T HE Landsat satellites have provided an extensive data set for land surface mapping and monitoring at local
and global scales [1], [2]. However, the utility of these data
is often hampered by missing values caused by cloud cover
and sensor-specific problems such as the scan line corrector
(SLC) error in Landsat 7 [3]. The extent of missing data is
even higher in multitemporal analyses due to the dynamic
nature of cloud cover and SLC gaps. Effective methods for gap
filling are needed to make these data more useful for assessing