Cloud Removal Of Multi-Spectral Remote Sensing Imageries

Atmosphere and clouds consist of tiny water drops and ice crystals, which can affect the surface characteristics in remote sensing imageries. The removal of clouds and cloud effect is an important and necessary step in data pre-processing. Valid algorithms of atmospheric correction have been developed. However, the cloud removal in optical data such as Landsat data remains a challenge. In this paper, cloud removal approaches based on a single optical imagery is considered.An algorithm for the removal of thin cirrus cloud as well as alto-thin clouds or thin clouds collectively within visible and near infrared bands was developed. The algorithm removed cirrus clouds using a spectral band centered around the wavelength of 1.38μm, and the remaining thin clouds using the replacement approach. Using a Landsat-8 subimage of 129/39(path/row) acquired on 16 December 2013, we evaluated the algorithm. Thin clouds disappeared visually. Reflectance values of Bands 1–4 decreased in both steps. Reflectance values of Band 5 decreased in step one, and then stayed the same. With a nearly cloud-free image acquired on 30 November 2013 as the “truth”, the spatial correlation coefficients of cloud-covered pixels within the December image were 0.84 or higher. Changes in reflectance values of Bands 1–5, and the high correlation coefficient values indicated the validity of the algorithm.However, there is the concern about the image replacement approach. The replacement relies on the ability of short-wave infrared radiation to penetrate through the clouds. If there is doubt in cloud penetration, the replacement can be questionable. In addition, if the cloud-covered area is over a water surface or the cloud-free area is water, the replacement approach becomes erroneous because water absorbs almost all the solar radiation in infrared bands(e.g., Bands 6 and 7 of Landsat-8).Of the concerns, another approach to remove clouds was developed with independent component analysis(ICA). Within cloud-covered areas, histograms were derived to quantify changes of the reflectance values before and after the use of the algorithm. Referred to a cloud-free image, changes of histogram curves validated the algorithm. Scatterplots were generated and linear regression performed for the reflectance values of each band before and after the algorithm, and compared to those of the reference image. Band-by-band, results in cloud removal were acceptable. The algorithm had little effect on pixels in cloud-free areas after the analyses of histograms, scatterplots, and linear regression equations.