82
s. LI, Z. CHENG, AND W.E WEEKS
divergence/convergence. Therefore, using open, closing, shear zones and ice floes identified in the tracer procedure as single entities, the error in the separate estimation of
divergence/convergence can be significantly reduced. An accuracy of 0.1% in estimation of deformation for a field of 50 x 50 km or larger is achievable.
Severe mislocation caused by an erroneous match of tie points is another issue.
Direct estimation of the impacts of this type of error is difficult. An evaluation of the
ASF ice motion product (Stern et al.1993) shows that 97-98% of the ice motion vectors
are free of this type of error. The implication of this statement is that about 2-3% of the
tie points are erroneous. However, four facts contribute to the minimization of the
influence of this type of error on our analysis. (1) A strict quality assurance procedure
is adopted at ASP. All of the questionable ice motion products, which are characterized
by low values in an automatic rating of motion tracking, are subject to manual inspection by an ASF operator, and only those which pass the manual inspection are delivered to users. (2) The procedural use of the edge point trimmer eliminates the majority of bad tie points by trimming the edge of the ice motion products where most bad
tie points are located. (3) Interior bad tie points with low values in the rating of motion
tracking are treated as gaps, and their locations on the target image are recalculated in
the gap filler. This tends to reduce the impact of location errors on the results of our
analysis. (4) Aggregation of divergence and convergence values in leads and large ice
floes through the tracer procedure further eliminates errors in deformation calculations when there are bad tie points in the interior of the concerned features. Furthermore, the random nature of this type of error tends to result in nonsignificant changes
in the aggregated statistics.
Nonlinearity of the boundaries of the deformed grid cells can also be a source of error.
It can have significant impact on the accuracy of the deformation calculation of the
individual cells with severe shear. However, Fily and Rothrock (1990) suggest that the
treatment of individual features, such as a lead or a large ice floe, as single entities also
reduces errors of this type, because aggregation reduces subdivisions, and as a result,
reduces the possibility of errors. Therefore, separate derivation of divergence and convergence values of the image area using the feature-aggregated deformation values
instead of values from the individual grids is adopted in the following analysis.
Some extreme cases of severe mislocation or nonlinearity may cause large errors in
aggregation of deformation values. However, those cases are also characterized by large
deformation values. Our method does not eliminate the necessity of inspecting cases
with large aggregated deformation values. However, it helps the investigator identity
those events where further inspection is warranted.
4.6
Examples and Discussion
The automatic algorithm described here makes possible the quick screening of a large
number of ASF ice motion data files for regions of significant ice deformation. It also
makes possible a quick extraction of intermediate-scale sea ice deformation data. Two
examples are given here to demonstrate the use of this algorithm.
In the first example, 140 GPS ice motion data files, covering a period between October 10, 1991, and March 30, 1992, were collected from a circular region with a radius of
200 km, centered at 81 0 N, 170 0 W in the Chukchi Sea sector of the Arctic Ocean.
s. LI, Z. CHENG, AND W.E WEEKS
divergence/convergence. Therefore, using open, closing, shear zones and ice floes identified in the tracer procedure as single entities, the error in the separate estimation of
divergence/convergence can be significantly reduced. An accuracy of 0.1% in estimation of deformation for a field of 50 x 50 km or larger is achievable.
Severe mislocation caused by an erroneous match of tie points is another issue.
Direct estimation of the impacts of this type of error is difficult. An evaluation of the
ASF ice motion product (Stern et al.1993) shows that 97-98% of the ice motion vectors
are free of this type of error. The implication of this statement is that about 2-3% of the
tie points are erroneous. However, four facts contribute to the minimization of the
influence of this type of error on our analysis. (1) A strict quality assurance procedure
is adopted at ASP. All of the questionable ice motion products, which are characterized
by low values in an automatic rating of motion tracking, are subject to manual inspection by an ASF operator, and only those which pass the manual inspection are delivered to users. (2) The procedural use of the edge point trimmer eliminates the majority of bad tie points by trimming the edge of the ice motion products where most bad
tie points are located. (3) Interior bad tie points with low values in the rating of motion
tracking are treated as gaps, and their locations on the target image are recalculated in
the gap filler. This tends to reduce the impact of location errors on the results of our
analysis. (4) Aggregation of divergence and convergence values in leads and large ice
floes through the tracer procedure further eliminates errors in deformation calculations when there are bad tie points in the interior of the concerned features. Furthermore, the random nature of this type of error tends to result in nonsignificant changes
in the aggregated statistics.
Nonlinearity of the boundaries of the deformed grid cells can also be a source of error.
It can have significant impact on the accuracy of the deformation calculation of the
individual cells with severe shear. However, Fily and Rothrock (1990) suggest that the
treatment of individual features, such as a lead or a large ice floe, as single entities also
reduces errors of this type, because aggregation reduces subdivisions, and as a result,
reduces the possibility of errors. Therefore, separate derivation of divergence and convergence values of the image area using the feature-aggregated deformation values
instead of values from the individual grids is adopted in the following analysis.
Some extreme cases of severe mislocation or nonlinearity may cause large errors in
aggregation of deformation values. However, those cases are also characterized by large
deformation values. Our method does not eliminate the necessity of inspecting cases
with large aggregated deformation values. However, it helps the investigator identity
those events where further inspection is warranted.
4.6
Examples and Discussion
The automatic algorithm described here makes possible the quick screening of a large
number of ASF ice motion data files for regions of significant ice deformation. It also
makes possible a quick extraction of intermediate-scale sea ice deformation data. Two
examples are given here to demonstrate the use of this algorithm.
In the first example, 140 GPS ice motion data files, covering a period between October 10, 1991, and March 30, 1992, were collected from a circular region with a radius of
200 km, centered at 81 0 N, 170 0 W in the Chukchi Sea sector of the Arctic Ocean.