Background
Chemical producers, petrochemical facilities, and similar industries that must maintain compliance with environmental regulations need a cost-effective means of assessing the extent of air, ground, or water contamination. Effective remote detection, search, and tracking of hazardous substance spills for site remediation requires real-time, site-specific, and high-spatial-resolution data. Opto-Knowledge Systems, Inc. (OKSI) of Manhattan Beach, Calif. has recently explored the use of aircraft-acquired digital multispectral and hyperspectral imagery to satisfy this need.
      Since its establishment in 1991, OKSI has built custom imaging spectrometers in the visible/near infrared (VNIR) and mid-wave infrared (MWIR) portions of the electromagnetic spectrum. OKSI has also developed novel algorithms, based on spectral discrimination performed in a transform domain rather than in the spectral domain, for real-time hyperspectral signature analysis. Previous applications of OKSI's technology have been directed specifically toward military target detection and discrimination.
      In an attempt to develop commercial applications for its technology, OKSI decided to pursue the subject of remote, passive detection of pollutants in the atmosphere, soil, and water. In the summer of 1993, OKSI was selected by the Advanced Research Projects Agency (ARPA) to develop a dual-use technology using a staring-array imaging spectrometer sensor that operates in the thermal infrared (TIR) portion of the electromagnetic spectrum. Under the ARPA program, OKSI built a prototype of the Thermal Infrared Imaging Spectrometer (TIRIS) sensor. OKSI then participated in NASA's Visiting Investigator Program (VIP) under the Commercial Remote Sensing Program Office at Stennis Space Center (SSC), Miss., to explore the state of the art in TIR data collection and processing. OKSI wanted to develop signal-preprocessing techniques and to prototype the TIRIS sensor using SSC's Airborne Terrestrial Applications Sensor (ATLAS) to assess the suitability of TIR data for the remote detection of pollutants and hazardous substances. The specific objective of OKSI's VIP project was to utilize the current capabilities of the ATLAS sensor at NASA/SSC to perform the following tasks: 1) assess the feasibility of and examine sensor requirements for remote detection of pollutants, 2) develop a better understanding of the issues related to signal processing, and 3) introduce OKSI to geographic information system (GIS) technology for incorporation into its end products.

Instrumentation and Implementation
ATLAS is a scanning imaging spectrometer with six VNIR, three short-wave infrared (SWIR)/MWIR, and six TIR channels (Table 1). Each scan line of ATLAS data is divided into two blocks: 1) the housekeeping data block that contains time-referenced, digitally recorded flight parameters, ambient conditions, and calibration data, and 2) the video data block which contains the time-referenced, digitally recorded reflected and emitted electromagnetic radiation (EMR) from the target.
      ATLAS data were collected over various areas, including oil refineries and industrial, residential, and agricultural zones. The TIR channels of the ATLAS data were analyzed, and ground temperature and emissivity maps were generated. Such maps are fundamental requirements for airborne remote detection of pollutants, either in the atmosphere or in the ground, because their spectral signatures are measured against the Earth's background.
      OKSI acquired the spectra of the 189 most hazardous air pollutants as defined by the Clean Air Act Amendment of 1992 and listed by the Environmental Protection Agency. The high-resolution spectra were de-resolved to the expected resolution of the TIRIS to demonstrate the capability of the sensor to detect such gases. Given the temperature and emissivity maps of the background terrain, the spectra of the airborne gases can be extracted from the spectrum of the observed thermal radiance and the toxic gases can be identified.
      Four areas of interest were targeted for the OKSI VIP project: 1) petrochemical storage and refinement facilities, 2) agricultural areas adjacent to the Mississippi River, 3) industrial areas adjacent to the Mississippi River, and 4) residential areas. During its VIP project, OKSI studied West Baton Rouge Parish near Baton Rouge and Port Allen, La. (mission date and time: March 31, 1994 at approximately 10:30 a.m. local time). Only the TIR bands of the ATLAS scanner were instrumental to this project; the VNIR, SWIR, and MWIR bands were used as ancillary data. The working environment for the image processing on the project was a UNIX-based Sun SPARC 10 Workstation. All of the corrections, georeferencing, and temperature and emissivity procedures were performed using ERDAS IMAGINE. U.S. Geological Survey 1:24,000 topographic maps were used in georeferencing the data.
      Geometric corrections were necessary to create an image with accurate, consistent geometric relationships between points on the ground and their corresponding representations in the digital imagery. Global Positioning System data (latitude, longitude, and altitude of the aircraft) and Inertial Navigational System data (true heading, pitch angle, and roll of the aircraft) were recorded in the housekeeping data block and used in the geometric correction algorithms. Atmospheric corrections were applied to the TIR data based on a standard atmospheric profile generated by LOWTRAN6 (radiosonde data were not available for this mission). Radiometric corrections were applied to the thermal data based on the thermal spectral response curves obtained from a pre-mission scanner calibration test. Atmospheric and radiometric corrections were necessary to control for atmospheric effects; to alleviate the possibility of scan errors; to incorporate scanner calibrations into the data; and to compensate for variations in scan angle, illumination of the target, and system noise. The output from these two correction procedures represented emitted radiance.
      Sensors detect radiation from the surface (approximately the first 50µm) of a target to determine its apparent surface temperature. Sensors cannot detect a target's kinetic ("true") temperature, which is a measure of the average thermal energy of molecules within a substance. Emissivity represents how efficiently an object absorbs and radiates energy compared to a blackbody (a hypothetical object that is completely non-reflective and is a perfect absorber and emitter of energy). Apparent surface temperature and emissivity were computed from the radiance data.

Observations
Figure 1 shows the derived apparent surface temperature of an agricultural area, a petroleum refinery, and a tank farm adjacent to the Mississippi River in Port Allen, La. Relative surface temperatures are shown with cooler surfaces in purple and warmer surfaces in red. Figure 2 is a false-color temperature-vegetation composite. The three-band combination uses temperature (displayed as red), the normalized vegetation index (displayed as green), and the ATLAS channel 5 "red light" band (displayed as blue) to demonstrate a negative correlation between apparent surface temperature and vegetation or soil moisture content. Figure 3 is an example of reflected energy (top: ATLAS bands 6,4,2 - RGB) versus emitted energy (bottom: ATLAS bands 14,12,10 - RGB).

Conclusions
OKSI's VIP project, though relatively limited in scope, provided stimulating information. The project demonstrated that the ATLAS can be used to imitate the TIRIS, and the ATLAS data can be used to develop data processing procedures and algorithms for calculating ground temperature and emissivity. Based on these results, the next logical step would be a more focused effort that includes the following elements:
1) Validation of the ground spectral emissivity analysis by collection of ground truth data, downwelling and upwelling radiation, and radiosonde data for better atmospheric corrections. Thermal emission needs to be separated from reflected EMR because emissivity is a function of wavelength and is a constant for the range of temperatures of interest, but solar reflection depends on the specific conditions at the time of measurement.
2) Data collection in the presence of controlled air pollutants (e.g., open barrels with organic compounds that release vapors in conjunction with other means, such as Fourier Transform Infrared Spectrometer, for concentration verification).
3) Data acquisition missions (including the acquisition of the appropriate ground-truth data) over known superfund sites and experiments that include controlled releases of a pollutant.
4) Review of efficient data preprocessing techniques for a hyperspectral (as opposed to multispectral) sensor.
5) Joint flights of the ATLAS and TIRIS sensors for cross validation of data.
6) Development of data recording formats for the TIRIS that are compatible with contemporary remote sensing/GIS practices as well as aircraft housekeeping data.
      OKSI personnel, having never before worked with GIS systems, appreciate the value of learning to calculate such parameters as ground temperature and emissivity from sensor data. Based on the results of the VIP effort, OKSI has received funding under the federal Small Business Innovative Research program and is currently developing chemometric algorithms for the detection of airborne pollutants. The TIRIS prototype is scheduled to undergo extensive laboratory characterization and field demonstration tests to determine the TIRIS' ability to detect toxic air pollutants. Once testing is completed, OKSI plans to incorporate all lessons learned into a new TIRIS prototype.

About the Authors:
Dr. Nahum Gat is the president of Opto-Knowledge System Inc. of Manhattan Beach, Calif. He may be reached at 310-372-6665. William Graham is a remote sensing scientist for Lockheed Stennis Operations in support of NASA's Commercial Remote Sensing Program at Stennis Space Center, Miss.

Back