2012
, Chengdu,
993. The
d by rader
;.P., Wang,
; and land
ology and
[.M., 2005.
layers in
se Journal
elation in
Geoscience
Ib. On the
differential
phys. Res,
idence and
of Tongji
and land
he Chinese
1-4.
orithms for
date and its
eying and
d by 973
al Natural
41074005,
he Central
U11ZT13,
0.112130).
1nance and
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
AUTOMATIC CALCULATION OF OIL SLICK AREA FROM MULTIPLE SAR
ACQUISITIONS FOR DEEPWATER HORIZON OIL SPILL
Batuhan Osmanoëlu!, Coskun Ozkan?, Filiz Sunar?, Gordon Staples*
University of Alaska - Fairbanks, P.O. Box 757320, Fairbanks, AK 99775, batu@gi.alaska.edu
?Erciyes University, Engineering Faculty, Geodesy and Photogrammetry Engineering Dept., 38039 Kayseri,Turkey, cozkan@erciyes.edu.tr
?Istanbul Technical University, Civil Engineering Faculty, Geomatics Engineering Dept., 34469 Maslak Istanbul, Turkey, fsunar(a/itu.edu.tr
4 MDA, 13800 Commerce Parkway, Richmond, V7S 1L5, Canada, gstaples@mdacorporation.com
Commission VII, WG VII/2
KEY WORDS: Marine pollution, Oil spill detection, Probability map, SAR
ABSTRACT:
The Deepwater Horizon oil spill occurred in the Gulf of Mexico in April 2010 and became the largest accidental marine oil spill in
history. Oil leaked continuously between April 20th and July 15th of 2010, releasing about 780, 000m? of crude oil into the Gulf of
Mexico. The oil spill caused extensive economical and ecological damage to the areas it reached, affecting the marine and wildlife
habitats along with fishing and tourism industries.
For oil spill mitigation efforts, it is important to determine the areal extent, and most recent position of the contaminated area. Satellite-
based oil pollution monitoring systems are being used for monitoring and in hazard response efforts. Due to their high accuracy, frequent
acquisitions, large area coverage and day-and-night operation Synthetic Aperture Radar (SAR) satellites are a major contributer of
monitoring marine environments for oil spill detection.
We developed a new algorithm for determining the extent of the oil spill from multiple SAR images, that are acquired with short
temporal intervals using different sensors. Combining the multi-polarization data from Radarsat-2 (C-band), Envisat ASAR (C-band)
and Alos-PALSAR (L-band) sensors, we calculate the extent of the oil spill with higher accuracy than what is possible from only one
image. Short temporal interval between acquisitions (hours to days) allow us to eliminate artifacts and increase accuracy.
Our algorithm works automatically without any human intervention to deliver products in a timely manner in time critical operations.
Acquisitions using different SAR sensors are radiometrically calibrated and processed individually to obtain oil spill area extent.
Furthermore the algorithm provides probability maps of the areas that are classified as oil slick. This probability information is then
combined with other acquisitions to estimate the combined probability map for the spill.
1 INTRODUCTION remote sensing sensors, as well as field observations can be com-
bined, and statistically provide an outcome better than any indi-
Marine oil spills are a common threat to all sea bordering coun- vidual resource.
tries. The environmental and related economical losses due to an
oil spill can be large, and many methods have been developed
to monitor oceans in operational conditions. In most cases it is
critical to respond to spills in a timely manner, and therefore it is
important for such operational systems to provide results in near
real time. Once a spill is detected, its behavior can be modeled
based on physical models.
In this paper we develop a fully automatic oil spill monitoring
systems that is capable of combining data from multiple SAR
sources. The focus of this paper is on radar imagery, however
the algorithm can be then be expanded to optical imagery and
ground measurements as discussed later. A brief background on
the technical aspects is provided in Section 2. Results obtained
from SAR data over the Deepwater Horizon oil spill are discussed
i ; 5 ; in Section 3.
Synthetic Aperture Radar (SAR) systems provide a viable option m SION
of oil slick monitoring. SAR intensity images are sensitive to
surface roughness which is altered in the case of an oil spill. The 2 BACKGROUND
scattering of oil-free ocean surface is dominated by Bragg scat-
tering (Bragg, 1913). A thin oil sheen covering the ocean will ^ Synthetic Aperture Radar imagery is sensitive to surface rough-
reduce the ocean-atmosphere interaction, and alter the smooth- ^ ness which is altered in case of an oil spill (Alpers and Hühnerfuss,
ness of the surface. Therefore, oil slicks appear slightly darker in 1988). Oil slicks change the smoothness of ocean surface and
moderate wind conditions in SAR imagery. SAR systems have appear darker compared to surrounding oil free ocean, however
an ideal range of wind speed and direction where they are most the amount of damping is affected with wind and wave condi-
sensitive, and do not perform well for oil spill detection under tions. Furthermore the speckle effect in SAR imagery limits the
very windy or very calm conditions. reliability of point measurements in the image, causing spurious
results (Brekke and Solberg, 2005). In our approach we apply a
There are of course other methods to monitor oceans and detect multiple step processing to limit the adverse effects of speckle,
oil spills, such as optical and infra-red remote sensing, and hy- instead of filtering the data with a speckle filter.
perspectral imaging (Brekke and Solberg, 2005). No matter how
complex or advanced a method is single-handedly all have dif- There are many methods developed to detect oil spills from SAR
ferent levels of uncertainties. Therefore it is important to com- intensity images: (1) machine learning and neural-network recog-
bine many observations complementing each other's weaknesses nition (Kubat et al., 1998, Ozkan and Sunar, 2007), (2) frequency
providing the best possible information. Results from different spectrum attenuation (Lombardini et al., 1989, Gade et al., 1998,
63
