Evaluation of Six Methods for Extracting Relative Emissivity Spectra from Thermal Infrared Images

The performance of six published methods for extracting relative spectral emissivity information from thermal infrared multispectral data has been evaluated. In the first part of this article, we recall those six methods and show mathematically that they are almost equivalent to each other. Then, using simulated data for the TIMS (Thermal Infrared Multispectral Scanner) instrument, we analyze the sensitivity of those methods to different sources of error which may occur in real data such as errors due to 1) method simplification, 2) instrumental noise and systematic calibration error, 3) uncertainties on the estimation of downwelling atmospheric radiance, and 4) uncertainties of atmospheric parameters in atmospheric corrections. In terms of resulting errors in relative emissivity, the results show that: a) all methods are very sensitive to the uncertainties of atmosphere. An error of 20% of water vapor in midlatitude summer atmosphere (2.9 cm) may lead to an error of 0.03 (rms) for Channel 1 (worst case) of TIMS. b) The effect of the atmospheric reflection term is very important. If this term is neglected in method development, this may lead to an error of 0.03 (rms) for Channel 1 and midlatitude summer atmosphere. This is the case for the alpha method. c) Instrumental noise commonly expressed by noise equivalent difference temperature (NEΔ T) from 0.1 K to 0.3 K results in an error on relative emissivity ranging from 0.002 to 0.005 for all methods. d) Error on relative emissivity due to the instrumental calibration error (systematic error) is negligible. The study also shows that the relative emissivity derived with deviate atmosphere is linearly related to its actual value derived with correct atmospheric parameters. Based on this property, we propose three methods to correct for the errors caused by atmospheric corrections under horizontally invariant atmospheric conditions. A practical analysis with the real TIMS data acquired for Hapex-Sahel experiment in 1992 supports the results of this simulation.