Construction of forest attribute maps using satellite imagery
typically requires reference data in the form of ground observations,
although interpretations of aerial photography are often used as
surrogates for ground observations. Two crucial tasks are necessary to
construct spatially accurate maps: first, the satellite imagery must be
rectified to a geographical coordinate system using ground control
points whose coordinates are known and that can be readily identified
on the imagery, and second, the locations of the reference observations
in the same geographical coordinate system must be determined.
The latter task is usually accomplished using Global
Positioning System (GPS) receivers that have been calibrated to
produce coordinates in the same system as the one to which the
imagery has been rectified. Following accomplishment of these two
tasks, the relationship between the land cover classes of the reference
observations and the image spectral data is estimated, and the land
cover class for each image pixel is predicted