µm, which is believed to be sufficiently representative of the whole sample. Random selection of these areas was made in order to account for heterogeneity in the grain size within the sample. Raw images from the scanning electron microscope were processed with Fiji by first setting the scale of the image based on the horizontal field of view of the tiles in nanometres per pixel. The grains were then analysed and data acquired on grain diameter, perimeter, area, circularity, and aspect ratio. Data returned by Fiji were saved in Excel format and further data management were automated through some Excel functions and Visual Basic for applications macros. Grain size was determined based on percentage Feret diameter by summing up the diameters in each sediment class.

1.5 Silt and Clay Orientation

Particle orientation of both silt and clay-sized grains was analysed using the particle size analysis function in Fiji. The orientation of each grain is measured through the best fit ellipse drawn around that grain. In order to focus only on the silt and clay-size particles separately, a macro was run on the orientation data in Excel in order to examine only particle sizes < 4 µm (i.e., clays) in the first instance, and then 4-63 µm (i.e., silt-size). A further filter was applied to exclude particles with an aspect ratio less than 2. This guaranteed that only silt and clay size particles that are elongated (typically mica and clay platelets) were used for the orientation analysis. Orientation data were divided into 18 bins and rose diagrams were constructed with Georient 9.5.1 software. In addition to the rose diagrams produced by Georient, the software gives as an output of circular statistical parameters, including circular variance, circular standard deviation, Kappa coefficient, and circular skewness, among others.

1.6 Mineralogy

Mineralogical composition was analysed via energy dispersive X-ray (EDX) analysis, within the scanning electron microscope. The EDX can provide information about the mineral phases through elemental composition. Mineralogical information through EDX were acquired on carbon-coated, polished thin-sections at 20 kV, high vacuum mode. EDX maps were acquired at 2 µm per pixel and 69 nm per pixel, such that the total horizontal fields of view were 1 mm and 70 µm, respectively. A total of 75 frames per map, with a scan time of 10 µs, were taken in order to maximise data quality at both resolutions. Mineralogical phase maps were constructed and quantified using AZtec software. The derived mineralogy from EDX requires observer intervention by interrogating several areas of the map to determine the elemental composition. A combination of elemental composition and mineralogical morphology form the basis for identifying minerals present in the EDX maps.

To directly compare the result of the EDX analysis with another technique, bulk X-ray diffraction was performed on all the samples. The samples for XRD analysis were oven dried at 60 ℃, ground with mortar and pestle and then mounted on a glass slide with acetone, similar to the smear mount method described by (Munson et al., 2016). X-ray diffraction data were collected at room temperature on a Bruker D8 advance powder diffractometer, operating with Ge-monochromatic Cu Kα1 radiation. The mineralogical composition was semi-quantified from the diffraction pattern using the intensity peak ratio and corrected with the multiplication factor presented in Piper (1977).

1.7 Pore Size Distribution and Porosity

Determination of pore sizes was based on Feret diameter and porosity was estimated as an area percentage using the Fiji software. Information retrieved on pore size, porosity and grain orientation through Fiji were saved in Excel format. Pores less than 15 nm were filtered out in Excel through Visual Basic Application (VBA). Further analysis and plotting were performed in Excel 2014 and Matlab 2016b (The MathWorks, Inc., Natick, MA, USA) for graphical presentations. Estimation of pore size and porosity was performed by progressively increasing the area of the image to achieve a less statistically varied value (Kameda et al., 2006; Bosl et al., 1998).

To investigate whether pore size distribution among subsets of the same sample varied, empirical quantile-quantile plots (Q-Q plots) were constructed. A Q-Q plot is a non-parametric statistical test to determine if two sets of data have a common distribution. The quantile of subset A was plotted against the quantile of other subsets for each of the samples on a log-log graph. A common distribution between two data sets is indicated by the Q-Q plot when it falls close to a reference line y=x. Further information about information on Q-Q plots can be found elsewhere (Lovie, 2005; Chambers et al., 1984).

To display variation in porosity at the microscale, hundreds of tiles of SEM images were analysed for porosity and subsequently turned into coloured contour maps. The maps express variation in pore distribution among subsets (Buckman et al., 2017). Because the polished thin-sections were prepared perpendicular to the bedding both vertical and lateral pore distribution were visually observed.

2 RESULTS 2.1 Lithological Characteristics

The sediments recovered from Site 1385 on the Portuguese continental slope are a very uniform series of nannofossil muds, with variable proportions of biogenic carbonate and terrigenous material. Bedding is very indistinct to non-existent, but a more or less distinct colour variation is evident throughout, from paler to darker greyish hues. These colour cycles correspond with more biogenic content (paler) and more terrigenous content (darker), respectively. The same cyclicity is also observed in physical property measurements, including natural gamma radiation, magnetic susceptibility, and density, as well as sediment colour spectral indices. The minor lithologies present include carbonate-rich nannofossil ooze, more clay-rich mud with biogenic grains.

There are no primary sedimentary structures present, and no discernible variation in the very fine grain size. Bioturbation and burrowing is pervasive, and the bioturbation index ranges from moderate to intense. The trace fossil assemblage comprises abundant Planolites, common Paleophycus, Thalassinoides and Taenidium, Chondrites and Zoophycos. Other non-specific traces are also present together with abundant biodeformation. Small-scale, sub-vertical microfaults are present at relatively few restricted intervals, and one thin interval of contorted strata was observed. Typical facies photographs with trace fossils and bioturbation are shown in Fig. 3.

Figure 3. Selected core sections representative of hemipelagite facies at IODP Site 1385. (a) Core shows part of colour cycles from light grey (more biogenic carbonate) to dark grey (more terrigenous clay). Intense multi-tiered bioturbation and burrowing throughout. Larger trace fossils observed include: Planolites, Zoophycos, Scolicia and Thalassinoides. Small-scale burrows in mottled background are probable Phycosiphon; (b) detail of Zoophycos, displaying three levels of a single specimen (Z1-Z3); (c) zoophycos (Z4), with axial tube (Z4-at), indicating a minimum burrow depth of 120 mm. Z4 cross cut by a second Zoophycos (Z5).

The Shackleton Site 1385 was drilled in order to provide a continuous marine record of Pleistocene millennial-scale climatic variability that can be correlated with both polar ice cores and European terrestrial records. For this reason, it has been intensively studied and has thus far yielded four separate age models, all of which are in very good agreement (Hodell et al., 2015). These show that sediment accumulation rates have been extremely uniform over the past 1.5 Ma, averaging around 11 cm/ka.

2.2 Grain Size

The results of the grain size analysis are presented more fully in a separate paper and further details on the procedures, results and discussion are given therein (Bankole et al., 2019). In brief, the samples are all within the mud-size field (Stow, 2005), ranging from silty clays to clayey silts. The mean size for all samples ranges from 7.6-6.8 phi (i.e., about 6-9 μm), with unimodal size distributions. They are very well to moderately well sorted, symmetrical to fine skewed, and mesokurtic to platykurtic. Significantly, from the methodological viewpoint, there is no substantial variation in the grain size among the subsets for any one sample (Fig. 4).

Figure 4. Ternary plot of grain size for several subsets. Mudrock classification in both ternary based on Stow (2005), subsets from samples 5 and 6 overlap and thus they plot with a cluster.
2.3 Particle Orientation

The full collection of particle orientation measurements is presented as rose diagrams for each of the sample subsets, and for silt and clay fractions separately, in Table 3. Sample 1 is dominated by random or mixed alignment, with two subsets that show preferred alignment. Sample 2 is typical of a more or less random or less completely random particle orientation for both silt and clay fractions separately. Only two of the subset samples show a more aligned to mixed orientation that appears at a high angle to the horizontal. Of the 60 individual orientation measurements from the 5 samples, 28 show random orientation, 18 display preferred sub-parallel orientation, and 10 show mixed (or polymodal) alignment. The samples that show sub-parallel alignment, are mostly horizontal or slightly oblique, whereas a further 4 samples show 'anomalous' orientations at a high oblique angle to the horizontal. In almost all cases, orientation of clays versus that of silt particles were found to be similar.

Table 3 Rose plots for silt and clay orientation for all the samples and their varying subsets. Arrow indicates direction parallel to the bedding. P. Bedding parallel; OB. oblique orientation to the bedding; R. random orientation.

In order to validate our visual observation of random versus preferred grain alignment, we calculated both circular statistical parameters and entropy values (Table 4). Circular variance measures dispersion of the orientation data and its value ranges between 0 and 1. Values close to 0 suggest distribution of data are well aligned with small dispersion, whereas values close to 1 suggest more random distribution (Berens, 2009; Davis, 1986). The Kappa coefficient (k) is inversely proportional to dispersion, such that high Kappa values (k > 0.5) suggests the angular data are restricted to a narrow arc of the circle with little dispersion i.e. preferred orientation (Suttle et al., 2017; Mardia and Jupp, 2008; Davis, 1986). Entropy is a different parameter for determining orientation of fabric for both unimodal and multimodal data, for which a high value signifies randomness, whereas a low value signifies greater alignment. Suttle et al. (2017) used a value of 2.80 to differentiate random and preferred alignment in fine-grained meteorites.

Table 4 Summary of circular statistical parameters

The data we present in Table 4 are based on averages of all subsets for each sample and, therefore, the ranges shown for circular variance, kappa coefficient and entropy include what we have interpreted as both random and preferred fabrics. These ranges typically span the random/preferred fabric values. Individual subset measurements, however, show better correlation (Table 3).

Furthermore, a plot of entropy values for clay size particles against that of silt particles shows strong correlation between the two which is also supported by a high value of Pearson correlation coefficient (0.91) (Fig. 5). This agrees with visual inspection of the rose diagrams which show that silt and clay orientations for each subset are very similar. At a 95% confidence interval, F-test statistics computed from the entropy values of silt and clay size particles showed that the null hypothesis can be accepted, which means that the calculated entropy for silt particles and clay size particles have equal variance (Table 5). All the statistical parameters, therefore, indicate that orientation analysis of silt particles is a proxy for clay size particle orientation.

Figure 5. Plot of entropy values for silt against entropy for clay. The graph shows a strong correlation between the two set of values which is an indication that the silt and clay particles have similar orientation.
Table 5 F-test two samples variances for entropies of clay and silt
2.4 Mineralogy

The composition of sediments at Site 1385 is a distinctive mixture of biogenic and terrigenous components. Smear-slide petrography reveals the biogenic fraction to comprise dominantly calcareous nannofossils, with minor to rare foraminifera, diatoms and sponge spicules. The terrigenous fraction is dominated by quartz silt, clay minerals and detrital carbonate silt, with minor to rare feldspar, accessory heavy minerals and authigenic dolomite. There is a notable difference in biogenic/terrigenous proportions between the dark and light-coloured cycles, as well as a slight increase in terrigenous fraction in the upper 40 m of section. The shipboard measurement of total carbonate content ranged between about 20% and 40% by weight.

The mineral composition determined by automated image analysis with EDX measurement is shown in Table 6, and compared with results from our own XRD analysis. Broadly speaking, the same minerals are identified by the different techniques, in the different subsets, and at the different EDX resolution settings used (i.e., 2 µm per pixel and 69 nm per pixel). However, the quantified measurements obtained may vary significantly for each of the three principal components-calcite, clay minerals (illite-kaolinite-smectite), and silt minerals (quartz-feldspar-mica).

Table 6 EDX average mineral composition from different subsets and bulk mineral composition

There is a systematic difference between the coarser and finer resolution measurements by EDX, with the finer resolution always being closer in amount to that determined by XRD. The difference between subsets in one sample is typically within 10 percentage points, rarely more. However, the difference between techniques is more marked, with the amounts of principal components varying by a factor of 2 or more.

2.5 Pore Size and Porosity

The mean pore sizes measured fall within the macropore range (Sing et al., 1985), with actual values from 541 to 1 000 nm (i.e., 0.54-1.0 µm) (Table 7). They vary significantly among subsets of the same sample, typically by between 10%-30%. The median pore sizes are systematically lower, wheras the modal pore sizes for all the samples are much less (21-177 nm) and fall within the micropore (2-50 nm) and mesopore range (> 50 nm). Computed skewness for the pore size distributions indicate that all the samples are positively skewed. Pore size distributions for the subsets in each sample shows that all samples have a unimodal distribution and are log-normally distributed (Figs. 6, 7).

Table 7 Computed statistical parameter for the pore sizes. There is wide disparity between the mean and the median because the pore sizes are highly skewed
Figure 6. Pore size distribution for Sample 5 at different size of areas (a) subset A and (b) subset B. The distribution of pores in both graphs indicate increase in number of pores with progressive increase in area analysed. The modal distribution in subset A varies with the size of the area while in subset B, the modal pore is with a sharp peak and consistent all through as the area analysed increases.
Figure 7. Log-normal plot of frequency vs. pore size (a) Sample 1 (b) Sample 3. All subset in both sample have unimodal distribution.

Samples of the Q-Q plots are presented in Fig. 8. Most of the Q-Q plots constructed showed that the subsets in each sample have similar pore size distributions with only few having slightly varied pore size distributions. However, the number of pores per subset differs substantially.

Figure 8. Empirical Q-Q plots for pore size distribution for Sample 1. (a) and (b) showed slightly varied distribution while (c) showed that the pore distribution between the two subsets are similar.

A summary of average porosity measurements is presented in Table 8. These show the full range of subset values obtained as well as the values at progressively increasing areas of measurement from 900 to 5 400 µm2. The representative area is taken as the area for which variation in porosity is less than 10% (Fig. 9) indicating that the change is statistically insignificant (Vanden-Bygaart and Protz, 1999). Although this is subset specific, it was noted that an area of 3 600 µm2 can be considered a representative area for all the subsets. For this area, porosity across the subsets typically varies by 10%-15% of the value, but can be much more in some cases (i.e., > 50% variation).

Table 8 Summary of porosity (%) values from different subsets
Figure 9. Plot of porosity values for different subsets in (a) Sample 4 (b) Sample 5. Note that the distribution of porosity across the subsets varies.

In general, the distribution of porosity varies from one subset to the other within individual samples, as well as within individual subsets. Based on porosity distribution maps (Figs. 10, 11), the subsets were classified into three groups: highly porous, partially porous and tightly porous (Table 9). Highly porous samples are those in which more than half of the area shows values of more than 15% porosity. Porosity in this group is evenly distributed and well-connected in 2-dimensions (Fig. 10c). The second group are those in which less than half of the area shows 15% porosity, and the pore spaces are either poorly linked or disconnected. This group are considered partially porous and characterised by clear separation between porous and non-porous areas (Figs. 10a, 10b, 11c). The last group are those in which more than half of the area shows less than 10% porosity (Figs. 11a, 11b). These are considered tightly porous.

Figure 10. Contour maps showing variation in porosity for Sample 2, (a)-(c) are subsets A to C. Subsets A and B are examples of partially porous while (c) is highly porous. Total horizontal field of view (HFOV) is 95 μm.
Figure 11. Contour maps showing pore distribution for Sample 1, (a)-(c) are subsets A to C. Subsets A and B are tightly porous while (c) is partially porous. Total horizontal field of view (HFOV) is 95 μm.
Table 9 Classification of subsets per sample based on porosity distribution. It is apparent that no single sample has homogenous porosity distribution across the subsets
3 DISCUSSION 3.1 Methodology

The automated image analysis (AIA) technique presented here differs from prior work in two important respects. Firstly, it images relatively large areas by a process of automated acquisition and stitching and, secondly, it incorporates multiple areas that have been randomly selected from the available polished thin section or ion-milled sample. Both these aspects help mitigate against relying on data from too small a sample area of a heterogeneous sediment. The methodology minimises human subjectivity and bias, but does require selected manual thresholding by image segmentation. This is non-trivial and may lead to varied results among different operators. It is likely that the discrepancies in results among users is minimal and for an individual, the process is expected to be reproducible. These factors can be quantified (Grove and Jerram, 2011; Francus and Pirard, 2004), but are not addressed here.

For the different attributes investigated, we make the following observations and assessment of the AIA method.

Grain size analysis. AIA provides an important and reliable technique for grain size analysis of very fine-grained sediments, for which high resolution measurement is required. It measures actual grain diameters (Feret diameter) for tens of thousands of grains, and is effective over a wide range of sizes from 10 nm to 5 mm. However, for most SEM measurements of mudrocks, we suggest a preferred range of 100 nm to 100 µm. The results obtained here were found to be closely comparable to those obtained by laser diffraction on the same samples (Bankole et al., 2019), and broadly comparable with the results of laser diffraction analysis carried out by Nishida (2016) on other samples from the same site. A significant advantage of the AIA methodology is that it can be applied equally to unconsolidated and consolidated sediments, as well as to compacted and cemented sedimentary rocks.

3.1.1 Particle orientation

Visual inspection of grain alignment is the only effective method of determining particle orientation in mudrocks. AIA, therefore, provides a fast, effective and objective method of measuring a large dataset and deriving meaningful quantification. The majority of previous works on the application of SEM to the orientation of clay particles are descriptive. There have been some attempts to quantify orientation of clay particles automatically, although these earlier efforts involve deduction of particle orientation from certain proxies (Martínez-Nistal et al., 1999; Tovey et al., 1992; Sokolov and O'Brien, 1990).

It is important to separate out measurements for silt and clay-size particles and to filter for elongate grains. It is also important to use circular statistical and entropy measures in conjunction with visual observation, as the circular statistical parameters are less reliable with multimodal orientation data (Mardia and Jupp, 2008; Fisher, 1993). An important result of this study is to demonstrate that silt orientation is a very good proxy for clay orientation. The results show an approximately equal spread of random and preferred fabrics, which we suggest is characteristic for hemipelagites (see discussion below). The slightly oblique to horizontal orientation, common to a number of subsets, may in some cases be due to a bedding curvature induced by the coring process, which is especially common at the very edge of cores. However, in most cases we suggest the grain alignment is due to the bioturbational fabric.

3.1.2 Mineralogy

By contrast, AIA does not appear to be such a good method for determining the sediment composition or, at least, not by using the SEM-EDX methodology in this study. There is poor repeatability between subsets of the same sample as well as significant differences in mineral composition at the two scales of resolution used (i.e., 2 µm and 63 nm per pixel). There is also poor agreement with the composition data obtained from the same samples by XRD measurement. Problems encountered with the AIA technique include: (a) an over-estimation of the dominant minerals present; (b) separation of calcite grains encased within a clay matrix; and (c) the separation of detrital mica silt grains (muscovite, biotite) from clay-fraction illite. Furthermore, high-resolution EDX maps of large areas is time-consuming and runs to several gigabytes of data, which is difficult to handle. The high-resolution EDX maps presented here are limited to about 4 200 µm2 and these are unlikely to be representative.

SEM-EDX does have the advantage of determining the mineralogy of specific grains and of making a visual assessment of their morphology or crystallinity and likely diagenetic or detrital origin. It can be used to pick out important heavy mineral components, such as the anatase and pyrite observed in this study, which may be useful for provenance or digenetic studies. However, further work is needed to enable the effective application of AIA. This should also include calibration of results with whole-rock x-ray fluorescence techniques.

3.1.3 Porosity

The usage of image analysis for determining porosity and pore size distribution is not well established compared to the more traditional methods, but the process involved is both easy and highly effective. Whereas previous work on pores and pore networks has mostly been qualitative (Curtis et al., 2012, 2010; Loucks et al., 2012, 2009; Desbois et al., 2009), we use AIA for quantitative analysis, with minimal user-bias, using high-resolution SEM images. From our results, there is relatively little variation between subsets of pore size distribution, which suggests repeatable results. The normal to slightly skewed distribution curves most likely reflect a close link to the grain-size distribution. The variation in porosity evident among subsets of the same sample is likely to be due to mudrock heterogeneity. Mapping the geometry of porosity distribution allows an inference about the connectivity of porosity and therefore provides a qualitative measure of permeability.

The majority of pores in mudrocks are within the microscale (< 2 nm) and mesoscale (2-50 nm) (Wang et al., 2014; Kuila and Prasad, 2013). Microporous mudrocks have a high capacity for gas adsorption (Yang et al., 2014). For this study by AIA, due to the spatial resolution limit of the SEM image, pores less than 15 nm were discounted in order to improve the signal to noise ratio. This technique, therefore, misses the greater percentage of the pores within micropore (< 2 nm) range, as well as some of the mesopores. The most common techniques currently used technique to determine pore size and porosity of mudrocks also have limitations in their measuring capabilities (Fig. 12). Nitrogen or carbon dioxide gas adsorption and NMR are suitable for measuring at the mesopore scale, whereas helium porosimetry and water intrusion porosimetry can measure the full range of pore sizes (Fig. 12). The AIA technique with SEM, however, has the added benefit of being able to place pores within their spatial context, which cannot be done with other techniques. Higher resolution SEM than used for this study and helium ion beam microscopes can provide some useful data on smaller pores, although HIB has better image clarity especially at lower kV.

Figure 12. Array of techniques for determining pore size and porosity. The variation in the measuring capabilty of one technique to the order is several orders of magnitude. The technique applied in the research for measuring pores and porosity is the scanning electron microscopy and its special resolution is limited to 5 nm.
3.2 Hemipelagites

The Shackleton Site 1385 was originally selected for construction of a marine reference section of Quaternary millennial-scale climate variability (Hodell et al., 2013; Stow et al., 2013). What was required was a site with a continuous sedimentary record and relatively 'high' rates of hemipelagic sedimentation to allow better resolution. Site 1385 was located at a water depth of 2 589 m on a spur of the Iberian margin, away from any potential turbidity current pathways and outside the slope region affected by the Mediterranean Outflow Water bottom currents. The 160 m section cored satisfied the criteria of a continuous sedimentary record, but it is also important to understand in detail the nature of sediments recovered at this marine reference site.

3.2.1 Classical hemipelagites

The entire section is made up of classical hemipelagites of mixed terrigenous-biogenic composition and a silt-rich terrigenous fraction--the marl-hemipelagite category of Stow et al. (2001) and Stow and Tabrez (1998). They show no primary sedimentary structures, but pervasive bioturbation. The trace fossil assemblage documented here is characteristic of the deep-water Zoophycos ichnofacies (Uchman and Wetzel, 2011), an indicator of well-oxygenated bottom waters and pore waters as well as good availability of organic matter and other nutrients. Several tiers of burrowing are evident, which suggests the thorough mixing of sediment over a few cm of section and partial mixing up to tens of cm. Two earlier papers (Rodríguez-Tovar et al., 2015; Rodríguez-Tovar and Dorador, 2014) have presented the results of more detailed studies of trace fossils at the Shackleton Site. They note a very similar ichnofacies assemblage and tiering structure to that presented here, with the addition of Scolicia and Nereites traces, as well as evidence for periods with less well-oxygenated conditions.

The mean sedimentation rate at the Shackleton Site of just over 11 cm/ka for the past 1.4 Ma is typical of open slope hemipelagites with moderate fluvial/eolian supply of terrigenous material. The Holocene rate is slightly higher, about 20-25 cm/ka, which may reflect an increase in both biogenic and terrigenous supply during this period. The sediments are all very fine clayey silts and silty clays and moderately well sorted with a unimodal grain-size distribution, which is compatible with the mixed biogenic-terrigenous model of Stow and Tabrez (1998) and similar to the results obtained by Nishida (2016).

3.2.2 Microstructures

The sediment microstructure we observe is mainly of random or mixed (polymodal) particle alignment and a secondary preferred grain alignment. There are few prior observations in the literature, but those summarised in Stow and Tabrez (1998) suggested that open-water hemipelagites are characterised by a partially sub-parallel (preferred) alignment that is disrupted into a random or mixed fabric by bioturbation and by the presence of larger grains, such as foraminifera. Our results would tend to support this interpretation, although the high-angle preferred grain alignment in four subset samples we interpret as the re-alignment of particles along the walls of sub-vertical burrows. The slightly oblique grain alignment may also reflect the bioturbational fabric. However, Nishida (2016) found only random fabrics in the Shackleton Site hemipelagites, which he interprets as the result of the vertical aggregation of loosely-bound flocs in the absence of current shear caused by either bottom currents or turbidity currents (Nishida et al., 2013; Moon and Hurst, 1984).

We suggest that the results presented herein clearly demonstrate the small-scale and micro-scale heterogeneity of mudrock properties. This can only be effectively elucidated by large-area imaging of multiple subsets. Further work is required, particularly to compare hemipelagite, contourite and turbidite mud fabrics (Nishida, 2016) and to consider the nature of fabric reorientation with burial compaction. Our samples range in burial depth from 10 to 80 m, so that we would expect little obvious compaction effect. We do note, however, that the deepest sample (Number 5) does show the most consistent parallel grain alignment of all the samples.

3.2.3 Hemipelagic deposition

The lithology, ichnofacies, composition, grain size and microstructural character of the hemipelagites at the Shackleton Site all support their deposition by vertical settling coupled with very slow lateral advection, as invoked by previous studies (Hoogakker et al., 2004; Stow and Tabrez, 1998; Hesse, 1975). The vertical settling component comprises biogenic planktonic material, wind-derived or storm-stirred terrigenous material, and material eroded from the seafloor by the action of internal tides and waves. The lateral advection of largely terrigenous material within the water column occurs as cross-shelf nepheloid-layer transport, eddies that peel off from along-slope bottom currents, and from the finest portions of low-concentration turbidity currents that become detached along density discontinuities. Progressive seaward distribution can occur via a process of suspension cascading.

The colour cyclicity observed, which reflects variation in the detrital/biogenic ratio in the sediments, is probably caused by variation the flux of silt and clay material from the continent rather than by changes in carbonate productivity (Hodell et al., 2015). These cycles are non-regular in thickness, but on average are close to 20 ka in duration—i.e., precession cycles (Hodell et al., 2015).


Microstructure is an important attribute that controls petrophysical properties of mudrocks and consequently the movement of fluid within them as well as their seismic elasticity. Scanning electron microscopy has advanced knowledge of microstructural characteristics, but understanding this complex and highly heterogeneous sediment type is still limited. Hemipelagic muds are very common in the marine environment, and one of the principal sediment types in deep-water. This study, therefore, has used samples of hemipelagites from the Iberian margin (Shackleton Site) to both develop an improved methodology for their microstructural characterisation, and advance our knowledge of hemipelagic sedimentation in general.

This study has developed a new methodology for the fast, reliable and effective characterisation of microstructure in mudrocks. Good sample preparation techniques are the first important step. The method has the following principal advantages: (a) a high degree of automation, which increases efficiency and reduces user-input and bias; (b) maximum areal coverage by using automated tiling and stitching of images; (c) quantification of orientation measurements; and (d) the ability to derive high quality data on grain size, grain orientation, pore size and porosity at the same time.

There are some disadvantages, which include: (a) the requirement for very large datasets (gigabytes per image), which has implications for computer memory and data handling; (b) the need for image segmentation, which is a non-trivial exercise requiring user input; (c) lack of resolution at the nano-scale for micropore characterisation. Further refinement of the technique and the use of higher-resolution instrumentation should help resolve these issues. The method also requires development to allow better compositional analysis.

The study has also provided important data on the nature of hemipelagic sedimentation at the Shackleton Site (IODP Site 1385) on the Iberian margin, and their deposition from a combination of vertical settling and very slow lateral advection. The sediments are typical mixed biogenic-terrigenous, silt-rich hemipelagites, with a tiered and diverse trace fossil assemblage of the deep-water Zoophycos ichnofacies. Sedimentation has been continuous at the site at an average rate of accumulation of 11 cm/ka over the past 1.4 Ma, and showing a pronounced but non-regular colour cyclicity. The two thin zones of contorted sediment do not appear to disrupt this continuity. The site is well-placed to become a valuable marine reference section for Quaternary climate change (Hodell et al., 2015). The sediment microstructures show small-scale heterogeneity in all properties, and an overall random fabric with secondary preferred grain-alignment. These results on the fabric differ, in part, from previous studies and suggest that further work is required on the microstructure of hemipelagites and their comparison with different deep-water sediment facies.


This research is part of Shereef Bankole's PhD programme at Heriot-Watt University, Edinburgh, United Kingdom. He appreciates the sponsorship received from Petroleum Technology Development Fund, Nigeria. The authors are grateful to Mark Curtis of the University of Oklahoma, United States of America for preparing the ion-milled samples. We thank the International Ocean Discovery Program for giving access to the core samples and Shereef is grateful to IODP technical staff (Walter Hale and Alex Wülbers) for their guidance and co-operation during the core sampling at MARUM IODP repository, Bremen, Germany. The final publication is available at Springer via

Aplin, A. C., Macquaker, J. H. S., 2011. Mudstone Diversity: Origin and Implications for Source, Seal, and Reservoir Properties in Petroleum Systems. AAPG Bulletin, 95(12): 2031-2059. DOI:10.1306/03281110162
Bankole, S. A., Buckman, J., Stow, D., et al., 2019. Grain-Size Analysis of Mudrocks: A New Semi-Automated Method from SEM Images. Journal of Petroleum Science and Engineering, 174: 244-256. DOI:10.1016/j.petrol.2018.11.027
Bankole, S. A., Stow, D. A. V., Lever, H., et al., 2016. Microstructure of Mudrock and the Choice of Representative Sample, In: Fifth EAGE Shale Workshop. EAGE, Catania, Italy
Berens, P., 2009. CircStat: AMATLAB Toolbox for Circular Statistics. Journal of Statistical Software, 31(10): 1-21. DOI:10.18637/jss.v031.i10
Bosl, W. J., Dvorkin, J., Nur, A., 1998. A Study of Porosity and Permeability Using a Lattice Boltzmann Simulation. Geophysical Research Letters, 25(9): 1475-1478. DOI:10.1029/98gl00859
Buckman, J., 2014. Use of Automated Image Acquisition and Stitching in Scanning Electron Microscopy: Imaging of Large Scale Areas of Materials at High Resolution. Microscopy and Analysis, 28: 13-15.
Buckman, J., Bankole, S., Zihms, S., et al., 2017. Quantifying Porosity through Automated Image Collection and Batch Image Processing: Case Study of Three Carbonates and an Aragonite Cemented Sandstone. Geosciences, 7(3): 70. DOI:10.3390/geosciences7030070
Camp, W. K., Diaz, E., Wawak, B. E., 2013. Electron Microscopy of Shale Hydrocarbon Reservoirs. Association of Petroleum Geologists, Tulsa
Chambers, J. M., Cleveland, W. S., Kleiner, B., et al., 1984. Graphical Methods for Data Analysis. Journal of the Royal Statistical Society, 147(3): 513.
Curtis, M. E., Sondergeld, C. H., Ambrose, R. J., et al., 2012. Microstructural Investigation of Gas Shales in Two and Three Dimensions Using Nanometer-Scale Resolution Imaging. AAPG Bulletin, 96(4): 665-677. DOI:10.1306/08151110188
Curtis, M. E., Ambrose, R. J., Sondergeld, C. H., et al., 2010. Structural Characterization of Gas Shales on the Micro- and Nano-Scales, Society of Petroleum Engineers—Canadian Unconventional Resources and International Petroleum Conference, Calgary
Davis, J. C., 1986. Statistics and Data Analysis in Geology, Wiley India. 656.
Desbois, G., Urai, J. L., Kukla, P. A., 2009. Morphology of the Pore Space in Claystones—Evidence from BIB/FIB Ion Beam Sectioning and Cryo-SEM Observations. eEarth, 4(1): 15-22. DOI:10.5194/ee-4-15-2009
DeVasto, M. A., Czeck, D. M., Bhattacharyya, P., 2012. Using Image Analysis and ArcGIS® to Improve Automatic Grain Boundary Detection and Quantify Geological Images. Computers & Geosciences, 49: 38-45. DOI:10.1016/j.cageo.2012.06.005
Fisher, N. I., 1993. Statistical Analysis of Circular Data. Cambridge University Press, New York, 277. https: //
Francus, P., Pirard, E., 2004. Testing for Sources of Errors in Quantitative Image Analysis. In: Francus, P., ed., Image Analysis, Sediments and Paleoenvironments. Springer Netherlands, Dordrecht. 87-102. https: //
Grove, C., Jerram, D. A., 2011. JPOR: An ImageJ Macro to Quantify Total Optical Porosity from Blue-Stained Thin Sections. Computers & Geosciences, 37(11): 1850-1859. DOI:10.1016/j.cageo.2011.03.002
Hemes, S., Desbois, G., Urai, J. L., et al., 2013. Variations in the Morphology of Porosity in the Boom Clay Formation: Insights from 2D High Resolution BIB-SEM Imaging and Mercury Injection Porosimetry. Netherlands Journal of Geosciences, 92(4): 275-300. DOI:10.1017/s0016774600000214
Hesse, R., 1975. Turbiditic and Non-Turbiditic Mudstone of Cretaceous Flysch Sections of the East Alps and other Basins. Sedimentology, 22(3): 387-416. DOI:10.1111/j.1365-3091.1975.tb01638.x
Hodell, D. A., Lourens, L., Stow, D. A. V., et al., 2013. The "Shackleton Site" (IODP Site U1385) on the Iberian Margin. Scientific Drilling, 16: 13-19. DOI:10.5194/sd-16-13-2013
Hodell, D. A., Lourens, L., Crowhurst, S., et al., 2015. A Reference Time Scale for Site U1385 (Shackleton Site) on the SW Iberian Margin. Global and Planetary Change, 133: 49-64. DOI:10.1016/j.gloplacha.2015.07.002
Hoogakker, B. A. A., Rothwell, R. G., Rohling, E. J., et al., 2004. Variations in Terrigenous Dilution in Western Mediterranean Sea Pelagic Sediments in Response to Climate Change during the Last Glacial Cycle. Marine Geology, 211(1/2): 21-43. DOI:10.1016/j.margeo.2004.07.005
Houben, M. E., Desbois, G., Urai, J. L., 2013. Pore Morphology and Distribution in the Shaly Facies of Opalinus Clay (Mont Terri, Switzerland): Insights from Representative 2D BIB-SEM Investigations on mm to nm Scale. Applied Clay Science, 71: 82-97. DOI:10.1016/j.clay.2012.11.006
Janssen, C., Kanitpanyacharoen, W., Wenk, H. R., et al., 2012. Clay Fabrics in SAFOD Core Samples. Journal of Structural Geology, 43: 118-127. DOI:10.1016/j.jsg.2012.07.004
Josh, M., Esteban, L., Delle Piane, C., et al., 2012. Laboratory Characterisation of Shale Properties. Journal of Petroleum Science and Engineering, 88-89: 107-124. DOI:10.1016/j.petrol.2012.01.023
Kameda, A., Dvorkin, J., Keehm, Y., et al., 2006. Permeability-Porosity Transforms from Small Sandstone Fragments. Geophysics, 71(1): N11-N19. DOI:10.1190/1.2159054
Keller, L. M., Schuetz, P., Erni, R., et al., 2013. Characterization of Multi-Scale Microstructural Features in Opalinus Clay. Microporous and Mesoporous Materials, 170: 83-94. DOI:10.1016/j.micromeso.2012.11.029
Kuila, U., Prasad, M., 2013. Specific Surface Area and Pore-Size Distribution in Clays and Shales. Geophysical Prospecting, 61(2): 341-362. DOI:10.1111/1365-2478.12028
Lemmens, H., Richards, D., 2013. Multiscale Imaging of Shale Samples in the Scanning Electron Microscope. American Association of Petroleum Geologists Memoir: 27-35. DOI:10.1306/13391702M1023582
Lonardelli, I., Wenk, H. R., Ren, Y., 2007. Preferred Orientation and Elastic Anisotropy in Shales. Geophysics, 72(2): D33-D40. DOI:10.1190/1.2435966
Loucks, R. G., Reed, R. M., Ruppel, S. C., et al., 2009. Morphology, Genesis, and Distribution of Nanometer-Scale Pores in Siliceous Mudstones of the Mississippian Barnett Shale. Journal of Sedimentary Research, 79(12): 848-861. DOI:10.2110/jsr.2009.092
Loucks, R. G., Reed, R. M., Ruppel, S. C., et al., 2012. Spectrum of Pore Types and Networks in Mudrocks and a Descriptive Classification for Matrix-Related Mudrock Pores. AAPG Bulletin, 96(6): 1071-1098. DOI:10.1306/08171111061
Lovie, S., 2005. Empirical Quantile-Quantile Plots. In: Encyclopedia of Statistics in Behavioral Science. John Wiley and Sons, Ltd., Chichester. 543-545. https: //
Macquaker, J. H. S., Howell, J. K., 1999. Small-Scale (< 5.0 M) Vertical Heterogeneity in Mudstones: Implications for High-Resolution Stratigraphy in Siliciclastic Mudstone Successions. Journal of the Geological Society, 156(1): 105-112. DOI:10.1144/gsjgs.156.1.0105
Mardia, K. V., Jupp, P. E., 2008. Directional Statistics, Directional Statistics. John Wiley and Sons, Inc. 432. https: //
Martínez-Nistal, A., Veniale, F., Setti, M., et al., 1999. A Scanning Electron Microscopy Image Processing Method for Quantifying Fabric Orientation of Clay Geomaterials. Applied Clay Science, 14(4): 235-243. DOI:10.1016/s0169-1317(98)00055-6
Moon, C. F., Hurst, C. W., 1984. Fabric of Muds and Shales: An Overview. Geological Society, London, Special Publications, 15(1): 579-593. DOI:10.1144/gsl.sp.1984.015.01.36
Munson, E. O., Chalmers, G. R. L., Bustin, R. M., et al., 2016. Utilizing Smear Mounts for X-Ray Diffraction as a Fully Quantitative Approach in Rapidly Characterizing the Mineralogy of Shale Gas Reservoirs. Journal of Unconventional Oil and Gas Resources, 14: 22-31. DOI:10.1016/j.juogr.2016.01.001
Nishida, N., 2016. Microstructure of Muddy Contourites from the Gulf of Cádiz. Marine Geology, 377: 110-117. DOI:10.1016/j.margeo.2015.08.017
Nishida, N., Ito, M., Inoue, A., et al., 2013. Clay Fabric of Fluid-Mud Deposits from Laboratory and Field Observations: Potential Application to the Stratigraphic Record. Marine Geology, 337: 1-8. DOI:10.1016/j.margeo.2012.12.006
Pal, N. R., Pal, S. K., 1993. A Review on Image Segmentation Techniques. Pattern Recognition, 26(9): 1277-1294. DOI:10.1016/0031-3203(93)90135-j
Pickering, K. T., Hiscott, R. N., 2015. Deep Marine Systems: Processes, Deposits, Environments, Tectonics and Sedimentation. John Wiley and Sons, Chichester. 657
Piper, D. J. W., 1977. Manual of Sedimentological Techniques. Departments of Geology and Oceanography, Dalhousie University, Halifax
Rodríguez-Tovar, F. J., Dorador, J., 2014. Ichnological Analysis of Pleistocene Sediments from the IODP Site U1385 "Shackleton Site" on the Iberian Margin: Approaching Paleoenvironmental Conditions. Palaeogeography, Palaeoclimatology, Palaeoecology, 409: 24-32. DOI:10.1016/j.palaeo.2014.04.027
Rodríguez-Tovar, F. J., Dorador, J., Grunert, P., et al., 2015. Deep-Sea Trace Fossil and Benthic Foraminiferal Assemblages across Glacial Terminations 1, 2 and 4 at the "Shackleton Site" (IODP Expedition 339, Site U1385). Global and Planetary Change, 133: 359-370. DOI:10.1016/j.gloplacha.2015.05.003
Saraji, S., Piri, M., 2015. The Representative Sample Size in Shale Oil Rocks and Nano-Scale Characterization of Transport Properties. International Journal of Coal Geology, 146: 42-54. DOI:10.1016/j.coal.2015.04.005
Schindelin, J., Arganda-Carreras, I., Frise, E., et al., 2012. Fiji: An Open-Source Platform for Biological-Image Analysis. Nature Methods, 9: 676-682. DOI:10.1038/nmeth.2019
Schindelin, J., Rueden, C. T., Hiner, M. C., et al., 2015. The ImageJ Ecosystem: An Open Platform for Biomedical Image Analysis. Molecular Reproduction and Development, 82(7/8): 518-529. DOI:10.1002/mrd.22489
Schneider, C. A., Rasband, W. S., Eliceiri, K. W., 2012. NIH Image to ImageJ: 25 Years of Image Analysis. Nature Methods, 9(7): 671-675. DOI:10.1038/nmeth.2089
Sing, K. S. W., Everett, D. H., Haul, R. A. W., et al., 1985. Reporting Physisorption Data for Gas Solid Systems with Special Reference to the Determination of Surface-Area and Porosity (Recommendations 1984). Pure Applied Chemistry, 57: 603-619. DOI:10.1351/pac198557040603
Sokolov, V. N., O'Brien, N. R., 1990. A Fabric Classification of Argillaceous Rocks, Sediments, Soils. Applied Clay Science, 5(4): 353-360. DOI:10.1016/0169-1317(90)90030-s
Stow, D. A. V., 1985. Fine-Grained Sediments in Deep Water: An Overview of Processes and Facies Models. Geo-Marine Letters, 5(1): 17-23. DOI:10.1007/bf02629792
Stow, D. A. V., Hernández-Molina, F. J., Llave, E., et al., 2013. The Cadiz Contourite Channel: Sandy Contourites, Bedforms and Dynamic Current Interaction. Marine Geology, 343: 99-114. DOI:10.1016/j.margeo.2013.06.013
Stow, D. A. V., Huc, A. Y., Bertrand, P., 2001. Depositional Processes of Black Shales in Deep Water. Marine and Petroleum Geology, 18(4): 491-498. DOI:10.1016/s0264-8172(01)00012-5
Stow, D. A. V., Tabrez, A. R., 1998. Hemipelagites: Processes, Facies and Model. Geological Society, London, Special Publications, 129(1): 317-337. DOI:10.1144/gsl.sp.1998.129.01.19
Stow, D. A. V., 2005. Sedimentary Rocks in the Field: A Color Guide. Taylor and Francis Group, Florida. 320
Suttle, M. D., Genge, M. J., Russell, S. S., 2017. Shock Fabrics in Fine-Grained Micrometeorites. Meteoritics & Planetary Science, 52(10): 2258-2274. DOI:10.13039/501100000271
Tovey, N. K., Smart, P., Hounslow, M. W., et al., 1992. Automatic Orientation Mapping of Some Types of Soil Fabric. Geoderma, 53(3/4): 179-200. DOI:10.1016/0016-7061(92)90054-b
Uchman, A., Wetzel, A., 2011. Deep-Sea Ichnology: The Relationships between Depositional Environment and Endobenthic Organisms. Developments in Sedimentology, 63: 517-556. DOI:10.1016/B978-0-444-53000-4.00008-1
Vanden-Bygaart, A. J., Protz, R., 1999. The Representative Elementary Area (REA) in Studies of Quantitative Soil Micromorphology. Geoderma, 89(3/4): 333-346. DOI:10.1016/s0016-7061(98)00089-5
Wang, Y., Zhu, Y. M., et al., 2014. Characteristics of the Nanoscale Pore Structure in Northwestern Hunan Shale Gas Reservoirs Using Field Emission Scanning Electron Microscopy, High-Pressure Mercury Intrusion, and Gas Adsorption. Energy & Fuels, 28(2): 945-955. DOI:10.1021/ef402159e
Wenk, H. R., Houtte, P. V., 2004. Texture and Anisotropy. Reports on Progress in Physics, 67(8): 1367-1428. DOI:10.1088/0034-4885/67/8/r02
Wenk, H. R., Lutterotti, L., Kaercher, P., et al., 2014. Rietveld Texture Analysis from Synchrotron Diffraction Images. Ⅱ. Complex Multiphase Materials and Diamond Anvil Cell Experiments. Powder Diffraction, 29(3): 220-232. DOI:10.1017/s0885715614000360
Wenk, H. R., Voltolini, M., Martin, M., et al., 2008. Preferred Orientations and Anisotropy in Shales: Callovo-Oxfordian Shale (France) and Opalinus Clay (Switzerland). Clays and Clay Minerals, 56(3): 285-306. DOI:10.1346/ccmn.2008.0560301
Yang, F., Ning, Z. F., Liu, H. Q., 2014. Fractal Characteristics of Shales from a Shale Gas Reservoir in the Sichuan Basin, China. Fuel, 115: 378-384. DOI:10.1016/j.fuel.2013.07.040
Zaitoun, N. M., Aqel, M. J., 2015. Survey on Image Segmentation Techniques. Procedia Computer Science, 65: 797-806. DOI:10.1016/j.procs.2015.09.027