1432 Volume 57, Number 11, 2003 APPLIED SPECTROSCOPY 0003-7028 / 03 / 5711-1432$2.00 / 0 q 2003 Society for Applied Spectroscopy Remote Detection of Heated Ethanol Plumes by Airborne Passive Fourier Transform Infrared Spectrometry TOSHIYASU TARUMI, GARY W. SMALL,* ROGER J. COMBS, and ROBERT T. KROUTIL Center for Intelligent Chemical Instrumentation, Department of Chemistry and Biochemistry, Clippinger Laboratories, Ohio University, Athens, Ohio 45701-2979 (T.T., G.W.S.); and Los Alamos National Laboratory, P.O. Box 1663, MS E543, Los Alamos, New Mexico 87545 (R.J.C., R.T.K.) Methodology is developed for the automated detection of heated plumes of ethanol vapor with airborne passive Fourier transform infrared spectrometry. Positioned in a xed-wing aircraft in a downward-looking mode, the spectrometer is used to detect ground sources of ethanol vapor from an altitude of 2000–3000 ft. Chal- lenges to the use of this approach for the routine detection of chem- ical plumes include (1) the presence of a constantly changing back- ground radiance as the aircraft ies, (2) the cost and complexity of collecting the data needed to train the classication algorithms used in implementing the plume detection, and (3) the need for rapid interferogram scans to minimize the ground area viewed per scan. To address these challenges, this work couples a novel ground-based data collection and training protocol with the use of signal process- ing and pattern recognition methods based on short sections of the interferogram data collected by the spectrometer. In the data col- lection, heated plumes of ethanol vapor are released from a portable emission stack and viewed by the spectrometer from ground level against a synthetic background designed to simulate a terrestrial radiance source. Classiers trained with these data are subsequently tested with airborne data collected over a period of 2.5 years. Two classier architectures are compared in this work: support vector machines (SVM) and piecewise linear discriminant analysis (PLDA). When applied to the airborne test data, the SVM classiers perform best, failing to detect ethanol in only 8% of the cases in which it is present. False detections occur at a rate of less than 0.5%. The classier performs well in spite of differences between the backgrounds associated with the ground-based and airborne data collections and the instrumental drift arising from the long time span of the data collection. Further improvements in classi- cation performance are judged to require increased sophistication in the ground-based data collection in order to provide a better match to the infrared backgrounds observed from the air. Index Headings: Remote sensing; Fourier transform infrared; FT- IR; Ethanol; Pattern recognition; Support vector machine. INTRODUCTION Passive Fourier transform infrared (FT-IR) spectrom- etry has been used for a variety of applications in envi- ronmental monitoring. Its capability of detecting and identifying analytes remotely allows regulatory monitor- ing of emissions from industrial stacks, 1 remote sensing of volcanic gases, 2 satellite atmospheric monitoring, 3 and toxic cloud imaging. 4 In most environmental monitoring applications, measurements can be performed either by the instrument xed on the ground to view a target or by a downward-looking airborne sensor passing over the tar- get. With the airborne instrument, the applications of pas- sive FT-IR spectrometry can be extended to monitoring Received 14 April 2003; accepted 20 June 2003. * Author to whom correspondence should be sent. gases released from the sites of chemical incidents or mapping air pollution in large areas. However, several technical difculties must be addressed for the successful implementation of these airborne measurements. One of the most challenging problems in the use of a downward-looking airborne spectrometer is the constant- ly changing background radiance from the ground. Un- like conventional FT-IR measurements made in labora- tory settings, this variance in the background radiance makes it impractical to obtain reference spectra for use in data processing. Without reference spectra, extracting analyte signatures from the background and identifying them becomes difcult. One way to solve this problem is to collect as many background spectra as possible with the airborne instrument and build classication algo- rithms that can distinguish the analyte signatures among the variety of background spectra. However, this ap- proach is not practical in terms of the cost and effort involved in performing such a large-scale data collection with the airborne instrument. A more practical approach is to apply appropriate signal processing methods to the data to remove the effects of differences in the back- ground radiance. If this background suppression strategy is effective, it may be possible to build the classication algorithms with controlled data collected on the ground, thereby rendering the training of the detection algorithms less costly and more practical. Another major challenge in the application of an air- borne instrument to ground-based targets is the integra- tion time (i.e., speed of interferogram collection) of the spectrometer. For example, a slow-ying aircraft with a ground speed of 100 knots covers 1 m in approximately 0.02 s. The spectrometer must acquire data at 50 inter- ferograms/s in order to receive information about a 1 m distance along the ight track. Stated differently, one air- craft overight of a 10 m target would produce at most 10 interferograms with information about the target. The airborne instrument must thus acquire interferograms much faster than a conventional laboratory spectrometer in order to pinpoint a ground target. In a series of studies, our laboratory has demonstrated that using short segments of digitally ltered interfero- grams in the analysis can help to overcome these tech- nical difculties. 5–10 The interferogram representation of broad background spectral features decays much faster than the corresponding representation of a narrower an- alyte feature, thereby allowing extraction of analyte in- formation by selecting an appropriate interferogram seg- ment. Furthermore, the use of short interferogram seg-