BACKGROUND
Clouds and their shadows are often a significant problem when using satellite imagery to examine features on the Earth's surface. Their masking, and exclusion from analysis, is an important pre-processing step in many applications. Automating the masking is especially important in operational situations where hundreds of images are routinely being processed.
Conventional pixel-based image classification techniques generally perform poorly for clouds and their shadows, especially in imagery that has limited spectral resolution, as they can be spectrally similar to many other features. The SICASM project has instead taken an object based approach, developing methods to identify cloud objects and their respective shadow objects based on four bands (green, red, near-infrared and mid-infrared) and the sun/sensor angles at acquisition time.
Some of the methods evolved from those described by Hegarat-Mascle and Andre (2009). This work was also used in developing the fmask algorithm for cloud and cloud-shadow detection in Landsat TM/ETM+ imagery by Zhu and Woodcock (2012), which probably works better than SICASM programs, as it uses all available TM/ETM+ bands. A matlab implementation of fmask is availble at google code.
REFERENCES
- Le Hegarat-Mascle, S and Andre, C (2009) Use of Markov Random Fields for automatic cloud/shadow detection on high resolution optical images. ISPRS Journal of Photogrammetry and Remote Sensing, 64, 351-366.
- Zhu, Z and Woodcock, CE (2012) Object based cloud and cloud shadow detection in Landsat Imagery. Remote Sensing of Environment, 118, 83-94.
PUBLICATIONS
The following papers have been published, describing the methods used in SICASM programs.
- Fisher, AG and Danaher, T (2011) Automating woody vegetation change detection at regional scales: the problem of clouds and cloud shadows. 34th International Symposium on Remote Sensing of Environment, Sydney, Australia, April 2011. Download
- Fisher, A (2014) Cloud and cloud-shadow detection in SPOT-5 HRG imagery with automated morphological feature extraction. Remote Sensing, 6, 776-800.