Abstract
Thermal infrared multispectral sensor data should be atmospherically corrected for estimating surface temperature and spectral emissivity. For land observations, a common method for it is using transmittance, path radiance and sky radiance estimated by a radiative transfer code such as MODTRAN. However, this method has the disadvantage that it requires both the atmospheric profile data simultaneously observed and the digital elevation model (DEM); if the quality of these data is low, the atmospherically corrected data is unreliable. We will, therefore, suggest a new method for estimating atmospheric correction parameters, surface temperature and spectral emissivity without the atmospheric profile data and the DEM.
The proposed method is based on using pixels with high emissivity for all channels (gray pixels). The method is as follows:
1) Gray pixels are selected from a target area.
2) For gray pixels, upward radiance at surface level is estimated by an extended multi-channel (EMC) method which produces the upward brightness temperature of each channel at surface level by the linear combination of the observed brightness temperatures of all channels.
3) The transmittance and path radiance of each channel are estimated by the linear regression between the observed radiance and the estimated upward radiance at surface level.
4) The sky radiance is estimated by using transmittance and path radiance. As the results, all the atmospheric correction parameters are estimated. We have named this the gray pixel (GP) method. 5) For all pixels, the surface temperature and the spectral emissivity are estimated by one of the temperature and emissivity separation methods from atmospherically corrected radiance.
6) As an option, gray pixels are selected again with estimated emissivity, and each parameter is recalculated for improved accuracy.
In the final part of the study, the effectiveness of the proposed method is evaluated using the Thermal Infrared Multispectral Scanner (TIMS) data. The traditional method, as well as the proposed one, is applied to the data, and the results of both methods are compared to each other. This evaluation demonstrates that the proposed method is feasible.
