Use of Multivariate Analysis to Determine Temperature from Low-Resolution Infrared Spectra of Carbon Dioxide

Abstract

The remote optical monitoring of gaseous contaminants is important for both military and industrial applications. An important parameter for quantifying chemical species and for predicting plume dynamics is the temperature. While in some industrial monitoring situations it may be practical to independently measure the temperature of stack emissions, for compliance monitoring and military chemical reconnaissance a remote optical means of estimating gas plume temperature is required. It was noticed that the band shape of low-resolution spectra of carbon dioxide in equilibrium with an exhaust plume was very sensitive to temperature. Spectra of carbon dioxide were acquired under controlled laboratory conditions in 5° increments from 20 to 200 °C. Various multivariate models were used to predict the temperature. It was found that partial least-squares (PLS) was unable to effectively model the simultaneous changes in amplitude and bandwidth with temperature. However, principal component regression (PCR) was found to be well correlated with temperature and allowed cross-validated prediction within 4% error.