2012
e maximum
when the an-
-PALSAR is
s, as shown
t-2 and En-
'gions in the
f Radarsat-2
n of Bragg
used in the
represents a
actor, which
ar wave and
amping fac-
The relative
sing angles,
udy we use
y data from
at 29.122N,
on oil rig.
15
the relative
ice from the
teratively at
formed in a
tage the im-
ie first stage
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
the image is multilooked to 2°, and to 2* at the second stage.
Both the intensity thresholding and damping factor methods pro-
vide results at each stage which are combined using:
P(O|n) * P(Old)
(On) * P(O]d) + P(W|n) * P(W|d)
JRO) == (4)
where J P(O) is the joint probability for oil, P(O|n), is the prob-
ability of oil given the NRCS, P(O|d) is the probability of oil
given the damping factor analysis, P(W |n) is the probability of
clear water given the NRCS, and P(W |d) is the probability of
clear water given the damping factor analysis. The main reason
for combining the two probability functions is that the damping
factor analysis will return low probabilities for the center of large
slicks. This is due to the fact that the damping factor is calculated
against a moving window average. For oil slicks larger than the
size of the moving window, the average and pixel values will be
very close, returning no dampening. Of course a larger window
size can be used to eliminate this problem, however, since that
would require human intervention. Contrary to the damping fac-
tor analysis, the NRCS based thresholding algorithm will return
high probabilities for the center of the slick.
The study area is shown in Figure 3, and is about 350km x 350km,
centered around the Deepwater Horizon oil spill. The locations
of the ALOS-PALSAR and Radarsat-2 imagery are shown in the
figure. It should be noted that even though the SAR imagery is
calibrated to NRCS, there is still a gradual change in intensity
along the range direction (Figure 3). The Gulf of Mexico oil spill
provides a great test case for the new algorithm, because it is lo-
calized and continuous over time. Furthermore there are many
published research and ground observations available for validat-
ing the method.
Figure 3: The black-box shows our designated study area.
GSHHS shoreline is shown in light green. The SAR intensity
images are from Radarsat-2, and the one on top is acquired on
April 24th 2010. Light green boxes show the footprints for the
ALOS PALSAR imagery. The legend shows distance in degrees.
3 RESULTS AND DISCUSSION
SAR images acquired from Radarsat-2 and Alos were processed
using the proposed method. Some of the imaging parameters
and environmental conditions are summarized in Table 1. Table
shows the observed wind-speed, relative angle between the wind
and radar wave, and the calculated damping factor (D.F.). The
damping factor calculation also takes into account the wave group
velocity, and radar incidence angle which are not listed in the ta-
ble. The damping factors listed for ALOS-PALSAR are rather
low, however they are still above the NESZ of the instrument,
65
which is about —29dB for the fine beam single (FBS) imaging
mode.
Date Sensor Wind Speed Wind Angle — D.F.
[m/s] [degrees] [dB]
04-27 . Radarsat-2 7.3 173.6 6.6
05-01 Radarsat-2 9.8 228.4 7.6
05-01 ' Alos-Palsar 7.8 77.0 -16.5
05-04 Alos-Palsar 3.2 50.3 -8.5
Table 1: Data Table
The results of oil spill detection algorithm is shown in Figure 4.
The probability maps calculated for the five iterative steps us-
ing two different methods and their combination are presented.
The results are shown in ascending order of resolution, where the
2° multilooked image is located at the left hand side. The final
solution for the processed imagery is shown in the bottom right
corner. The joint probability (JP(O)) is calculated using the
NRCS based oil probability (P(O|n)) and damping factor based
oil probability (P(O|d)) as shown in (4).
Multilook 5 4 3 2 1
P(Oln)
=
9
x
Figure 4: Results for the first (point probability) and second (spa-
tial probability) processing steps, for the Radarsat-2 data acquired
on April 27th, 2010.
The complementing behavior of the two methods can be seen in
Figure4. While the P(O|n) has very little noise at high multi-
looking, the opposite is true for P(O|d), which becomes less and
less noisy with decreasing multilooking. Furthermore, the void
in the center of the oil spill is visible in P(O|d) results for level
five.
Final results of all the images processed in this study are shown in
Figure 5. The analysis using Radarsat-2 imagery obtained better
results compared to the Alos imagery. This is very likely due to
the small damping factors that are obtained at L-band, as shown
in Figure 1. The current algorithm does not employ any weighting
to the data, therefore the combined probability of all observations
are inconclusive. This can be improved however, by implement-
ing a more complex filter to the final stage, such as a Kalman
filter, or by simply adding more C-band data to dominate the re-
sults. We recently acquired additional imagery from Envisat to
test our hypothesis. Utilizing a Kalman filter at the final step of
the algorithm will allow for utilizing a larger spectrum of methods
and data sets, which may only be useful under certain conditions.
It is also worth noting that the Alos-PALSAR imagery, acquired
on May Ist, 2010 at 04:10 UTC shows almost no brightness vari-
ation.
