|
Accurate retrieval of surface emissivity from longwave infrared hyperspectral imaging data is necessary for many scientific and defense applications. Emissivity retrieval requires an atmospheric scene model and consists of two steps: atmospheric compensation (AC) and temperature-emissivity separation (TES). AC converts the at-aperture radiance to ground radiance, and TES uses the ground radiance to produce a temperature and emissivity estimate. TES assumes smooth emissivity spectra for solids, compared to atmospheric features. These model-based techniques find a high-resolution atmospheric model which provides the smoothest corrected target emissivity. The high-resolution model is then band-averaged to match the sensor’s spectral response function (SRF), which is difficult to characterize and maintain. Any SRF errors are propagated to the atmospheric spectra, emissivity estimates, and lead to an incorrect atmospheric retrieval. The proposed technique improves the quality of the retrieved atmospheric model and emissivity by correcting SRF errors concurrently with model retrieval. High emissivity pixels are found using an in-scene AC technique. The emissivity and temperature of each pixel are identified using a chosen atmospheric model, material library, and featureless atmospheric bands. At-aperture radiances are then generated for reference. The linear relationship between the generated reference and measured at-aperture radiances is used to determine model correction factors. The correction magnitude is dependent on the calibration and model error. The correct atmospheric scene model is determined as the model requiring the smallest correction factor. Using simulated data, we demonstrate the capability of this technique to substantially reduce the effects of calibration error on atmospheric model and emissivity retrieval.
|
