Comparison of Spectral and Interferogram Processing Methods Using Simulated Passive Fourier Transform Infrared Remote Sensing Data

Abstract

Computer-generated synthetic single-beam spectra and interferograms provide a means of comparing signal processing strategies that are employed with passive Fourier transform infrared (FT-IR) sensors. With the use of appropriate radiance models and spectrometer characteristics, synthetic data are generated for one-, two-, and four-component mixtures of organic vapors (ethanol, methanol, acetone, and methyl ethyl ketone) in two passive FT-IR remote sensing scenarios. The single-beam spectra are processed by using Savitsky–Golay smoothing and first-derivative and second-derivative filters. Interferogram data are processed by Fourier filtering using Gaussian-shaped bandpass digital filters. Pattern recognition is performed with soft independent modeling of class analogy (SIMCA). Quantitative models for the target gas integrated concentration-path-length product are built by using either partial least-squares (PLS) regression or locally weighted regression (LWR). Pattern recognition and calibration models of the filtered spectra or interferograms produced comparable results. Discrimination of target analytes in complex mixtures requires a sufficiently large temperature differential between the infrared background source and analyte cloud. Quantitative analysis is found to be possible only when the temperature of the analyte cloud is stable or known and differs significantly from the background temperature. Net analyte signal (NAS) methods demonstrate that interferogram and spectral processing methods supply identical information for multivariate pattern recognition and calibration.