Feature ArticlesMonitoring Oil Slicks From Space To Aid in Offshore Petroleum Exploration
By Dr. Magnus Wettle
Remote Sensing Scientist
Petroleum and Marine Division
Offshore hydrocarbon reservoirs can experience oil and gas seepage through the seafloor and water column, resulting in telltale slicks on the sea surface. Detecting and identifying these intermittent and often remote slicks can contribute to directing exploration resources in both producing and frontier basins, as active oil seepage provides information about source rock type and maturity, location of source kitchens and migration pathways. These form part of a petroleum prospectivity analysis of a basin, which Geoscience Australia presents to industry to reduce its exploration risk. This type of precompetitive information also forms part of the offshore acreage releases.
Remote sensing, the acquisition of image data from sensors mounted on airborne or spaceborne platforms, is seen as a promising tool for detecting, mapping and identifying these seepage-derived slicks. Simply put, it offers a cost-effective means of scanning extensive and/or remote regions on an ongoing basis.
Challenges of Remote Sensing
The bulk of remote sensing of marine hydrocarbon slicks to date has focused on detecting and mapping relatively thick oil pollution and spills, often with a known extent, date and/or location. Utilizing remote sensing for petroleum exploration purposes poses significant additional challenges: Natural, seepage-derived oil slicks are typically considerably thinner on the sea surface, and their timing, extent, location and chemical properties are often unknown.
One type of remote sensing, synthetic aperture radar (SAR), has been shown to effectively detect oil slicks on the sea surface. Under suitable environmental conditions, a sea-surface slick has a dampening effect on capillary waves, thereby reducing the amplitude of the wave-dependant backscatter signal from the sea surface. This is visible as a dark area (negative contrast) in SAR imagery.
In principle, SAR technology is ideally suited to scan for remote and transient natural oil slicks, as it covers large geographic areas (one scene can have a footprint of several hundred kilometers across the sea surface) and can image through clouds and at night. However, the volume of data collected when scanning an area as large as the Australian Marine Jurisdiction (AMJ) prohibits manual, operator-based inspection of the imagery. This problem is further compounded where multitemporal datasets are involved. In addition, new data are continuously being acquired.
A second important limitation of SAR data is that it only contains information in one spectral channel. SAR imagery can therefore be thought of as monochromatic (black and white). Hence, other phenomena that cause dampening of capillary waves, such as biological films, hydrodynamic effects, bathymetric features and even weather-induced artifacts, lead to a range of false-positive interpretations of oil slicks in SAR imagery.
Optical remote sensors, on the other hand, typically collect data in three or more spectral channels. As an example, DigitalGlobe’s (Longmont, Colorado) spaceborne QuickBird sensor is referred to as multispectral, and it acquires information in four channels, the first three of which are broadly equivalent to the blue, green and red wavelengths of human vision (the fourth channel is a near-infrared channel). Hyperspectral sensors, such as the airborne HyMap sensor manufactured by Integrated Spectronics (Castle Hill, Australia), collect information in hundreds of spectral channels. These additional dimensions of wavelength-dependent information in optical remote sensing data theoretically allow for discrimination of false positives such as biological films, and may even allow for differentiation between oil types. Admittedly, this remains to be proven.
However, there is a trade-off with using optical remote sensing. Specifically, the spatial coverage of a typical scene is limited (e.g., one QuickBird scene covers approximately 15 by 15 kilometers of the sea surface), with costs ranging from thousands to tens of thousands of dollars per scene. In addition, optical remote sensing requires sunlight and cannot image through clouds. The latter constraint translates to potentially requiring several scene acquisitions over the same area in order to have one useful image.
In other words, it is too costly to use optical remote sensing data when attempting to monitor the AMJ for intermittent, seepage-derived slicks, particularly on an ongoing basis.
One solution is to try and harness the advantages of these complementary remote sensing data types. To this end, Geoscience Australia is developing a two-pronged remote sensing-based approach to study seepage-derived slicks in the AMJ: first, building a semiautomated processing and classification system in order to scan large numbers of SAR scenes for potential natural slick targets, and second, investigating the potential of optical remote sensing as a diagnostic tool for further, targeted study in identified areas of interest.
The Semi-Automated Marine Slicks SAR Analysis (SAMSSARA) system, which was developed using Definiens (Munich, Germany) software, combines object-oriented data analysis and classification tools with batch-processing and thorough reporting schematics in order to provide a first screening of large SAR datasets.
An identified object with contiguous, lower values (i.e., negative contrast as defined by local statistics) is considered a potential slick candidate object. The shape of each potential object is then examined using a set of criteria that include length versus width and perimeter versus area. A natural slick typically has an elongated, thin form, with one or more twists (due to surface currents).
Identified objects are merged using a set criteria that include the object’s orientation in relation to neighboring objects and the extent and orientation of shared perimeters. Merged objects are then reclassified based on the contrast and shape rules previously described, and the process is repeated iteratively until every subset in the scene has been classified.
The classification scheme currently uses three categories of potential slick targets: dark feature, moderately slick-shaped and slick-shaped, with the latter the most likely slick target. SAMSSARA retains the number of classified targets in each category for each scene and outputs these project statistics, together with thumbnail imagery and geocoordinates, in both spreadsheet and HTML format.
Finally, SAMSSARA outputs georeferenced maps (shapefiles) of all classified potential targets. This allows for spatial and contextual analysis of an identified potential slick feature, together with ancillary information, such as geological data.
In one example, SAMSSARA’s output from the screening of 238 SAR scenes of the southwestern AMJ covered a sea surface area of approximately 1.35 million square kilometers. While the manual and subjective analysis of one SAR scene can take a trained operator approximately 30 minutes, the 238 scenes were objectively analyzed by SAMSSARA in approximately six hours, producing an array of easily interpretable outputs.
Slick-shaped features of interest in the SAMSSARA output can be investigated in the context of ancillary data and revisited with additional SAR data from a different date. The benefit of using such multitemporal analysis is that the reoccurrence of a very slick-shaped feature in both space and time rules out several false positive interpretations and supports directing additional resources, such as optical remote sensing data, to investigating this particular sea surface area.
Optical Remote Sensing
Optical remote sensing has been shown to be able to identify certain sea surface slicks through an ability to collect imagery data over several wavelength channels. One example is algal blooms, where a combination of spectral channels can detect the biological signature. Errors of commission in SAR data can therefore be reduced by eliminating these types of false positives using the additional information in optical data.
The applicability of optical remote sensing data to positive identification of seepage-derived slicks is less straightforward. The absorption of light by oil is dependent on both wavelength and oil thickness; there is an increase in light absorption with increasing oil thickness, as well as an increase in absorption toward the blue end of the visible spectrum.
Petroleum is composed of complex mixtures of hydrocarbons, and the chemistry of individual oils can vary substantially. This variation in chemistry affects a variety of bulk properties, such as density and viscosity, and has a direct effect on the optical properties of oils. Thus, the chemistry of an oil can influence both the wavelength and thickness-dependent absorptive properties. Importantly, it is these absorptive properties that offer the oil slick diagnostic potential in optical data.
This diagnostic potential was investigated through a series of small-scale laboratory experiments. Samples of oils known to occur in the AMJ were sourced from the Geoscience Australia archive, and the conditions under which a remote sensor would collect data in the environment were simulated. Known amounts of each oil type were added to a layer of water, and a hyperspectral radiometer was used to measure the resulting change in reflectance as a function of oil thickness. These changes in reflectance were then normalized to the response functions of remote sensing platforms likely to be used for oil slick detection.
The results confirmed that remote sensing-based oil slick detection is both thickness and oil type-dependent. What is more noteworthy is that the work demonstrated an objective method for quantifying the sensitivity of a given sensor to a given oil type at a given oil thickness. This has numerous applications. For example, if a basin is known to contain a certain oil type, the sensitivity analysis will reveal if optical remote sensing is applicable. Indeed, one of the oil types studied was shown to be undetectable up to a thickness exceeding those of typical natural slicks, while another oil type was shown to be detectable by several commercially available sensors at thicknesses of only 10 micrometers. This type of information is relevant to a prospective purchaser of optical remote sensing data, particularly given the conflicting information from the literature and data suppliers with regards to this application of optical imagery.
Building on the findings of these initial laboratory simulations, a more robust and easy-to-implement method has been developed to obtain the optical properties of a given oil type, making this type of sensitivity analysis more accessible to a range of users, provided that they have access to samples of the oil type in question.
This sensitivity analysis method gives an indication of the applicability of a given sensor to a given oil type, if the approximate layer thickness is known. In addition, this type of information can be used to start identifying oil types and/or estimating thicknesses. If an area is known to contain a certain oil type, and optical remote sensing data is able to detect a slick, assumptions can be made as to the minimum thickness of the layer. As an alternative example, the detection of a slick can be used to rule out the presence of oil types that have been assessed to be undetectable.
Future development of SAMSSARA will focus on furthering the multitemporal analysis component, since a reoccurrence of a sea surface slick is a strong indicator of a seepage origin. The anticipation is that results will continue to be added to a database of oil type versus sensor sensitivity. This will provide a reference for the applicability of a given optical remote sensing platform for detecting a given type of marine oil slick, effectively directing future exploration resource deployment.
Dr. Magnus Wettle leads the aquatic remote sensing team within the National Earth Observation group of Geoscience Australia. The team specializes in detecting offshore hydrocarbon slicks as well as producing high-spatial-resolution bathymetry models using remote sensing.