The identification of cloud shadow pixels is important for land and cloud–atmosphere remote-sensing applications. In this study, a stepwise cloud shadow detection approach for Moderate Resolution Imaging Spectroradiometer (MODIS) 1 km reflectance data, which combines a geometry-based
method, a threshold-based automated training data extraction, and a support vector machine (SVM) classification-based spectral detection process, is presented. An extended potential cloud shadow mask is generated according to the satellite and solar geometry and the positions of clouds. An
automated training sample data-extraction process, which is based on the reflectance characteristics of cloud shadows, is performed to acquire training samples. Accurate cloud shadow pixels are then confirmed by the SVM classification algorithm. The advantage of this approach is that only
reflectance data, geolocation data, and a cloud mask are required; no further cloud or atmospheric information, such as cloud-top height, cloud type, or aerosol information are needed in the workflow. The reduced input requirements benefit rapid-response remote-sensing applications such as
flood detection and monitoring. Experimental results were compared with the spectral-based cloud shadow detection scheme, which was employed in the MOD35 product. The comparisons indicate that the new approach detects cloud shadows better than the results generated by using spectral threshold
No Reference information available - sign in for access.
No Citation information available - sign in for access.
No Supplementary Data.
No Article Media
Document Type: Research Article
Department of Geography and Geoinformation Science, College of Science,George Mason University, Fairfax,VA,22030, USA
Center for Satellite Application and Research,NOAA/NESDIS, Camp Spring,MD,20748, USA
Publication date: January 10, 2013
More about this publication?