The measurement of GCPs can be automated if suitable reference datasets are available in the form of georeferenced imagery covering the scene area. The reference images need an accuracy as good as or better than the desired target accuracy. The spectral response of the reference images must also match the target image in at least one spectral band so that they are correlated enough for successful matching.

The automatic model adjustment is based on strategies involving:

• Automatic selection of reference images and DEM covering the image to be corrected.
• Selection of large numbers of potential GCP locations.
• Hierarchical matching in different resolutions.
• Robust methods for gross error detection.
• Least-squares adjustment of LOS model parameters.

After the successful model adjustment, the target scene is ready for orthorectification. If the model adjustment fails to produce the required accuracy, the scene is passed on to a fallback of manual GCP identification in the reference data sets and subsequent re-adjustment.

The approach described to automatic model adjustment is presently in use in operational environments. For example, it is used within the Keystone server installed at DMCii Ltd at their UK receiving station where it performs automatic adjustment of all acquired DMC imagery. A global coverage of orthorectified Landsat imagery is successfully used as the reference dataset in this system. A similar function has also been implemented at the JRC in Ispra for automatic orthorectification of imagery for the agricultural subsidy control programme Control with Remote Sensing (CwRS). In this case the reference imagery consists of aerial orthophotos. A presentation of this technology appears in [Westin, Edgardh, 2006].