squares approach. An influence cin;le around each observation restricts the extension of
the local plane and thus the number of corresponding grid nodes. The size of this circle
depends on the orbit characteristics taking into account the density of the satellite track
pattern. It is chosen slightly larger than the equatorial separation of the satellite tracks (80
km for ERS-1.ALT.OPR2 due to the 35 day repeat cycle and 30 km for ERSl.ALT.QLOPR) to ensure a contribution of at least one ascending and one descending
track for each grid node. Due to the high orbit quality of TOPEX data an influence circle
extension (30 km) much smaller than the equatorial separation was chosen (310 km),
which implies that only grid nodes very close to TOPEX track were considered. A
weighting scheme inversely proportional to the square of the distance was chosen. In
order to preserve the maximum along-track gradients only grid nodes very close to the
track receive high weighted contributions.
The size of the local planes, which determine how much grid nodes are influenced by a
single measurement and therefore how smooth the resulting mean sea surface will be,
must be considered very carefully. On one hand, the model should preserve as many
along track gradients as possible. On the other hand, the model should have cross track
properties without gradients from unremoved orbit errors or seasonal effects. Adjacent
tracks could have semiannual time differences (ERS-l 168 day cycles), which can lead
to 20 to 40 cm amplitude differences (Wunsch 1994). Unremoved orbit errors and
temporal effects result in errors of all wavelength, which must be eliminated within the
gridding. If these errors are not eliminated a shaded relief representation will reveal sharp
along track gradients which cannot be attached to ocean signatures. Error removal and
gridding must attempt to minimize track related problems and maximize high frequency
information. These contrary conditions lead to local planes of MSS95A, that result in
lower gradients in the gradient test as one part of the quality control, but error free cross
track representations.
QUALITY CONTROL
For the validation of MSS95A, three different quality control procedures were applied:
1)
Comparisons to external models
2)
Gradient method
3)
Relief representation
As mentioned above each quality procedure reveals important results, but only the
combination of the three shows, how good or bad the model actually is.
COMPARISONS TO EXTERNAL MODELS
This test was performed by subtracting the new OSU mean sea surface, that combines
altimeter data of ERS-l, TOPEX and Geosat (RAPP 1995), from MSS95A:
212
the local plane and thus the number of corresponding grid nodes. The size of this circle
depends on the orbit characteristics taking into account the density of the satellite track
pattern. It is chosen slightly larger than the equatorial separation of the satellite tracks (80
km for ERS-1.ALT.OPR2 due to the 35 day repeat cycle and 30 km for ERSl.ALT.QLOPR) to ensure a contribution of at least one ascending and one descending
track for each grid node. Due to the high orbit quality of TOPEX data an influence circle
extension (30 km) much smaller than the equatorial separation was chosen (310 km),
which implies that only grid nodes very close to TOPEX track were considered. A
weighting scheme inversely proportional to the square of the distance was chosen. In
order to preserve the maximum along-track gradients only grid nodes very close to the
track receive high weighted contributions.
The size of the local planes, which determine how much grid nodes are influenced by a
single measurement and therefore how smooth the resulting mean sea surface will be,
must be considered very carefully. On one hand, the model should preserve as many
along track gradients as possible. On the other hand, the model should have cross track
properties without gradients from unremoved orbit errors or seasonal effects. Adjacent
tracks could have semiannual time differences (ERS-l 168 day cycles), which can lead
to 20 to 40 cm amplitude differences (Wunsch 1994). Unremoved orbit errors and
temporal effects result in errors of all wavelength, which must be eliminated within the
gridding. If these errors are not eliminated a shaded relief representation will reveal sharp
along track gradients which cannot be attached to ocean signatures. Error removal and
gridding must attempt to minimize track related problems and maximize high frequency
information. These contrary conditions lead to local planes of MSS95A, that result in
lower gradients in the gradient test as one part of the quality control, but error free cross
track representations.
QUALITY CONTROL
For the validation of MSS95A, three different quality control procedures were applied:
1)
Comparisons to external models
2)
Gradient method
3)
Relief representation
As mentioned above each quality procedure reveals important results, but only the
combination of the three shows, how good or bad the model actually is.
COMPARISONS TO EXTERNAL MODELS
This test was performed by subtracting the new OSU mean sea surface, that combines
altimeter data of ERS-l, TOPEX and Geosat (RAPP 1995), from MSS95A:
212