Abstract
Masking of cirrus clouds in optical satellite imagery is an important step in automated processing chains. Firstly, it is a prerequisite to a subsequent removal of cirrus effects, and secondly, it affects the atmospheric correction, i.e., aerosol and surface reflectance retrievals. Cirrus clouds can be detected with a narrow bandwidth channel near 1.38 μ m and operational detection algorithms have been developed for Landsat-8 and Sentinel-2 images. However, concerning cirrus removal in the case of elevated surfaces, current methods do not separate the ground reflected signal from the cirrus signal in the 1.38 μ m channel when performing an atmospheric correction, often resulting in an overcorrection of the cirrus influence. We propose a new operational algorithm using a Digital Elevation Model (DEM) to estimate the surface and cirrus cloud contributions in the 1.38 μ m channel and to remove cirrus effects during the surface reflectance retrieval. Due to the highly variable nature of cirrus clouds and terrain conditions, no generic quantitative results could be derived. However, results for typical cases and the achieved improvement in cirrus removal are given for selected scenes and critical issues and limitations of the approach are discussed.
Highlights
Satellite imagery is frequently contaminated by high-altitude cirrus clouds in the upper troposphere and in the stratosphere
The proposed method consists of two parts, (i) cirrus detection based on elevation thresholds, and (ii) cirrus removal with the remaining 1.38 μm signal component after subtracting the estimated ground surface contribution
Cirrus masking is conducted with the λ = 1.38 μm channel using a certain threshold on the TOA reflectance, e.g., ρ∗(1.38) > 0.01, to avoid very thin cirrus or instrument noise
Summary
Satellite imagery is frequently contaminated by high-altitude cirrus clouds in the upper troposphere and in the stratosphere. The occurrence of cirrus clouds is larger than 50% over the midlatitude and tropical regions [1]. Thin cirrus is difficult to detect with visible/near infrared (NIR) bands because land surfaces show a high degree of spatial nonuniformity. The detection and removal of cirrus, which may strongly vary in a scene, is of particular interest for an improved quantitative processing, and it helps in the interpretation of data. It is of scientific interest as well as of practical importance
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have