Without GPS, the position error incurred in measuring the
spacecraft ephemeris (10", after 2 days) is 500 m along
track and 100 m cross track. When GPS is operational, the
ephemeris errors will be reduced, and the error sources
are indicated as follows:
8 m
Ephemeris
Attitude
Other
Total (RSS)
175 m
36 m
180 m
where the total resultant error is shown by a root-sum-
square (RSS) value.
Eventually, the GPS will provide more accurate ephemeris
information in the future» Attitude error becomes the major
error source that requires attention. Without precise
attitude data to indicate the pointing, ground data proc
essing systems rely upon processing of ground control
points (GCP) to correct geometric error in order to regis
ter images. Registration accuracy then depends on the
number and quality of GCP used and how well these control
points are distributed. Increased numbers of well-dis
tributed control points per image must be available to
improve accuracy. Significant processing capability, human
labor, and time are required for Identifying ground control
points, selecting ground control features, and determining
accurate coordinate values for each GCP. However, some
desirable ground control features cannot be easily identi
fied on an image. For instance, cloud coverage or snow may
obscure the desired features.
Inaccurate GCP locations, lack of well-distributed control
points, or errors in ground control features can seriously
constrain the processing systems. Current Landsat 2-3
ground control data contain errors between 6 to 10 m and
are insufficient to achieve subpixel registration for 10 m
resolution images. In addition, the absence of GCP for
remote and unmapped land areas has limited the registration
since the multiple images cannot be corrected for pixel
location errors. Therefore, ground control is inadequate
for image registration in many areas and new methods of
correcting pointing errors must be provided.
One key in achieving the desired registration accuracy is
to improve attitude accuracy by removing the source of
large position error. Current methods of using GCP for
processing data on the ground require computation, measure
ments, estimations, correlation, etc., which induce numerous
errors including GCP location error, correlation error,
measurement error, computation error, etc.. These residual
errors will eventually translate into a registration error
with a resulting root-mean-square (RMS) value of about 30
meters range even with perfect maps (reference 4, page 2-
41). The resulting RMS exceeds the allowable limit for
subpixel accuracy of TM or MLA data, and is very difficult
to eliminate. Although improvements in ground processing
techniques are anticipated, the image position error caused
by an inadequate attitude system cannot be totally re