Word Template for Authors, EIAS Style B s1

Development of an Analytical Method for Ground-based Multi-spectral Mapping: The Deepwater Deposits at Big Rock Quarry, Arkansas

Mariana I. Olariu*†, John F. Ferguson†, Carlos L.V. Aiken†,

and Mohamed G. Abdel-Salam‡

†Geosciences Department, University of Texas at Dallas, P.O. Box 830688, FO 21, Richardson, TX 75083-0688

‡University of Missouri-Rolla, Rolla, MO 65409-0410

This study assesses how well and under what conditions sandstone may be detected versus shale in a deepwater sedimentary succession using ground-based remote sensing techniques. Multi-spectral images of the outcrop at Big Rock, Arkansas were co-registered and displayed in Red-Green-Blue colour space to create false colour images that are useful for highlighting lithologic variation. Remote sensing analysis for geologic mapping is common, but here it is applied to digital acquisition from the ground at close range, obliquely to demonstrate its use in detailed outcrop mapping. Analyses of the spectral characteristics of sandstone and shale samples over the visible, near infrared and thermal infrared part of the electromagnetic spectrum identified spectral responses of these sedimentary rocks indicating that these rocks can be discriminated using the information contained over the thermal interval. Normal color images were acquired using a conventional high-resolution digital camera and infrared images using a thermal infrared camera. Recently available, handheld, infrared cameras have a spectral range of 1 to 20 µm and can map rocks at wavelengths that usually can not be used when remotely measuring spectra through the Earth’s atmosphere due to the presence of absorption bands. However from the ground this is possible and a suite of images from visible through thermal infrared of this outcrop have been used to discriminate lithology.

Keywords: multi-spectral, ground-based infrared thermal image, deepwater, lithologic mapping

*Corresponding author. Email:

1. Introduction

In this study, multi-spectral imaging is used to identify lithological units on outcrops. These methods normally applied in remote sensing from high altitude airborne and space platforms are modified here for application to close range imaging.

Imaging spectroscopy is a technique used to spectrally identify and spatially map rocks based on their specific chemical/mineralogical composition. Today spectrometers are used in the laboratory, in the field, on aircrafts, and on satellites. Absorption bands in the atmosphere (due to water vapour, carbon dioxide) limit the use of spectral data in remote sensing (figure 1). Therefore laboratory spectrometers will have a higher spectral resolution compared to spectrometers that remotely measure spectra through the Earth’s atmosphere (Floyd 1997). However, these spectral regions can be used at close range outcrops, since the atmospheric path lengths are shorter.

Technological advances now provide automatic acquisition of satellite (ASTER, LANDSAT) and airborne (SEBASS - Spatially Enhanced Broadband Array Spectrograph System, TIMS - Thermal Infrared Multi-spectral Scanner) photography, but the remotely sensed data are usually too coarsely sampled (low resolution) for geological outcrop study purposes. Even when instruments are flown at low to medium altitudes the acquired data is at a spatial resolution of tens of meters/pixel (Hubbard 1998) or at the best meters/pixel (Smailbegovic 2000).

Detailed geological features exposed on near-vertical cliff faces are difficult to image from overhead. Newly developed handheld thermal infrared cameras are now available for the large public, but yet they have not been widely used for geologic purposes. There are some studies that have successfully used infrared thermal photography acquired from the ground to map high temperature features such as lava lakes (Oppenheimer and Yirgu 2002) or for identifying hydraulically active fractures (Rosenbom and Jacobsen 2005). To our knowledge this is the first study to map sedimentary features using ground-based thermal infrared imagery.

2. Test Site: Big Rock Quarry, Arkansas

2.1. Geologic Setting

Big Rock Quarry is located in the south-eastern part of the Ouachita Mountains along the north bank of the Arkansas River in North Little Rock, Arkansas. The cliff faces of Big Rock Quarry expose a three-dimensional view of the lower part of the upper Jackfork Group (Jordan et al. 1993). The exposure is oriented at different angles and is up to 60 m high and almost 1 km long (figure 2).

The Jackfork Group is a succession (about 2000 meters thick) of deep-water sedimentary rocks that were deposited in the Ouachita Basin during late Carboniferous. Sediments were derived mainly from the northern and eastern shelves (Coleman Jr. 2000).

The four facies that crop out at Big Rock Quarry are represented by massive to parallel-laminated, fine-grained sandstone, shale intraclast breccia with a sandy matrix, shale intraclast breccia with a shale matrix, and finely laminated shale (Link and Stone 1986; Cook 1993).

Matrix-supported breccias are the product of cohesive debris flow. Clasts consist of deepwater sediments eroded by the gravity flows during transport. These beds commonly show erosional base.

Amalgamated, thick to very thick tabular to irregular-bedded, fine-grained, and massive sandstones were deposited from high-concentration turbidity currents. Individual turbidite beds rarely exceed 30 cm in thickness, but they commonly occur in amalgamated units that have an average thickness of 1-2 m and frequently attain thickness of 6 to 8 m.

Thin bedded, massive to planar stratified, fine–grained sandstone deposited from low-density turbidity currents are separated by discontinuous centimetre scale siltstone and shale. Sandstone bed thickness ranges from centimetres to 1 meter.

3. Methodology

In order to accomplish this study a specific succession of steps had to be followed. Rock samples were collected from the outcrop and their reflectance as well as emittance spectra measured in the lab. Multi-spectral analysis is required to identify spectral characteristics of the main types of rocks exposed at the outcrop. Ground-based thermal infrared photography is integrated with conventional photography to identify lithological units on outcrops. Finally digital image analysis is performed using a principal component transformation.

3.1. Multi-spectral Analysis

Reflectance and emittance spectroscopy are sensitive to specific chemical bonds in rocks. Every spectral feature is due to an interaction of photons of particular energy with the electrons (atoms) in the rock. At different wavelengths, photon interaction gives rise either to absorption, transmission, or scattering. The nature of the absorption is unique to the specific chemical structure and therefore diagnostic for a particular mineral. Absorption features usually are concentrated in limited ranges of wavelength and between them are portions of the spectrum that contain little information (Clark et al. 1990).

A reflectance and an emission spectrum of sandstone and shale samples collected from Big Rock have been acquired in the laboratory using a Geophysical and Environmental Research (GER) 3700 spectrometer (spectral range from 0.35 to 2.5 μm) and a GX Fourier Transform Infra Red (FTIR) spectrometer (spectral range from 2 to 14 μm).

The GER 3700 spectrometer has a spectral sample interval of 1.5 nm over the range 350-1050 nm (spectral resolution 3 nm); 6.5 nm over the range 1050-1900 nm (spectral resolution 11 nm); and 9.5 nm from 1900 to 2500 nm (spectral resolution 16 nm). Two measurements were required in order to determine the reflectance of the rock samples: the spectral response of a reference sample and the spectral response of the sample. The reflectance spectrum was computed by dividing the spectral response of each rock sample by that of the reference sample.

The Spectrum GX FTIR spectrometer is capable of collecting spectra in the near and mid-infrared region with a spectral resolution of better than 1.5 nm. The Spectrum GX has a microscope accessory with a liquid nitrogen cooled detector that can be used in either transmittance or reflectance mode. The solid sample was finely pulverized with pure, dry Potassium Bromide (KBr), the mixture was pressed in a hydraulic press to form a transparent pellet, and the spectrum of the pellet measured. Potassium Bromide is commonly used for infrared transmission windows within FTIR spectrophotometers because KBr has no absorptions in the infrared above 0.4 μm.

In order to explore the spectral signatures of rock samples, a combination (figure 3) of different data analysis methods (least square modelling, nonlinear smoothing, and principal component analysis) has been applied. This way, the spectral signatures of rock samples are related to their distinct mineralogical compositions.

Reflection spectra are commonly dominated by surface effects and scattering phenomena that can soften spectral signatures. The source strength is controlled by sunlight in the reflective region and target temperature in the emissive region and in both cases the atmosphere contaminates the spectra. Even in the lab the amount of light that hits the sample controls the intensity of return.

First the data is smoothed with a running median smoother (4253H) and decimated. 4253H consists of a running median of 4, then 2, then 5, then 3 followed by Hanning. Hanning is a running weighted average, the weights being 1/4, 1/2 and 1/4. Nonlinear smoothers were used instead of linear smoothers, such as running means because they are resistant to outliers and remove narrow “spikes” (figure 3(a)). Since the intensity of reflected/emitted radiation is directly proportional to the incident ‘light’, the observations were converted to a logarithmic form to overcome the multiplicative effect of the ‘light’ source. This logarithmic transformation helped to better visualize trends in the data set (figure 3(b)). Then the data is modeled by a combination of the second degree polynomial trend plus sinus and cosines terms of wavelengths (figure 3(c)). The model contains a constant term that adjusts for individual reflectance level differences. Subtraction of additive model removes multiplicative effect of illumination (figure 3(d)). Discrimination is based on the difference in spectral shape. Therefore we standardized the amplitudes ofeachsamplebysubtractingthemedianacrossall wavelengthsanddivided by the standard deviationacrosswavelengths tomakethesamples comparable.

A principal component analysis (PCA) was performed to isolate independent components in the spectra. Principal component analysis transforms a number of correlated variables into a smaller number of uncorrelated variables called principal components and possibly isolates spectra typical of shale and sandstone rock types. The first principal component accounts for most of the variance in the spectra and each succeeding component accounts for decreasing percentages of the variance (Davis 1986). In our case because the number of variables is large only the first PCs with eigenvalues contributing with more than 1% to the total variance were considered.

3.2. Ground-based Infrared Thermal Imaging

There are two principal types of detectors that are used for thermal imaging—photovoltaic and thermal. Photovoltaic IR detectors produce electric current in proportion to the number of incident photons that are shorter than a threshold wavelength. Thermal detectors make use of the changes in material properties, most commonly resistance, as absorbed light heats the lattice of the detector material (Matthews 2004).

Photovoltaic arrays are more sensitive—by an order of magnitude or more—than thermally-based detectors. However a major drawback to these detectors is that they must be cooled to cryogenic temperatures (77 K). To maintain these very low temperatures, the detectors are enclosed in a Dewar with a window transparent at the required infrared wavelengths. However, such detectors are not yet available for the large public at wavelengths higher than 11 µm.

Thermal detectors have a significant advantage over photovoltaic types: they do not require cryogenic cooling. While less sensitive than photovoltaic detectors, microbolometer pixels will readily respond to a temperature change of less than 0.1° C (Matthews 2004). Uncooled microbolometer arrays are now commonly available in commercial imaging systems.

Multi-spectral images of the outcrop at Big Rock have been acquired using a conventional digital camera for the visible range and a PV320 digital camera for the thermal infrared interval (2-14 μm). The PV320 has an uncooled microbolometer focal plane array detector. Radiometric calibration of the PV320 camera is done using internal blackbody source references. This method does not account for the intervening atmosphere emitting radiant energy into the instantaneous field of view (IFOV) of the sensor system or absorbing energy emitted from the ground before it reaches the detector. Therefore the data must be at a high enough pixel resolution to discriminate.

The radiometric, spatial and spectral resolutions of the thermal infrared imaging system control the detection and mapping of geologic features at different scales. There is a trade-off between resolution and noise amplification: the poorest value of resolution corresponds to the maximum value of noise, and decreasing the noise necessarily produces a higher resolution. There is no way that both quantities can be improved simultaneously (Milman 1999). Therefore slicing the spectrum into narrower bands can give more information about surface composition, but sacrifices resolution.

In our case we were concerned about the individual shale layers which typically have a thickness of less than 10 - 15 cm. The question was whether or not the signal from a pixel of 10 by 10 cm and a band pass of 200 nm will be strong enough to be recorded at the camera sensor.

The thermal band and the three visible bands were co-registered and displayed in Red-Green-Blue (RGB) colour space to create false colour images that are useful for highlighting lithologic variation of the outcrop.

Discrimination of different lithology based on the spectral characteristics was also made using other image processing techniques, such as principal component analysis. The principal component transformation is effective in enhancing information present in the scene when there is high correlation between bands. Therefore it is commonly used to compress multi-spectral data sets. Each successive principal component image accounts for a progressively smaller proportion of the variation of the original multi-spectral dataset. Since principal components are orthogonal (uncorrelated) the spectral differences between different types of rocks may be more apparent in the PC images than in individual bands.

4. Results

4.1. Spectral Analysis

4.1.1. Analysis of the Reflectance Spectrum

Reflectance spectra of 51 samples (34 sandstones and 17 shales) collected from Big Rock were measured with a GER 3700 spectrometer. Repeated measurements have been done for each sample and the results compared. Apart from an overall albedo change (due to variations in light intensity and to the roughness of surfaces which cause scattering to occur) the spectra did not significantly differ (figure 4(a)).

In order to investigate the spectral characteristics of the two types of rocks the data was modeled using a combination of second degree polynomials and Fourier series of order 15 that were ulterior removed.