Fast Quantification of Air Pollutants by Mid-Infrared Hyperspectral Imaging and Principal Component Analysis

Abstract

An imaging Fourier-transform spectrometer in the mid-infrared (1850–6667 cm−1) has been used to acquire transmittance spectra at a resolution of 1 cm−1 of three atmospheric pollutants with known column densities (Q): methane (258 ppm·m), nitrous oxide (107.5 ppm·m) and propane (215 ppm·m). Values of Q and T have been retrieved by fitting them with theoretical spectra generated with parameters from the HITRAN database, based on a radiometric model that takes into account gas absorption and emission, and the instrument lineshape function. A principal component analysis (PCA) of experimental data has found that two principal components are enough to reconstruct gas spectra with high fidelity. PCA-processed spectra have better signal-to-noise ratio without loss of spatial resolution, improving the uniformity of retrieval. PCA has been used also to speed up retrieval, by pre-calculating simulated spectra for a range of expected Q and T values, applying PCA to them and then comparing the principal components of experimental spectra with those of the simulated ones to find the gas Q and T values. A reduction in calculation time by a factor larger than one thousand is achieved with improved accuracy. Retrieval can be further simplified by obtaining T and Q as quadratic functions of the two first principal components.

Keywords

• infrared imaging
• multispectral and hyperspectral imaging
• air pollution monitoring
• remote sensing and sensors
• spectroscopy
• fourier transform