254
R.KwOK
is saved in twenty I-dB bins. The backscatter history will allow the correlation of surface properties of the sea ice to atmospheric and surface conditions.
Date of Melt Onset/Freeze-Up. There is a fairly well-defined change in the backscatter of the snow/ice at the onset of melt and freeze-up. The analysis algorithm will detect
and estimate the date this seasonal transition using the time series of backscatter product. Different parts of the Arctic go through these transitions at different dates and this
product would provide a high spatial resolution of the date of transition. The temporal resolution is,however, dependent on the repeat observations of the Lagrangian cells.
Daily Gridded Fields of Pressure, Temperature and Geostrophic Wind. The gridded
surface pressure and wind fields are derived by interpolation of the pressure and wind
fields obtained from the National Meteorological Center. The analyzed temperature field
is a 6-hourly, 2-m temperature data set (Martin and Munoz 1997) provided by the
POLES (POLar Exchange at the Sea Surface), a NASA interdisciplinary project. The gridded air temperatures are estimated using an optimal interpolation scheme with temperature inputs from drifting buoys, manned drifting stations, coastalland weather stations' and ship reports. These are air temperature estimates used in the conversion of
ice age to ice thickness.
11.3.3
Sampling Issues and Measurement Errors
Spatial and Temporal Undersampling. The spatial sampling of the small scale motion
field is limited by the density of grid points. In this case, the spatial sampling of the
motion field is on the order of 5 km, thus smaller-scale motion details are not resolved
and the field is undersampled. The deformation of a region of ice is represented by
these deforming polygons, so the geometry of the region is very coarsely represented.
We do not know the quantitative effect of this spatial undersampling. The motion field
is also temporally undersampled. The sampling rate is dependent on the repeat observation of a region which in our case is constrained by orbit geometry and repeat cycles.
Again, the motion details between observations are not resolved and the temporal
record is undersampled. The temporal undersampling has a direct effect on the computation of total open water production and ridging. These parameters will be underestimated, as was pointed out by Fowler et al. (1996), because opening/closings ofleads
probably happen at a time scale which is shorter than the sampling period possible
with current SAR observations. Coon et al. (1996) studied the effects of undersampling using buoy data and suggested correction factors to compensate for the weekly
sampling issue.
Measurement Errors. There are two primary sources of error in measuring ice
motion from satellite imagery: (1) the geolocation uncertainty of each image pixel and
(2) the tracking error which is the uncertainty in identifying common features in an
image pair. A more detailed discussion of these errors can be found in Holt et al. (1992).
For ERS-l imagery, the uncertainty in ice displacement is approximately 0.3 km. Deformation computations are typically not affected by geolocation errors if these errors are
positional biases. For SAR imagery, the geolocation errors are biases and the variations
are negligible within an image. Stern et al. (1995) estimated uncertainty in divergence
computed from the GPS ice motion products to be approximately 0.008. For
RADARSAT, these errors need to be reassessed.
R.KwOK
is saved in twenty I-dB bins. The backscatter history will allow the correlation of surface properties of the sea ice to atmospheric and surface conditions.
Date of Melt Onset/Freeze-Up. There is a fairly well-defined change in the backscatter of the snow/ice at the onset of melt and freeze-up. The analysis algorithm will detect
and estimate the date this seasonal transition using the time series of backscatter product. Different parts of the Arctic go through these transitions at different dates and this
product would provide a high spatial resolution of the date of transition. The temporal resolution is,however, dependent on the repeat observations of the Lagrangian cells.
Daily Gridded Fields of Pressure, Temperature and Geostrophic Wind. The gridded
surface pressure and wind fields are derived by interpolation of the pressure and wind
fields obtained from the National Meteorological Center. The analyzed temperature field
is a 6-hourly, 2-m temperature data set (Martin and Munoz 1997) provided by the
POLES (POLar Exchange at the Sea Surface), a NASA interdisciplinary project. The gridded air temperatures are estimated using an optimal interpolation scheme with temperature inputs from drifting buoys, manned drifting stations, coastalland weather stations' and ship reports. These are air temperature estimates used in the conversion of
ice age to ice thickness.
11.3.3
Sampling Issues and Measurement Errors
Spatial and Temporal Undersampling. The spatial sampling of the small scale motion
field is limited by the density of grid points. In this case, the spatial sampling of the
motion field is on the order of 5 km, thus smaller-scale motion details are not resolved
and the field is undersampled. The deformation of a region of ice is represented by
these deforming polygons, so the geometry of the region is very coarsely represented.
We do not know the quantitative effect of this spatial undersampling. The motion field
is also temporally undersampled. The sampling rate is dependent on the repeat observation of a region which in our case is constrained by orbit geometry and repeat cycles.
Again, the motion details between observations are not resolved and the temporal
record is undersampled. The temporal undersampling has a direct effect on the computation of total open water production and ridging. These parameters will be underestimated, as was pointed out by Fowler et al. (1996), because opening/closings ofleads
probably happen at a time scale which is shorter than the sampling period possible
with current SAR observations. Coon et al. (1996) studied the effects of undersampling using buoy data and suggested correction factors to compensate for the weekly
sampling issue.
Measurement Errors. There are two primary sources of error in measuring ice
motion from satellite imagery: (1) the geolocation uncertainty of each image pixel and
(2) the tracking error which is the uncertainty in identifying common features in an
image pair. A more detailed discussion of these errors can be found in Holt et al. (1992).
For ERS-l imagery, the uncertainty in ice displacement is approximately 0.3 km. Deformation computations are typically not affected by geolocation errors if these errors are
positional biases. For SAR imagery, the geolocation errors are biases and the variations
are negligible within an image. Stern et al. (1995) estimated uncertainty in divergence
computed from the GPS ice motion products to be approximately 0.008. For
RADARSAT, these errors need to be reassessed.