Unmixing of satellite thermal images: simulation and application to TM/Landsat data

The Multi-Sensor Multi-Resolution Technique (MMT) was applied to unmix a lower resolution thermal TM image using higher resolution multispectral images obtained in the reflective TM channels. It allowed the authors to reveal high-resolution features, which were smeared or not recognizable in the original thermal image. The MMT accuracy was tested on a simulated data set and proved to be 0.8 K for the mean error of class radiant temperature retrieval and 1.6 K for the pixel error of the low-pass corrected unmixed image when the optimal number of classes was selected.