IEEE Geoscience and Remote Sensing Magazine - September 2021 - 146

One of the difficulties of deriving OSCs from GOCI
hourly imagery is that the features are typically not produced
by individual objects but are calculated as spatial
distribution statistics for one bio-optical product.
In some cases of clear and uniform water, using a small
template window might create ambiguity and result in
a wrong match within the search region. An incorrect
match might even have a higher correlation coefficient
value that cannot be ruled out by simply setting a larger
threshold. Therefore, some quality control filtering approaches
have to rely on information from neighborhood
vectors that may also be suspicious. The fact that
the selection of the threshold value for each quality control
filtering process always involves a tradeoff between
unreliable information and less matching motivated us
to propose two objective criteria-size-irrelevant and
majority filtering-to initially rule out suspicious vectors
with minimum preset thresholds.
Taking robust feature characteristics into consideration,
the derived vectors should remain the same even if the STW
is slightly changed. In other words, a reliable feature should
0.12
0.16
0.2
0.25
0.29
be irrelevant in terms of the STW. By contrast, suspicious
vectors are derived from unreliable features that are more
sensitive to the STW. A very different result is obtained once
the STW is modified. Based on this characteristic, the second
image's coregistration process is repeated by specifying
a smaller STW (in this work, it is four pixels smaller in
the x and y directions). In Figure 4, the displacement vectors
derived from the original STW are labeled in red. They
are green if the result from the smaller STW is unchanged.
That is, the remaining vectors labeled in red are identified
as suspicious by introducing a smaller STW through sizeirrelevant
filtering.
After carefully examining the green displacement vec0.12
0.16
(a)
0.2
tors,
however, a few can still be found that are rather different
from their neighborhood ones. This ambiguity is
mainly created in regions where the water is clear or uniform
and where there are no apparent features. The incorrect
matches resulted from weak features being too persistent
to rule out through size-irrelevant filtering. The good thing
is that the number of error-prone vectors was quite small
at that point, so it could be reasonably assumed that the
majority of the displacement vectors were correct. For each
displacement vector, an estimated vector could be interpolated
from the remaining displacement ones. The deviation
between this vector and its estimated one could then
be used as an anomaly indicator. To avoid selecting a fixed
threshold value, we followed the approach described in [15]
and [16] to determine a value of 1.63 by searching the inflection
point in the histogram of the anomaly indicator,
as in Figure 4(b). The displacement vectors with anomaly
indicator values lower than 1.63, namely, the majority filtering,
are labeled in yellow. Note that both size-irrelevant
filtering and majority filtering are implemented at the pixel
level and that a total of seven elliptical regions is specified
for a corresponding number of eddies. They were used for
the later quantitative analysis.
0.25
0.29
(b)
FIGURE 5. (a) The streamlines derived from all the displacement
vectors [labeled in red, green, blue, and yellow in Figure 4(a)].
(b) The streamlines derived from only the reliable displacement
vectors [labeled in yellow in Figure 4(a)]. Five regions affected by
suspicious vectors are denoted by cyan ellipses to illustrate that
the unrealistic flows are corrected after removing the suspicious
vectors, resulting in more realistic streamlines.
146
PATHLINES VERSUS STREAMLINES
The main concept for deriving OSCs from GOCI hourly images
involves tracking features and continuously connecting
the vectors derived from each pair of consecutive images,
as illustrated in [33, Fig. 3]. To be precise, these vectors
are pathlines that record the path of a fluid element in the
flow every hour. As mentioned, haze and clouds limit the
availability of GOCI images, and it is rare to have a complete
pathline from eight cloudless units of GOCI hourly data. In
other words, 8-h pathlines might not be an ideal product in
terms of data availability. By contrast, once we have a time
slot when there is a clear sky and acquire a pair of consecutive
images, as in Figures 2(b) and 4, not one but many vectors
can be derived. These images can be used to provide
streamline products that are instantaneously tangent to the
velocity vector of the flow, as demonstrated in Figure 5. Generally
speaking, the streamlines match the vortex structures
very well. The density and pattern of the streamlines show
the velocity and direction in which a massless fluid element
IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE SEPTEMBER 2021
IEEE Geoscience and Remote Sensing Magazine - September 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2021
