Analysis of hyperspectral and lidar data: Remote optical mineralogy and fracture identification

It is the quest of the geologist to convey complex, three-dimensional relationships quantitatively in both space and time. Passive methods such as visible-light photography and multi-even hyperspectral imagery offer a semiquantitative template to understand natural systems but collapse the Earth into a flat map view. In recent years, active remote-sensing techniques that include laser scanning have become a reality for many geologists. The union of high-resolution spatial data and traditional field methods has inspired numerous projects, publications, and international conferences in the last decade (the GSA Penrose Conference held in Durham, UK, in September 2006; Manchester meeting in 2007; Bellian et al. 2005 and the references therein). From the early days of creating coarse-grid outcrop models with surveying instruments (Xu et al., 1999, 2000; Adams et al., 2004, 2005) at a rate of a few hundred points per day to many tens to hundreds of thousands of points per second, the spatial resolution of laser-based surface models has increased by four orders of magnitude. Lidar (light detection and ranging) has drastically changed the way geoscientists collect quantitative field data. Application of this technology parallels the innovation of photography versus a field-book sketch. We can now rotate new-age “three-dimensional photographs” in space and extract quantitative measurements directly. Moreover, computer simulations can be constructed directly from outcrop measurements that constrain them—data from the real world.

Currently available hyperspectral imagers that include the Hyperion satellite-based system and the AVIRIS (airborne visible infrared imaging spectrometer) aircraft-based system collect data in discrete bands of the electromagnetic spectrum and can be used to determine the chemical composition of a target. The natural light reflected off the target (passive imaging) can be recorded by instruments like those discussed in this paper from the lower end of the visible-light spectrum (Fig. 1) through the short wave infrared (SWIR). The advantage of this technology is that it offers the user the ability to determine anomalies in the target material within and above the visible range of the electromagnetic spectrum very quickly and effectively. For this project, we constructed a spectral library of target materials using both field and laboratory spectral measurements. From this rock catalog, it was possible to delineate variations in carbonate mineralogy, including calcite and dolomite composition.

The fusion of DEM (digital elevation model), ground-based lidar DOMs (digital outcrop models sensu Bellian et al., 2005), and hyperspectral data were used to map dolomitization patterns exposed in the Upper and Lower Ordovician Montoya and El Paso Groups (Fig. 2) in the Franklin Mountains of El Paso, Texas. The study area (Fig. 3 and Animation 1,¹) is exceptional for testing this technique because the dolomitization patterns are well documented (Lucia, 1968, 1995; Kerans, 1988; Tillotson, 2003) and usually follow a visible color change. The two common phases of dolomitization in the Franklin Mountains are an earlier phase related to the tidal-flat environment and later karst-related breccia dolomitization (Lucia, 1968, 1995; Stepanek, 1988). The hyperspectral map (Fig. 4) was field verified and also compared with the original unpublishedfield maps of Lucia (summarized in Lucia, 1995), and an excellent correlation was observed. Ground-based lidar data were acquired and processed in Transition Canyon (Fig. 3 and Animation 1) and were compared with Lucia's interpretation and recent work (Tillotson, 2003), and again, an excellent correlation between dolomitized collapse breccia and textural analysis of the lidar-based surface was confirmed.

The findings of this study highlight the need for development of a compact, light-weight, full-range spectral imager for use in conjunction with laser scanning in a field setting. Our experience has shown that the combination of high-resolution shape or texture analysis with multispectral and hyperspectral imagery helps us to see and better understand the geology and geochemistry of the target. Current technology available from both satellite and airborne platforms allows us to conduct visible through upper–short-wave infrared (Fig. 1) spectral analyses (.35–2.5 μm) at the multimeter scale, as utilized in this study, although the imagery is too coarse in spatial resolution (20- to 30-m pixels) for more detailed stratigraphic mapping and is acquired from the wrong orientation (map view) to be fully exploited in outcrop settings.

A 10-m DEM (digital elevation model) from the online U.S. Geological Survey National Elevation Database (http://seamless.usgs.gov) was extracted for the Franklin Mountain area (Animation 1) and merged with higher spatial resolution, ground-based lidar data from Transition Canyon (Fig. 5A)—a canyon along the west side of the mountain. An Optech ILRIS-3D time of flight lidar scanner was used for field-data acquisition. Measured points were aligned using Polyworks (InnovMetric, Ontario, Canada). Lidar textural processing was conducted at the Bureau of Economic Geology using a combination of Polyworks and custom internal software.

A series of ten planar extractions were conducted, effectively “slicing” the lidar data in Transition Canyon into discrete segments (Fig. 5B). The planes defined to slice the data were created by picking points from the lidar data in the undisturbed limestone section (Fig. 5A left side). The plane was then extended to slice through all the data in the Montoya and El Paso Groups, and all lidar points within ±5 cm were extracted and exported into a new point set. This process was repeated for each of the ten segments and is displayed in Figure 5B. In addition, visible fractures in the Montoya (Fig. 5A) were digitized and displayed (Fig. 5B) to highlight fracture intensity above collapse breccia. Since it is difficult to see this relationship in a flat image, the extracted point sets have been posted as a text file for download (TransitionCanyon.txt; see footnote 1).

Spectral signatures of limestone and dolomite in Transition Canyon were collected on the basis of field mapping by Lucia (1968, 1995) and Tillotson (2003), who noted distinct diagenetic alteration of limestone host rock in and around the vicinity of collapse breccias in the Great McKelligon Sag and Transition Canyon. Each spectral sample was verified using 10% HCl in the field to confirm limestone or dolomite presence after each spectral measurement, and a sample for later petrographic analysis was taken. Spectral anomalies were noted directly in the field from the limestone and dolomite sections away from and within the breccia zone using a 350- to 2500-nm spectroradiometer (ASD FieldSpec®). These measurements were used as a natural-light, field calibration of rock type. In addition, hand samples were collected from the Great McKelligon Sag on the opposite side of the mountain to the east (Fig. 3) and then were scanned under controlled, simulated-sunlight conditions in the laboratory using the same spectroradiometer as in the field. From these data, we constructed a rock spectral library or “rock catalog” to better constrain the specific spectral bands that would be useful for limestone-dolomite discrimination over the study area (Animation 1).

AVIRIS (aircraft-mounted) hyperspectral image data were acquired in 1999 at 20-m pixel resolution by National Aeronautics and Space Administration's ER-2 aircraft and were contributed to the project by the Jet Propulsion Laboratory (JPL) Graduate Research Program. The AVIRIS data were a subset to the SWIR (2000–2500 nm) and were atmospherically corrected using ATREM software (Cooperative Institute for Research in Environmental Sciences, University of Colorado) before artifact suppression in ENVI 4.1 (ITT Visual Information Solutions, Boulder, Colorado). The atmospherically corrected imagery was distilled by forward principal component analysis (FPCA) to isolate the least correlative variables. Principal components 2, 5, and 6 were displayed as RGB (red, green, and blue), respectively, resulting in limestone dominating the green band and dolomite dominating the red and blue bands. Red-purple pixels are therefore most likely dolomite, and green are most likely limestone (Fig. 4). We found the SWIR spectral range to be most effective with regard to limestone versus dolomite delineation, which also offered an independent spectral validation by excluding the visible spectrum from the spectral analysis.

The Lower Ordovician El Paso Group exposed in the southernmost Franklin Mountains was deposited on a giant carbonate bank (Fig. 6) that accumulated mixed carbonates and siliciclastics for ∼20–25 m.y. (Stacy et al., 1992). These strata are capped by the 33-m.y. unconformity that contains the Sauk C Supersequence boundary (Lucia, 1968, 1995; Goldhammer et al., 1993). Six third-order sequences (Goldhammer et al., 1993) are present in the El Paso Group, which includes the Bowen, Hag Hill, Chamizal, McKelligon Canyon, Cindy, and Ranger Peak Formations (Fig. 7). These formations consist of a succession of carbonates and siliciclastics composed of alternating sub-tidal to tidal-flat limestone and dolomite with evidence of multiple subaerial exposure events (Wilson et al., 1993). The lowermost third-order sequence, composed of the Bowen Formation, is a mixed peritidal succession of siliciclastic sandy dolomite and marine dolowackestones. The second third-order sequence begins with the Hag Hill subtidal to peritidal thrombolitic wackestones and continues with the Chamizal peritidal to supratidal crossbedded siliciclastic-rich, algal dolowackestones. The third third-order sequence begins with the lower McKelligon Canyon Formation subtidal limestone and never quite reaches the peritidal before the deposition of the next sequence begins. This was determined on the basis of biozonation (Cloud and Barnes, 1948) and cycle stacking patterns as documented in Goldhammer et al. (1993). The fourth third-order sequence is composed of the upper subtidal McKelligon Canyon Formation and the supratidal Cindy Dolomite. The fifth sequence is the lower half of the Ranger Peak Formation composed of burrowed, dolomitized, shallow, subtidal limestone. The final sequence is the upper Ranger Peak, also burrowed, dolomitized, subtidal limestone that is truncated by the uppermost Lower Ordovician and Middle Ordovician unconformity. Ages of these sequences (Fig. 7) were assigned by Goldhammer et al. (1993) on the basis of cycle-stacking patterns, regional correlations, and biozonation. The same timescale is used here (Harland et al., 1989) for clarity.

The southern Franklin Mountains offer superb exposure of Precambrian through Silurian carbonate-shelf deposits along the southwest edge of the Diablo Platform of west Texas. Strike of the El Paso and overlying Upper Ordovician Montoya group runs northwest-southeast, and the preserved thickness of the Lower Ordovician rocks thins depositionally to the east onto the north-south–trending Diablo Arch (LeMone, 1988, Goldhammer et al., 1993). The lowermost Cambro-Ordovician Bliss sandstone through uppermost Ranger Peak (Fig. 7) shallow-water limestones and dolomites consists of a 500-m-thick (1500-ft) stack of cyclic shelf deposits that constitute the Sauk C Second-Order Supersequence Set.

A likely period of extensive erosion followed the deposition of the El Paso Group, at which time karst development is proposed to have occurred (Lucia, 1968, 1995). It is unclear at this time whether higher frequency (third-order) karst events can be confirmed within the El Paso Group or whether all dissolution was restricted to the Sauk-Tippecanoe unconformity. Normal, open-marine deposition resumed in the Late Ordovician, as recorded by the Montoya Group, apparently in stratigraphic-bedding continuity of the El Paso Group (Fig. 7). Deposition of the Montoya was also terminated by a period of exposure and karst development related to the Late Ordovician-Silurian unconformity (LeMone, 1987, 1988). On the basis of mapping by Lucia (1995), both Montoya and Silurian Fusselman Formations appear to have been disrupted by collapse into large, vertically extensive paleocaves. The dominant lithology of the southern Franklin Mountain outcrop is limestone, with early dolomite in the relatively thin supratidal formations (Chamizal and Cindy) and sparse, later stage dolomite that tends to be confined to karst-related breccia, with minor facies-specific dolomitization penetrating into the host rock (Lucia, 1968, 1995; Tillotson, 2003).

The strata of the Franklin Mountains were exhumed by early-phase Rio Grande uplift (30–15 m.y.) and are now exposed spectacularly in a north-south–trending fault block that dips to the west in an extremely arid environment with little vegetation. Due to excellent exposure and low humidity, the Franklin Mountains are an ideal test site for ground-, air-, and space-based studies for geological remote sensing using spectral and textural instruments.

The lidar data analysis of Transition Canyon clearly indicates the ability to distinguish intact bedding from breccia on the basis of textural analysis (Fig. 5), which coincides with spectral response from the hyperspectral image processing (Figs. 8 and 9) and previous work by Lucia (1968, 1995) and Tillotson (2003). The ten slices through the cliff in Transition Canyon (Fig. 5B and associated text file, TransitionCanyon.txt) show erosional profile changes in the vertical cliff face as the karst breccia is intersected. Numerous fractures within the lowermost Upper Ordovician Montoya Group (Upham Formation) that could be easily identified manually in the sun-shaded lidar model (colored sub-vertical lines in Fig. 5A) were digitized. Careful observation of the bedding parallel slices (red lines 4 and 5 from top of outcrop in Fig. 5A) will illustrate the relative down-drop of the Upham Formation into the underlying Ranger Peak Formation in the vicinity of maximum fracture density, which corresponds to brecciation in the Ranger Peak.

On the basis of field mapping, in addition to previous petrographic work (Lucia, 1995; Tillotson, 2003), the collapse breccia is dolomitized, and the surrounding host rock is less altered limestone. In the spectral analysis of the SWIR spectral range, the same pattern is observed. Note the green-to-purple/red transition in Figures 8 and 9, which corresponds to the spectral response of limestone and dolomite, respectively. It is important to note that although the dolomite and calcite in the southern Franklin Mountains can be very accurately distinguished visually (Figs. 10 and 11), no visible spectral information was used in this classification. Thus, chemical signatures of dolomite and calcite outside the visible spectral range may be diagnostic, even when visible spectra may not be. In addition, it is important to note that when dealing with naturally weathered surfaces, additional complications may exist. However, both freshly broken-face and weathered surfaces gave very similar spectral responses under controlled conditions in the laboratory; therefore, this effect is thought to be minimal.

The large footprint of the AVIRIS data (20 m) offers ample averaging of potentially noisy information (vegetation, localized mineral staining, etc.) into a smoother spectral response, and the airborne versus satellite platform with AVIRIS versus Hyperion (respectively) improved the signal-to-noise ratio significantly, which is potentially due to shorter atmospheric transit time. One problem with Transition Canyon as a study area is the relatively small number of pixels from which to extract spectral signature, even with the higher resolution AVIRIS data set. To test this, a larger breccia body, also dolomitized, was selected for sample analysis.

The Great McKelligon Sag on the east side of the mountain was chosen for spectral analysis because of its proximity to Transition Canyon, as well as its several orders of magnitude greater size. Relatively sharp transitions of limestone to dolomite are observed between “green” limestone host rock and red-blue dolomitized breccia (Figs. 8 and 9). In addition, other alteration zones within the Ranger Peak Formation are identified to the south of the Great McKelligon Sag, near the northern extent of the Quarry Breccia (Fig. 9, white arrow). This area has not been mapped as part of this study, but does show up on Lucia's interpretation (Lucia, 1995). The extremely difficult to access vertically extensive breccia body in the Great McKelligon Sag exhibits the dominant spectral signature of dolomite, as sampled from both Transition Canyon and the Great McKelligon Sag.

Principal component analysis of the SWIR spectral data alone has enabled the clear classification of the dolomite calcite transitions in the southern Franklin Mountains, both as related to brecciation and stratal dolomite (including the unbrecciated tidal flats of the Cindy Formation). The next step will be to try similar spectral analysis in a region with less obvious visible dolomite transitions and attempt a similar classification. The results from combined field and laboratory analysis can be used to assess quantitatively diagenetic patterns and relative proportions over much larger areas in a relatively short period of time.

Dolomite-calcite transitions have been imaged using hyperspectral image data in the SWIR, indicating this spectral zone as a potential focus area for future spectral research. The ability to confirm this relationship using natural reflected light, tested using simulated sunlight in the laboratory, ground-truth mapping, and petrographic analysis (Lucia, 1968, 1995; Tillotson, 2003), leaves little, if any, discussion as to the existence of dolomite within the collapse breccias. However, it is not clear from the spectral data used that different phases of dolomitization can be separated. Application of this technique to additional field areas that do not have clear visible dolomite coloration (although visible spectra were not used in the processing of the hyperspectral map) will further test this procedure for large-scale field mapping. Of particular interest would be areas where the dolomite visible spectra appear to be unaltered to the naked eye. Extension of this study into finer pixel dimensions would greatly improve the utility of this research, but will come at the price of larger and larger data sets to process.

Both the high monetary cost and lack of portability of full-spectrum hyperspectral imagers have contributed to high-resolution (submeter), hyperspectral acquisition not yet being a reality for most outcrop-based research. Larger pixel imagers, such as those used on AVIRIS, are commercially available; however, they are not compact enough to take into a remote field setting as are terrestrial scanning lidar instruments. In summary, the geoscience community is on the threshold of simultaneously collecting and utilizing high-resolution hyperspectral and lidar data in much the same way that we currently use lidar. In the same way that photography did not replace our need to conduct first-hand field observations, new imaging technology that blends lidar and hyperspectral data will only assist our focus in constructing quantitative geological models and help us see with new eyes what we may not have seen previously. This technology was able to assist in the delineation of calcite from dolomite in a remote-sensing application. These delineations correlate to breccias of collapse origin. The breccias created preferential fluid pathways later exploited by fluid migrating through the system.

We would like to thank the Reservoir Characterization Research Laboratory at the Jackson School of Geosciences, Bureau of Economic Geology, who sponsor this ongoing research. Current research sponsors include Anadarko, Apache Corporation, British Petroleum, Chevron, ConocoPhillips, ENI, ExxonMobil, Kinder Morgan, Marathon Oil Corporation, NorkHydro, Oxy, Petroleum Development of Oman, Pioneer, Repsol YPF, Saudi Aramco, Shell, and Statoil. Publication was authorized by the Director of the Bureau of Economic Geology, The University of Texas at Austin.