This is taken fom the Satellite Imagery FAQ, and was written in 1995 by Nick Kew, now of WebÞing. In view of the explosion of Internet resources since 1995, the FAQ is now largely obsolete and not maintained, although parts such as this may be worth reproducing.
Because they're fun, of course! :-)
Continental to global scale images are useful if they show information that is studied at a large scale, such as the state of the global biosphere. One major measure is NDVI, which characterises 'greenness' (see RS/Vegetation FAQ for details). Global NDVI images taken regularly over time - at intervals between one and two weeks - enable scientists to study change in the biosphere in detail.
The AVHRR Pathfinder and Global 1KM projects have created global land datasets showing NDVI (together with lower-level data) from AVHRR imagery, at resolutions up to 1.1KM. The global images are created by mosaicing a large number of individual scenes, taken over ten-day periods. Individual scenes are first stitched to generate half-orbits (in principle south to north pole, but generally broken because only daytime data is used)! The half orbits are then stitched together, with reference to a digital chart of the world.
The key to compositing for NDVI is that each point on the Earth's surface is replicated in several images over the sampling period. Only the best NDVI value is selected, so bad data (such as cloud cover) is discarded.
Yes, Landsat data is just as well-suited to computing NDVI as is the AVHRR.
The NOAA satellites, in a polar orbit at an altitude of 833 KM, orbit the Earth fourteen times per day. The AVHRR instrument images a 2400-KM wide swath as it passes. Thus every point on the Earth's surface is viewed at least about once per day (the exact frequency of course varies with latitude).
The Landsat series (4-5), in near-polar orbits at 705 KM, also orbit the Earth fourteen times per day. However, the swath imaged is just 185KM, so a point on the equator may be viewed only once in sixteen days. The data with which to generate weekly, ten-day or fortnightly global composites is simply not available. A sixteen-day composite would of course be subject to considerable cloud-cover (see below).
Having said that, it is certainly possible to make large-area Landsat mosaics. NASA's Landsat Pathfinder Project (see http://www.bsrsi.msu.edu/) has created such datasets for the study of tropical deforestation.
As noted above, only the best NDVI values from each input dataset is used. Clouds will necessarily generate very low NDVI values - clouds are not green!. Hence clouds are automatically filtered out in the compositing process, provided there is at least one cloudless view of a point within the sample. Thus cloudlessness is not in fact guaranteed, but is statistically far more likely than for a single pass. Alternatively, it can be assured by collecting data over an unlimited time period; c.f. the GeoSphere project).
Clearly this will work if and only if the characteristics being studied are dissimilar to any cloud in at least one of the available bands!
Several outdated links in the original SATFAQ have been deleted.
All the above references deal with global land datasets. NASA's pathfinder program created also Ocean and Atmospheric datasets: