Overview of three study areas (yellow boxes) near Santa Barbara, Calif.
Accurately determining the location of hyperspectral measurements made in the field is critical for supporting airborne mapping programs. The spectral signatures collected at field sites must be spatially correlated with individual picture elements (pixels) of airborne hyperspectral imagery. Field sites are critical for understanding how to map different materials at the earth’s surface, such as oil-impacted soils and onshore oil seeps, using airborne hyperspectral technology. The need to detect oil-impacted materials such as soils, water bodies and man-made surfaces within refineries, industrial sites, brownfields and oil fields is increasing as environmental regulations are tightened and alternative land uses are advocated. In addition, the quest for increasing petroleum reserves has explorationists looking for new tools to detect onshore oil seeps.

Last year, HJW GeoSpatial (HJW) Inc., Walnut Creek, Calif., and The Geosat Committee (Geosat) Inc. initiated a cooperative Research and Development (R&D) study designed to locate with GPS and measure with ground and airborne hyperspectral technology different materials associated with onshore oil seeps and oil-impacted soils in southern California. The primary objectives of the study were to develop an understanding of the spectral characteristics of oil-impacted soils and seeps, and to build a spectral library that would make the detection process more rapid and reliable. This oil-focused hyperspectral library may be the first of its kind in the commercial sector.

Hyperspectral image draped over DEM—an oil seep found

at the red area.

Hyperspectral Remote Sensing

Airborne hyperspectral imagery is used to map different materials at the surface of the earth based on their spectral characteristics. Hyperspectral sensors measure the intensity of solar energy reflected from materials at the earth’s surface. These sensors record different wavelengths of reflected light. They can record visible light (comprised of relatively short wavelengths) as well as longer, near-infrared and shortwave-infrared light. Reflected light is collected into picture elements (pixels) by flying a sensor over terrain or into an instantaneous field of view (IFOV) by pointing a hand-held instrument at the ground. The reflected visible and infrared light is subdivided into 100 to 200+ discrete samples or wavelength bands within each pixel or IFOV. This large number of spectral bands is the basis of the name “hyperspectral.”

The amount of energy recorded by the hyperspectral sensor within each pixel varies across the wavelength spectrum because different materials on the earth’s surface scatter or absorb solar energy in varying amounts based upon the material’s physical properties and composition. Hyperspectral sensors are unique in that they have enough spectral resolution to identify different surface materials based solely on the spectral signature.

In order to correlate spectral signatures with specific materials, scientists obtain “pure” samples of the material and collect highly accurate, reflected light measurements in the lab or in the field (using a portable spectrometer that is carried in a backpack). The measurements allow spectral libraries to be built that contain a group of hyperspectral signatures that have been positively identified with specific materials at the earth’s surface. Spectral libraries have been constructed for numerous minerals, plants and man-made materials.

Hyperspectral signatures spanning blue light to shortwave-infrared wavelengths for different materials.

Problems with Accurately Locating Oil-Impacted Surfaces

In 1998, HJW joined an airborne hyperspectral group shoot initiated by Geosat. The sophisticated sensor used in the study spanned the visible, near-infrared through shortwave-infrared wavelengths and was built by Integrated Spectronics of Australia. Earth Search Sciences, Inc. (ESSI), Kalispell, Mont., acquired the imagery for the group shoot over a mountainous site east of Santa Barbara.

The sensor (designated Probe-1 by ESSI) subdivides light into 128 wavelength intervals, creating a hyperspectral “datacube.” These datacubes were acquired as flight strips with a 5-meter ground sampling distance (gsd) and were collected by line-scanner technology. The instrument collects data in a cross-track direction by mechanical scanning and an along-track direction by forward motion of the aircraft. Each flight strip covered an area approximately 2.5 km across and 25 km in length.

HJW analysts correlated spectra in the library with spectra derived from airborne imagery pixels to help identify materials in each pixel. In this study, the pixels covered an area of 25 square meters, much larger than the area measured at the field site (typically the IFOV at field sites was ~1 square meter). In addition, airborne pixels typically contain a mixture of materials. Our analysts “unmix” materials occurring within a pixel using ENVI image processing software on NT workstations (Research Systems Inc., Boulder, Colo.) in order to identify individual materials of interest such as oil-impacted soils.

Field sites with exacting portable spectrometer measurements are used as training sites to help guide the analysis of the airborne imagery and to confirm that the airborne sensor “sees” the same spectral signature as recorded on the ground. However, in order to use these training sites, the analyst has to be confident that he/she can accurately locate the airborne pixel that contains the field site. Visual and spectral correlation is done between the airborne and ground datasets, however, this process does not provide for accurate coordinates.

The coordinates of individual pixels on the rectified flight strips were not reliable for determining the locations of field sampling sites. Published USGS topographic maps, geologic maps from the Dibblee Geological Foundation and orthophotographs were also not adequate for deriving accurate coordinates as these have scales of 1:24,000 (± 12 meters accuracy). In order to construct a spectral library and GIS of field measurements that had reliable coordinates, we deployed GPS technology to the field. The spectral library, field sampling program, and ground photographs were designed to have locations based on GPS coordinates so that they could be easily integrated into a GIS and would provide a consistent spatial framework for the study.

Coastal field site showing oil seep’s GPS location (red triangle) and ground photograph—taken from GIS.

The GPS Program

During early 2000, our remote sensing team analyzed the airborne hyperspectral flight strips and published geologic maps for areas that could have oil-impacted soils and/or onshore oil seeps. We obtained preliminary coordinates for these features from the USGS maps and loaded these coordinates into a GIS.

In July, we sent an experienced crew into the field to verify if these areas were impacted by oil and to determine optimum access routes. The crew evaluated field conditions, vegetation (including canopy closure), and surrounding topographic relief. The primary sites were located in areas of limited vehicle access. Some sites required trekking two to three miles over steep terrain. While in the field, photo-identifiable points associated with proposed measurement sites were located on enlarged hard copy plots of the imagery and USGS maps. The crew developed a prioritized list of field sites and recommended the most time-effective sequence for revisiting these sites in August with a portable Analytical Spectral Devices spectrometer.

The discontinuance of Selective Availability several months earlier did not help us achieve our accuracy goals as it would only provide a location accuracy of between ± 10 to 20 m. We needed cost-effective, GPS technology that would ensure a 1 to 5 m accuracy for the field sites. Such a level of accuracy would enable other researchers to revisit the oil seep outcrop or impacted soil area in the future. It would also provide a consistent level of accuracy for spectra, samples and ground photographs in our GIS. In addition, we hoped to use these coordinates to improve the geometry of the hyperspectral imagery.

To achieve the 1 to 5 m accuracy, we used one high-end, hand-held GPS receiver, a moderately priced GPS receiver and post-processing of code-phased GPS data. Trimble’s (Sunnyvale, Calif.) online mission planning program, SatView, was used to determine satellite availability and expected signal health for the days the crew was in the field.

The Trimble GeoExplorer II was used as the “rover” and mounted onto a tripod to ensure stability. The field crew recorded the required 10 minutes of signal at each site where field spectrometer measurements, samples and photographs were obtained. During this 10-minute, high-accuracy acquisition, an auxiliary GPS unit, the LEI Eagle Explorer (Eagle Electronics, Catoosa, Okla.), was used to monitor the Position Dilution of Precision (PDOP) and location of each satellite vehicle within the constellation. This Eagle Explorer served as a backup to ensure we were collecting correct data while in the field.

Field measurements of oil seep material using portable spectro-meter with GPS unit.
The portability of the GPS receivers was essential to the operation, as the crew had to carry the spectrometer that weighed more than 60 lbs. while collecting many samples of oil-impacted soils, hardened, oil-seep deposits and rocks. The hyperspectral measurements needed to be collected close to solar noon to optimize illumination. Temperatures exceeded 100 degrees F during the data-gathering phases of the three day operation. Hardened tar deposited around the oil seep outcrops partially melted and flowed down the slope while the GPS data were being collected. All the equipment functioned normally in spite of the heat.

The DGPS files were post-processed using Trimble’s Pathfinder Office software and the appropriate base station correction files. The correction files for this project were downloaded via the Internet from the U.S. Forest Service reference station in Porterville, Calif., which is approximately 100 miles north of the study area and well within the Trimble-prescribed, 300-mile, base station-to-rover range for code phase corrections.

The GeoExplorer II could store up to 192kb in memory, enough for about 20 high accuracy files. As a precaution, however, the rover files were downloaded onto a PC each day and backed-up on diskette.

Because the USGS 7.5' geologic maps from the Dibblee Geological Foundation were fundamental to this study, the GPS coordinates were converted from WGS-84 to NAD-27 using Pathfinder Office software. These coordinates were attached to all field spectra, ground photographs and field samples via Excel spreadsheets.

The Excel spreadsheets were loaded into ArcView GIS. The DGPS coordinates are being used to upgrade the location of those airborne pixels characterized by oil seeps and oil-impacted soils. Improving the locational accuracy of those pixels that have these unique signatures improves our ability to integrate all the field observations with the airborne hyperspectral datacube.

Collecting GPS data along the coast at oil seep deposits.

The Spectral Library

This spectral library includes measurements of field samples with varying amounts of oil-impacted surface. This library can be used with other hand-held, airborne and satellite sensors that span the visible through short-wave-infrared spectrum. The library enables skilled analysts to detect oil-impacted surfaces within dense urban sites (industrial sites, refineries, brownfields, tank farms and pipeline corridors) and across remote geologic structures. This spectral library expands the use of hyperspectral technology to monitoring changes at active and abandoned industrial sites, establishing environmental baselines and supporting oil exploration.

GPS technology enables the critical field measurements used to create the spectral library to have accurate ground locations. The portability of the GPS units, the reliability of the hardware and software in the field and office, and the consistent results makes this technology cost-effective and mission critical for constructing field-based hyperspectral libraries. The GPS locations enable the library to be confidently integrated with other spatial datasets within a GIS and also enables revisits to the relatively remote measurement sites with new sensors in the future.

We plan to use these accurate GPS points to reprocess the flight strips and improve the geometry of the hyperspectral datacubes. Our cooperative R&D study clearly demonstrates that GPS technology is fundamental to hyperspectral remote sensing.

Acknowledgements

Hattie Davis and Patrick Caldwell of HJW GeoSpatial, and Dr. Joseph Zamudio of ESSI, provided key image processing, field operations and field spectrometer measurements, respectively. The sponsors of this cooperative R&D effort were Royal Dutch Shell, Chevron and ExxonMobil. Coordination and funding was through The Geosat Committee Inc. For more information, contact James Ellis by E-mail at jellis@hjw.com or by phone at 510/638-6122.