Thin cloud removal for Landsat 8 OLI data using independent component analysis

Using independent component analysis (ICA) coupled with the quality assessment (QA) band of Landsat 8, an approach for thin cloud removal in Landsat 8 operational land imager (OLI) data was developed. After the ICA transformation of the visible, near infrared, short-wavelength and cirrus bands of OLI data, cloud component was identified by the mixing matrix. Then, a cloud mask derived from the analysis of the QA band was formed such that an image pixel with and without cloud cover was delineated. The cloud component and cloud mask were used to remove the thin clouds. Thin clouds disappeared visually within the OLI data. Using another cloud-free image acquired in the previous overflight as the reference, we assessed the accuracy level of the cloud removal. Before and after the cloud removal, the spatial correlation coefficients increased from 0.69 to 0.83 in band 1, 0.75 to 0.86 in band 2, 0.81 to 0.88 in band 3, 0.87 to 0.91 in band 4, and no change in bands 5, 6, and 7 for pixels identified with cloud cover.