A Revised ICA Algorithm to Remove Cirrus Cloud Effects in Spectral Data of Landsat-8 Bands 1–7

A revised independent component analysis (ICA) algorithm was developed to improve the performance in the removal of cirrus cloud effects within spectral data of Landsat-8 Bands 1-7. The prior was that the reflectance values in the cirrus spectral band should be zero or near zero after the cloud removal. Then, the cloud component derived using the forward ICA transformation was revised and use to remove the cloud effects. After the inverse ICA transformation, the cloud removal was completed. Applying the algorithm to the Landsat-8 sub-image of 041/036 (path/row) acquired on 14 March 2015, we noted that cirrus clouds in Bands 1-7 disappeared visually. The cloud-free sub-image of the same location acquired on 1 April was extracted and used as the reference image in the algorithm verification. After the algorithm, R
²
values of the March image and reference image increased significantly. The revised algorithm was effective and valid. The revised one outperformed the original ICA algorithm in the comparison of both algorithms.