Professional Documents
Culture Documents
Ben Lord1
Abstract
Remote-sensing data are highly useful for the oil and gas
industry. Remote sensing has proven to be an integral tool for
Downloaded 12/27/16 to 132.239.1.231. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
1
NEOS GeoSolutions. http://dx.doi.org/10.1190/tle36010024.1.
whole-pixel and subpixel analysis, analyzing angular differences specific absorptions that need to be highlighted. Using classifiers
in spectra and overall wavelength land-cover contributions to such as SAM and LSU, the entire spectrum is analyzed and may
in-scene pixels. These classification algorithms included spectral not capture some of the key wavelengths that need to be considered.
angle mapper (SAM), linear spectral unmixing (LSU), continuum Band ratios allow for this specific wavelength analysis. For this
removal, and band ratios. These were all combined into a decision approach, the band-math preset function in ENVI was used, and
tree approach to create final classifications. mathematical indices were created and applied in order to best
highlight these specific wavelengths in the spectrum. The math-
Spectral angle mapper (SAM). Spectral signature matching ematical equations created for each target endmember calculates
is very important in the identification of earth surface objects. the depth of the specific absorption as well as gradient changes
Endmember spectra collected in the field (NEOS spectral library) of the spectra. The output of these ratios are single-band images
as well as spectra from the USGS spectral library were used to in gray scale (values between 0 and 1), also known as a rule image,
identify oil seep and other target pixels such as rock types and in which the brightest sections (values closer to 1) signify the
mineral alterations in the images. A SAM algorithm, which is a deepest section of endmember absorption dip.
spectral-based classifier, was used to compare pixel values to Spectral endmembers used as a basis for creating these indices
user-derived endmembers or signatures using the various spectral and ratios were taken from known spectra from field operations
libraries, both USGS and proprietary (Kruse et al., 1993). SAM or spectral libraries, similar to the reference endmembers in the
compares the angle between reference spectra vectors and spectra previous classification algorithms. These endmembers were used
from each pixel in an image. It compares each pixel in the image as references and are compared to spectral profiles from the study
with every endmember for each class in a spectral library and area for generating the surface hydrocarbon classifications.
assigns a weighting value between 0 and 1. Endmembers can be
taken directly from the image or from signatures measured directly Combined decision tree approach. All gray-scale rule images
in the field or laboratory. The smaller the angle, the closer the produced by the SAM, LSU, and band ratio images put into a
relationship between a given reference spectrum and a pixel. The decision tree matrix where thresholds were applied. These thresh-
main disadvantage of the SAM algorithm is that it does not olds are derived from and compared with SAM, LSU, and ratio
consider the subpixel attributes. The two main outputs of SAM values from known targets from a priori knowledge (ground truth
analysis are a SAM classification (which is generated based on operations, expert geologic knowledge, known geologic map, etc.)
one particular threshold value of angle calculation) and a gray-scale and applied in the decision tree. This decision tree process produces
rule image where the lower values signify angles closest to the final hydrocarbon seepage classifications, mineral alteration clas-
input spectra. sifications, and lithology classifications. Figure 2 shows an example
decision tree path classifying for alteration minerals.
Linear spectral unmixing (LSU). LSU outputs the same prod-
ucts as the SAM classification (a classified image and a gray-scale Indirect mineralogy (satellite/airborne)
rule image) but specifically takes into account the spectral mixing In addition to the direct detection of hydrocarbon spectra, both
that can take place within one pixel since inevitably, despite the mineral analysis and surface geochemistry play a large role in the
highest resolutions capable by sensors, there will be spectral vari- overall understanding of the hydrocarbon system. With respect to
ability within the pixel itself (Plaza and Plaza, 2011). The method indirect mineralogy, a number of relationships can be established
assumes that within one scene pixel there are many different factors between surface hydrocarbon seepage and mineral occurrence. The
that can affect the average endmember spectra that come out of classification of minerals in the various study areas becomes part
that particular pixel. That is to say that in one pixel there will be of the indirect hydrocarbon detection methodology. The indirect
a series of different landforms that can affect the overall spectra detection method for remote sensing data relies on the detection
of that location, including vegetation, different mineral alterations, of sediments and rocks altered at the surface by the microseepage
and different soil and ground properties. LSU investigates these of hydrocarbons. The hydrocarbons that escape from the under-
aspects and analyzes the images incorporating these factors based ground reservoir cause an oxidation-reduction reaction either in
on the endmember spectral library used in the analysis. situ or along vertical migration pathways (Saunders et al., 1999).
The resulting mineral alterations can be detected at the surface in
Continuum removal. Throughout the images, background sediments and soils. The mineralogical alteration induced by hy-
spectra from vegetation and general quaternary cover can mix drocarbon seepages include bleaching of red beds, enrichment of
with some of the more pure spectra expected to be classified. One ferrous iron, and the alteration of clay minerals and carbonates
approach to minimize this effect is via linear spectral unmixing leading to mappable alteration zones above and near subsurface
suggested previously; another method to highlight some of the accumulations. Figure 3 shows an example diagram of key surface
more subtle variability in in-scene spectra is via continuum removal observations for this microseepage model. It is through these mineral
Although many areas already have been well researched and postprocessing/classification to validate the results. Classifications
highly explored for oil and gas, remote sensing can still play a were done for hydrocarbon seepage as well as methane gas seepage
helpful role in these areas. The study area of Tioga County in from the combined SWIR spectra from the airborne platform
Pennsylvania has produced hydrocarbons for more than 100 years, and the thermal data collected from the helicopter-based FTIR
and many of the well sites are in full operation. However, a number platform. Field crews were also deployed to take analytical spectral
of wells and pipelines have been decommissioned, plugged, or device (ASD) readings of oil seepage from documented abandoned
abandoned. Some of these sites have since started leaking, adding wells. Figure 4 shows ASD readings taken from land-cover types
to the natural seepage already occurring in the area. Companies surrounding seepage areas. These readings were later used as
wishing to sustain ongoing production require the ability to drill reference spectra input into ENVI classification algorithms such
wells while taking into account the environment and overall safety as SAM and LSU.
of the local population. In Pennsylvania,
this, in effect, means avoiding pipelines
through all potential pathways of meth-
ane, whether they be man-made or
natural (McKee and Beasley, 2013), in
order to protect fresh ground-water
aquifers. The ability to map these hy-
drocarbon occurrences using remote
sensing technology has proven to be
highly useful for environmental analysis
and for future infrastructure planning
as described by Smejkalova and Bujok
(2012) and Schneising et al. (2014).
For this study, a SPECTIR airborne
hyperspectral survey was conducted in
Figure 2. Example decision tree constructed for alteration mineral classifications associated with hydrocarbon
late summer 2011 over the entire county. microseepage. As annotated above, each decision node has thresholds within it to characterize the pixel. The output
During the same period, a smaller test of this decision tree node can lead into another node (leaf node) or an edge, which can be a classified pixel or,
area was covered using a Fourier eventually as you go down the nodes, an unclassified pixel.
Figure 4. ASD sample readings taken from an abandoned oil well location
Figure 3. Diagram illustrating microseepage in oxidizing zones and accompanied with surrounding land covers, including soil and vegetation altered by oil
minerology alterations (modified from Schumacher, 1996). contamination.
generated at each X and Y location of the FTIR data for further readings (up to 19,660 ppm). Other areas where origins were
analysis and mapping purposes. The thermal emissivity data were determined to be natural had oil sheens in stagnant water sur-
gridded and stacked using proprietary NEOS algorithms, creating rounded by vegetation and usually were located on the banks of
thermal image files for each flight. wetlands or springs.
Methane gas shows characteristic absorption at the 7800 nm Results from spectral mapping coupled with field observations
region of the emissivity spectrum (Figure 5). Thermal data wave- discovered seepage occurring from broken and abandoned well
length ranging from 7500 to 12,000 nm were used to detect sites as well as natural seepage indicators in numerous positions
methane plumes. A band ratio was calculated at 7800 nm from that at the time of the survey had not been documented. Figure 7
the thermal image files. Band threshold parameters as well as shows some of the key classified discoveries in the survey area
absorption characteristics of CH4 spectra and surrounding areas differentiated between liquid hydrocarbon and gas hydrocarbon
(vegetation, soil, etc.) were used in the binary decision tree algo- seep endmembers.
rithm described in Figure 2 to yield possible CH4 plume polygons In summary, the project allowed for better understanding of
for the study area. where surface seepage was occurring. It also provided real-time
Ground-truthing verification work was performed to validate analysis of potential leakage in already developed areas, making
oil classification points. Field validation sites were selected based it an ideal tool for future pipeline monitoring because of its rapid
on a subset of classified locations with regard to accessibility. Eleven and noninvasive nature (Schneising et al., 2014). The area was
classified locations were targeted to validate the hyperspectral also an ideal location for sampling surface leakage and measuring
classification. Field visits were conducted with a ground crew using spectral characteristics of the leakage.
Figure 5. Example methane spectra with key target absorption at 7800 nm.
Figure 7. Tioga County study area showing classified liquid hydrocarbon seepage
points in green and gas hydrocarbon seepage points in purple. Ground survey
crews were deployed in a number of these positions to verify classification targets.
Figure 6. Seep location field photo taken over brown algal soil surrounding an Surveys showed an 89% success rate with the classification between airborne and
abandoned well. ground-based methods.
from all pixels in the ROI. Sampling locations were divided into a decision tree analysis to produce highest-scoring rule images
various land-cover types based on known geology and TOC and generate a final classification of only pixels closely matching
sample values. TOC samples were grouped into values of high, the endmember input spectra.
medium, and low TOC based on TOC lab readings. These were
grouped as follows:
rock. The majority of seepage points occurred on Jurassic and that these are measurements of the surface that do not in any way
Late Cretaceous rock. The highest densities of this classification
coincided well with the historical seepage points (Figure 13).
For this project, hyperspectral data were combined with 200 m
resolution radiometric data to obtain enhanced surface lithologic
mapping. An averaged 22-band mosaic was stacked with a 3-band
radiometric data set. An unsupervised isodata classification was
applied and compared with known surface geology maps for in-
terpretation and to attempt further enhancement of detail to the
surface geology already mapped. For this process, each class output
from the unsupervised classification was compared with the known
geology map, and links were made as to the overall surface variation
and the area’s geologic setting (Figure 14).
The geologic setting will impact the overall spectral signature
collected from the sensor, whether it is from variations in vegeta-
tion, variations in soil composition, or variations in surface lithol-
ogy characteristics. From these correlations, a surface variation
map was made, leading to an enhanced surface lithology map
(Figure 15). This product yielded improved insights on where the
mature source rock might be and higher precision on the exact
contact boundaries between geologic units.
Figure 13. Bitumen classification map using SPECTIR hyperspectral data. Two Figure 15. Final surface variation map where variations in color signify
areas pointed out are known documented hydrocarbon seepage locations, which differences in surface lithology obtained by combining radiometric and
coincide well with the classification results. hyperspectral data sets.
are younger than that of the latest tectonic event. For example, classifications existed, thereby increasing the likelihood of its
this holds true for surface seepage, as varying migration pathways existence. Then, similarly to the Argentina case study, compari-
can mean that reservoirs or producing source rocks are not always sons were made with known geologic maps of the region, and,
directly below the seepage point. Similarly with outcropping rocks, using unsupervised classifications of hyperspectral data, greater
due to high complexity of variance in directional stresses, the detail was given to the surface lithology, which in turn resulted
outcropping rock may be dipping away (Abrams, 2005) in another in improvements in the geologic maps. The Lebanon project is
direction, or may be completely folded over as we have seen in an ideal study showing the applicability of remote sensing in
Argentina. In the analysis of the hydrocarbon system in these frontier areas where little ground data exists.
situations, additional remote sensing data types such as magnetic Advancing technology with newer sensors and newer satellite
or gravity data assist in creating a better understanding of sub- payloads, with higher spatial and spectral resolutions, will only
surface anomalies that would cause some of these migration increase the overall applicability of remote sensing (Lisica, 2015).
pathways to form. Adding seismic data to the integration with Lisica also makes reference to new superspectral sensors such as
remote sensing data can enhance areas of potential faulting and that of WorldView3 (launched in August 2014) that combine
identify where the structures extend beyond the surface. An higher spatial/spectral resolutions in larger footprints, making
understanding and integration of both subsurface and surface data this new technology ideal for basin-wide and regional analyses.
is of great assistance to any analysis (Schumacher, 2016). As previ- Remote sensing, as shown here and in countless other studies,
ously mentioned, however, this is expensive; what remote sensing has proven to be a unique, cost-effective way of understanding
can do is establish high-priority areas for further study. broad frontier areas and high-grading areas for further study.
This being said, remote sensing is a tremendously useful form
of data analysis for the earth’s surface in the oil and gas realm. In Corresponding author: blord@neosgeo.com
Pennsylvania, USA, it was shown that seepage classifications
provided insights into leaking abandoned wells in addition to References
naturally occurring seepage. Individual surveying of these numer- Abrams, M. A., 2005, Significance of hydrocarbon seepage relative
ous wells by foot can be challenging. Therefore, being able to map to petroleum generation and entrapment: Marine and Petroleum
and locate seepage that could prove harmful to the populous and Geology, 22, no. 4, 457–477, http://dx.doi.org/10.1016/j.marpet-
the environment (regardless of the source being natural or man- geo.2004.08.003.
Agar, B., and D. Coulter, 2007, Remote sensing for mineral explo-
made) in a quick and cost-effective way can be extremely useful
ration – A decade perspective 1997-2007, in B. Milkereit , ed.,
for oil and gas planning and services.
Proceedings of Exploration 07: Fifth Decennial International
In Argentina, hyperspectral airborne data was used to map Conference on Mineral Exploration, 109–136.
hydrocarbon seepage and mineral alteration associated with Agrawal, G., and J. Sarup, 2011, Comparison of QUAC and
microseepage. The lithologic variance allowed for key formations FLAASH atmospheric correction modules on EO-1 Hyperion
such as the Vaca Muerta to be mapped and classified. The com- data of Sanchi: International Journal of Advanced Engineering
bination of field data and expert geologic knowledge resulted in Sciences and Technologies, 4, no. 1, 178–186.
the advancement in detail of lithologic mapping in the area. Also, Almeida-Filho, R., F. P. Miranda, and T. Yamakawa, 1999, Remote
with the combined assistance of TOC lab data (analyzed from detection of a tonal anomaly in an area of hydrocarbon micro-
field samples) surface TOC was mapped, establishing a better seepage, Tuscano basin, north-eastern Brazil: International Jour-
assessment in the quality of potential source rocks and assisting nal of Remote Sensing, 20, no. 13, 2683–2688, http://dx.doi.
org/10.1080/014311699212029.
in the evaluation of shales and their locations.
Clark, R. N., and T. L. Roush, 1984, Reflectance spectroscopy:
Finally, in the Lebanon study, it was shown that a great deal
Quantitative analysis techniques for remote sensing applications:
can be gained from remote sensing — even with a lack of field Journal of Geophysical Research, 89, no. B7, 6329–6340, http://
data. This was shown through classification of hydrocarbon dx.doi.org/10.1029/JB089iB07p06329.
seepage using bitumen samples and previously attained hydro- Curtis, B., 2013, Quantifying bitumen in natural oil sands using
carbon spectra. This was validated with the correlation of the NIR technolog y: ASD Inc., http://w w w.asdi.com/
classification with known historical hydrocarbon show points. getmedia/63538819-9aca-40fc-aa57-82a2e58c5a1c/Quantify-
However, issues do exist when picking reference spectra. In the ing_bitumen_in_natural_oil_sands_using_NIR_technology.pdf,
most ideal cases, reference endmembers should be collected from accessed 2 December 2016.
the scene itself for direct correlation between field and airborne Du, Y., C. Chang, H. Ren, C. Chang, J. O. Jensen, F. M. D’Amico,
spectra. For the Lebanon study, this was not available and hence 2004, New hyperspectral discrimination measure for spectral
characterization: Optical Engineering, 43, no. 8, 1777–1786,
reference spectra were taken from regions outside of the study
http://dx.doi.org/10.1117/1.1766301.
area; specifically, two spectra from North America were used.
This can lead to additional uncertainty in the classification as
Lillesand, T. M., R. W. Keifer, and J. W. Chipman, 2009, Remote sensing and image in-
terpretation: John Wiley & Sons, Inc.
Lisica, R., 2015, A new age for oil and gas exploration remote sensing data and analytics
are changing the industry: Earth Imaging Journal, http://eijournal.com/print/articles/a-
new-age-for-oil-and-gas-exploration-remote-sensing-data-and-analytics-are-chang-
ing-the-industry, accessed 12 November 2016.
Kruse, F. A., A. B. Lefkoff, and J. B. Dietz, 1993, Expert system-based mineral mapping
in northern Death Valley, California/Nevada using the airborne visible/infrared imag-
ing spectrometer (AVIRIS): Remote Sensing of Environment, 44, nos. 2–3, 309–336,
http://dx.doi.org/10.1016/0034-4257(93)90024-R.
McKee, B. and C. Beasley, Use of remote sensing technologies to detect surface and near-
surface stray gas occurrence and potential migration pathways in Tioga County, Penn-
sylvania, 2013, Search and Discovery Article #80317.
Naeth, J., R. di Primio, B. Horsfield, R. G. Schaefer, P. M. Shannon, W. R. Bailey, and
J. P. Henriet, 2005, Hydrocarbon seepage and carbonate mound formation: a basin
modelling study from the Porcupine Basin (offshore Ireland): Journal of Petroleum
Geology, 28, no. 2, 147–166, http://dx.doi.org/10.1111/j.1747-5457.2005.tb00077.x.
Plaza, A., and J. Plaza, 2011, Parallel implementation of linear and nonlinear spectral un-
mixing of remotely sensed hyperspectral images: proceedings of High-Performance
Computing in Remote Sensing, http://dx.doi.org/10.1117/12.897326.
Prelat, A., Gunaratne, S., Huebner, L., Freeman, C., Cook, A., and C. Soriano. 2013.
Airborne hyperspectral detection of natural offshore and onshore hydrocarbon seeps,
in F. Aminzadeh, T. B. Berge, and D. L. Connolly, eds., Hydrocarbon Seepage:
From Source to Surface: SEG/AAPG Geophysical Developments Series 16, 171–
182, http://dx.doi.org/10.1190/1.9781560803119.ch11.
Prost, G. L., 2002, Remote sensing for geologists: A guide to image interpretation: Tay-
lor & Francis.
Saunders, D. F., K. R. Burson, and C. K. Thompson, 1999, Model for hydrocarbon mi-
croseepage and related near surface alterations: AAPG Bulletin, 83, no. 1, 170–185.
Schneising, O., J. P. Burrows, R. R. Dickerson, M. Buchwitz, M. Reuter, and H. Bovens-
mann, 2014, Remote sensing of fugitive methane emissions from oil and gas produc-
tion in North American tight geologic formations: Earth’s Future, 2, no. 10, 548–558,
http://dx.doi.org/10.1002/2014EF000265.
Schumacher, D., 1996, Hydrocarbon-induced alteration of soils and sediments, in D.
Schumacher and M. A. Abrams, eds., Hydrocarbon Migration and Its Near-Surface
Expression: AAPG Memoir 66, 71–89.
Schumacher, D., 2016, Improving success of surface geochemical surveys: 7 pitfalls to
avoid: Search and Discovery Article #80504.
Smejkalova, E., and P. Bujok, 2012, Remote sensing methods in the identification of oil con-
taminations: GeoScience Engineering, 58, no. 1, 24–33, http://dx.doi.org/10.2478/v10205-
011-0010-6.
Taranik, J. V., and A. P. Crosta, 1996, Remote sensing for geology and mineral resources,
an assessment of tools for geoscientists in the future: International Archives of Pho-
togrammetry and Remote Sensing, 31, B7, 689–699.
van der Meer, F., P. Van Dijk, H. Van Der Werff, and H. Yang, 2002, Remote sensing
and petroleum seepage: a review and case study: Terra Nova, 14, no. 1, 1–17, http://
dx.doi.org/10.1046/j.1365-3121.2002.00390.x.
Veiga, G. D., L. A. Spalletti, J. A. Howell, and E. Schwarz, 2005, The Neuquen Basin,
Argentina: A case study in sequence stratigraphy and basin dynamics: Geological So-
ciety of London, Special Publications, 252, 1–14.
Young, S. J., R. B. Johnson, and J. A. Hackwell, 2002, An in-scene method for atmo-
spheric compensation of thermal hyperspectral data: Journal of Geophysical Research,
107, D24, ACH 14-1–ACH 14-20, http://dx.doi.org/10.1029/2001JD001266.