An empirical and radiative transfer model based algorithm to remove thin clouds in visible bands

• Published in: Remote sensing of environment Vol. 179; pp. 183 - 195
• Main Authors: Lv, Haitao, Wang, Yong, Shen, Yang
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 15.06.2016
• Subjects: Algorithms; Clouds; correlation; Empirical analysis; Histograms; Image acquisition; land cover; land use; Landsat; Landsat-8; Pixels; radiative transfer; Radiative transfer model; remote sensing; Spectra; spectral analysis; Standard deviation; Thin cloud and its removal; Visible bands; Thin cloud and its removal; Landsat-8; Visible bands; Radiative transfer model; Empirical analysis
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2016.03.034

An algorithm for cloud removal in visible bands was developed. Thin clouds inside visible Bands 1–4 of Landsat-8 data acquired on 8 January 2014 disappeared after the algorithm. Values of mean and one standard deviation decreased, band-by-band. The reduction was supported by the leftward shift of the histogram curve in each band. To validate the algorithm, we used the cloud-free image acquired on 23 December 2013 of Landsat-8 as the reference image. Among the January image before the algorithm, the January image after the algorithm, and the reference image, values of mean and one standard deviation of the January image after the algorithm were much closer to those of the reference image. Histogram curves of the January image after the algorithm and the reference image were almost overlapped entirely. Spatial correlation coefficients of the January image before the algorithm and reference were 0.496, 0.547, 0.656, and 0.730 for Bands 1–4, respectively. Coefficients of the January image after the algorithm and reference image became 0.782, 0.822, 0.840, and 0.885 for Bands 1–4. In cloud-free areas, the algorithm did not alter spectral characteristics of cloud-free pixels. Thus, the algorithm was not only able to remove thin clouds, but also to preserve spectral characteristics of cloud-free pixels. The algorithm was then applied to other land use and land cover (LULC) types, and images acquired in other locations and seasons by Landsat-8 and WorldView-2 sensors. Results in cloud removal were satisfactory. Finally, this algorithm outperformed three widely-used cloud removal algorithms in comparison. •The characteristics of visible and near infrared bands were analyzed and summarized.•Six parameters of the algorithm were determined.•The algorithm for removing thin clouds in visible bands was developed and assessed.•The algorithm recovered ground features well in cloud-covered areas.•The algorithm was applicable in different land use and land cover types.