nd (lower
ssimilated
) to each
Figure 1 illustrates a typical ebb tide surface current field in
Juan de Fuca Strait, British Columbia. The measured field
using two SeaSonde radars on opposite sides of the strait is
shown in the upper panel. The modelled field, without
assimilation, is shown in the centre panel, and the blended
field is shown in the lowest panel. The averagé x-field for 3
weeks of hourly assimilations is shown in Figure 2 and
illustrates the expected decrease in k towards the edges of the
coverage area and along the baseline zone where surface
currents cannot be estimated from the radial components.
Figure 2 Spatial Distribution of the Mean Weight « for 500
SeaSonde Current Assimilations.
Oil Slicks
Previous studies using ERS-1 synthetic aperture radar (SAR)
indicated that hydrocarbon oil slicks could be detected in the
imagery under suitable low-wind conditions (Bern et al.,
1992). Using recent RADARSAT SAR images of the Sea
Empress oil spill at Milford Haven in 1996 and the Nakhodka
fuel oil spill in the Sea of Japan in January 1997, Hodgins et
al. (1997a, 1997b) have shown that it is possible to monitor
oil distributions on the sea and classify the images for areas
that are heavily oiled, areas that are lightly oiled and areas
that are free of oil. A composite 3-scene SAR image obtained
on January 11, 1997, for the Nakhodka spill in the Sea of
Japan is shown in Figure 3. The main source of the spilled
fuel oil was from the grounded bow section of the tanker
approximately 200 m offshore of Mikuni. This image
suggests that the heavy oil patch originating at Mikuni is
surrounded by a transition zone of lighter oiling, consistent
with the application of dispersants used by the response
agencies. It also displays low-brightness areas associated
with unstable atmospheric conditions behind a front passing
over the area.
Figure 3 Composite SAR Image of the Nakhodka spill, Jan.
11, 1997.
Image classification involves speckle reduction using a 15x15
Lee adaptive filter, followed by calculation of frequency
distributions of reflectance from selected training areas
spanning heavy oil to background sea and selection of
thresholds to delineate areas with differing oil coverage.
Applying three thresholds to the digital values (DV) over the
entire SAR scene image yields a classified image as shown in
Fig. 4. Matching the thresholds with categories in
Environment Canada's oil observing scheme provides a
relative scaling to oil thickness (Table 1).
The classified images contain the slick features, but also false
features that are atmospheric in origin. Moreover, the area
associated with each slick class will be sensitive to the
thresholds set for the class. In order to provide for human
judgment in the assimilation process, an interactive scheme
allowing the SEACAST modeller to view the SAR image, the
classified SAR image and the oil distribution predicted by the
model within the oil spill model's geographic canvas has been
developed. The software also provides a graphical editor to
map and re-classify the delineated slicks as a set of polygons
linked to oil classes (Fingas et al., 1979). Once the polygon
set has been derived, the volume of oil on the water V(x;,t) is
mapped into the polygons weighting the volumes by the
relative thickness of the classes. The final result is a new
allocation of volume matched to the polygon locations. This
reinitializing editor provides the man-machine link between
SPILLSIM and either the enhanced SAR image, or the
classified SAR image, both of which are viewed as data
layers through the interface.
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 431