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
Kim et al., 2010), (3) segmentation techniques (Barni et al., 1995,
Solberg et al., 2007), (4) slick feature extraction (Fiscella et al.,
2000, Del Frate et al., 2000). Some algorithms also combine
ocean drift models to assist data analysis (Espedal, 1999, Cheng
et al., 2011). In addition to SAR intensity, co-polarization differ-
ences of multiple polarization SAR data is also suggested for oil
spill detection (Migliaccio et al., 2009).
In this paper, our focus is not on developing a method that per-
forms better than any one of the methods mentioned earlier, but
instead to develop a practical framework where results from dif-
ferent methods and sources can be combined to provide a joint
solution, which is likely to have less uncertainty. The method
can be easily extended to include any geospatial observation (e.g.
optical remote sensing, ground measurements), however in this
paper our focus is on SAR imagery. The algorithm is composed
of three calculation steps: (1) Pixel (point) probability, (2) spatial
probability, and (3) spatio-temporal probability. Point probabil-
ity is calculated based on normalized radar cross section, where
darker pixels get higher probabilities for oil contamination (Barni
et al., 1995). Spatial probability is based on the damping ratio
given the current wind conditions and imaging parameters (Gade
et al., 1998, Kim et al., 2010). Point and spatial analysis results
are then combined to provide a joint probability for oil slick at
each acquisition. The spatio-temporal probability is estimated
from multiple SAR acquisitions separated shortly in time, result-
ing in a time varying probability of oil slick over target area. All
analysis in this paper is done over intensity calibrated, geocoded
SAR imagery. Images are calibrated to normalized radar cross
section (NRCS). The imagery is resampled to a common geome-
try using a sinc interpolator, and land-masked using Global, Self-
consistent, Hierarchical, High-Resolution Shoreline data (GSHHS)
(Wessel and Smith, 1996).
First and second steps of the algorithm are iterative, and are per-
formed at the same time. The first step of the algorithm is a sim-
ple dark-object selection routine based on intensity thresholding.
At this step, dark areas of the image are assigned a higher proba-
bility for an oil spill. The initial threshold for the first step is the
noise equivalent sigma zero (NESZ), which is the sensors noise
floor. A probability value for each pixel is assigned based on it’s
intensity such that:
P(Wi|es) =
P(Oloo)
(co — min(co))/(T — min(co)) (1)
1 — P(W|oo) (2)
|
where P(W |oo) is probability of oil-free water given the NRCS,
T 'is the threshold, and P(Oloo) is the probability of oil given the
NRCS is the complement of P(W oo). The P(W |oo) is modi-
fied by bringing all larger values to 1, constraining the probability
values between 0 and 1. Furthermore, iterations start using a mul-
tilooked imagery, to reduce the effect of speckle noise. It is worth
noting that the multilooking operation changes the dynamic range
of the radar imagery. In order to keep the thresholds equivalent at
each iteration, multilooked images are further calibrated to have
the same dynamic range as the full resolution image.
The second step of the algorithm starts with calculation of damp-
ing factor after Gade et al. (Gade et al., 1998, Kim et al., 2010).
The estimated damping factors at different wind speeds as a func-
tion of Bragg wavenumbers are shown in Figure 1. Figure 1 also
shows the maximum theoretical observation range for Radarsat-
2, Envisat-ASAR, and Alos-PALSAR SAR systems. The Bragg
wavenumber is defined as (Gade et al., 1998):
ks = 2ko sin(¢) (3)
where kg is the Bragg wavenumber, ko is the radar wavenum-
64
ber, and ¢ is the incidence angle. Figure 1 shows the maximum
expected damping amount which would be observed when the an-
gle between the wind, and radar wave is zero. ALOS-PALSAR is
not expected to be effective in mild wind conditions, as shown
in the figure. The SAR sensors on-board Radarsat-2 and En-
visat are both C-Band, and therefore cover similar regions in the
wavenumber domain. The slightly larger coverage of Radarsat-2
is due to it’s larger range of incidence angles.
=>
N
oo
BA
Damping Factor [dB]
0 200 300
Bragg Wavenumber
Figure 1: Estimated damping factor as a function of Bragg
wavenumber. The theoretical range of SAR systems used in the
system are marked with horizontal lines. Each curve represents a
different wind velocity.
Wind is an important factor to estimate the damping factor, which
is also dependent on the relative angle between radar wave and
wind direction. Figure 2 shows a plot of expected damping fac-
tors at speeds between 2.5 m/s and 12.5 m/s. The relative
wind direction is plotted at counter-clockwise increasing angles,
starting from zero at the horizontal axis. In this study we use
wind speed, direction and ocean wave group velocity data from
National Data Buoy Center station 42040, located at 29.122N,
88.207W, about 40km NNW of the Deepwater Horizon oil rig.
180
Figure 2: Estimated damping factor plotted against the relative
angle between radar and wind direction. Radial distance from the
center indicates the wind speed.
The first and second steps of the algorithm are run iteratively at
five different spatial resolutions. The analysis is performed in a
pyramid structure with five different stages. At each stage the im-
age is multilooked to the power of two, such that at the first stage
