Get documents about "Use and analyse of satellite SAR images
Document Sample
CENTRE DE DOCUMENTATION
DE RECHERCHE ET D’EXPERIMENTATIONS
SUR LES POLLUTIONS ACCIDENTELLES DES EAUX
USE AND ANALYSE
OF SATELLITE SAR IMAGES
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 1
USE AND ANALYSE OF SATELLITE SAR IMAGES
FOR OIL SPILLS DETECTION
CONTENTS
1 INTRODUCTION 3
2 HOW TO DETECT AND IDENTIFY OIL SPILLS IN SAR IMAGES 4
2.1 GENERAL CONSIDERATION .................................................................. 4
2.2 TECHNICAL CONSIDERATIONS .............................................................. 4
2.3 SAR CAPACITIES TO MONITOR THE OCEAN SURFACE ............................. 6
2.3.1 Examples of atmospheric SAR signatures ................................................................6
2.3.2 Examples of oceanic SAR signatures .......................................................................8
2.3.3 Ship detection on SAR images..................................................................................9
2.4 IDENTIFICATION OF AMBIGUOUS SLICKS .............................................. 10
3 HOW TO ANALYSE SAR IMAGE AND HOW RELIABLE IS THE IDENTIFICATION OF OIL
SPILLS? 13
3.1 CONTEXT ANALYSIS AND OBJECTIVES ................................................. 13
3.1.1 Monitoring accidental disasters ...............................................................................13
3.1.2 Routine surveillance of oil pollutions from tanker operations ..................................13
3.1.3 Statistical analysis of SAR images for long assessment of oil pollutions................14
3.2 METHODOLOGY ................................................................................ 14
3.2.1 Analysis of the slicks candidates .............................................................................14
3.2.2 How can the use of ancillary data improve the detection?......................................15
3.3 CLASSIFICATION OF THE DETECTED SLICKS ......................................... 16
4 OPERATIONAL USE OF SAR IMAGES 17
4.1 USER REQUIREMENTS ....................................................................... 17
4.2 OIL SPILL DETECTION PORTFOLIO ...................................................... 17
4.3 DESCRIPTION OF THE OIL SPILL DETECTION SCHEME............................ 17
4.3.1 On the choice of appropriate SAR image modes ....................................................17
4.3.2 Methodology ............................................................................................................18
4.3.3 Where to collect ancillary information......................................................................19
REFERENCE 20
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 2
1 INTRODUCTION
Spillage of oil either in open sea or in coastal waters may have a severe impact on the natural resources and
the economic health of the area at stake. Thus, there is a need to set-up operational processes to rapidly and
reliably detect, track, and monitor oil spills and to predict their drift. The accidental pollution caused by ships in
distress usually catches the imagination of people and the dramatic consequences on the sea environment
are widely illustrated and broadcast through mass media. However, these pollution events only represent a
small amount of the whole marine oil-pollution problem. The most important source of ship pollution arises
from ship intentional oily discharges, which may result from various operations: ballast water, tank washings,
and engine-room oily waters discharges [15].
The distinction between accidental oil spill disaster and oil pollution due to ship intentional discharges
is a crucial issue. Not only do the remote sensing signatures of these events differ in terms of shapes and
extension, but these different situations will also induce different approaches in terms of
• the choice of appropriate remote sensing techniques and ad hoc image parameters (e.g. coverage
versus spatial resolution of Synthetic Aperture Radar imagery modes)
• the methodology for remote sensing analysis (exhaustive reporting versus minimisation of false
positive alarms)
This document intends to provide guiding information concerning the use of Satellite Synthetic Aperture Radar
(SAR) imagery for Oil Spill Detection (OSD).
The second chapter presents the current state of the art of the detection and identification of oil spills in SAR
images. The SAR capacity to detect oil spills on the sea surface is presented in terms on the local sea state
conditions as well as technical characteristics of the sensor configuration. It is also shown that the SAR
imagery offers to the end-users much larger possibilities to retrieve valuable information on the on-going
oceanic and atmospheric situations and/or human activities at the time of the sensor acquisition. Some these
observable dynamic processes having SAR signatures similar to oil spill, they may sometimes induce
ambiguous interpretations which turn into positive false alarms in the SAR oil spill detection schemes.
The third chapter presents a methodology to analyse SAR images as well as an attempt to assess the
reliability of slicks identification.
The operational use of SAR images to detect oil spill will also be addressed in the last part.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 3
2 HOW TO DETECT AND IDENTIFY OIL SPILLS IN SAR IMAGES
2.1 General consideration
Sea-surface slicks, oil spills but also films of natural surfactant material, modify surface tension, therefore
having a strong impact on radar backscatter level which is decreased and results in dark patches in
comparison with the surroundings [2][3][4]. Local sea-state conditions stand as one of the most critical
limitations. Indeed, since the sea-surface roughness is primarily driven by local wind conditions, sufficient wind
speed is required to allow enough contrast between spilled and clean areas. Thus, low wind conditions are the
major source of false negative alarms (e.g. situations where oil spills are actually present on the sea surface
but could not be detected by SAR imagery).
Low wind conditions do not provide enough signal contrast for detection. In the case of the VV-polarised C-
band ERS-1/2 and ENVISAT SAR sensors, this minimum wind speed is about 2.5 m/s.
In the case of high winds situations, the radar inability to observe oil spills on the sea surface is due to
physical process as surface turbulence and waves dispersion drag slicks into the ocean subsurface. It is
generally agreed that the upper wind limit for oil spill detection ranges between 12-14 m/s [15]. In fact the
ability to detect an oil slick will depend on the importance and nature of the slick but also on the evolution of
the local conditions (wind, currents) from the very beginning of the spill (there is a cumulative dispersion and
spreading which is of uttermost importance, but we seldom have the full history of those local conditions that
are determining the detectability of the slick, as well as its drift)
The local wind sea surface being the primary indicator to a priori assess the feasibility oil spills detection, it is
then essential to take this parameter into consideration when performing statistical analysis of oil spills
occurrence over specific areas. For example, an oil-spill detection demonstration using Envisat SAR imagery
was performed during year 2004 over the French Zone de Protection Ecologique (ZPE) in the Mediterranean
Sea. As part of this demonstration, high resolution wind fields were extracted from approximately 80 ENVISAT
ASAR Wide Swath images. Based on the criterion that the oil spill detection is a priori limited to wind speeds
ranging between 2.5 and 12.5 m/s, it was shown that only 60% of the sea surface observed by SAR images
was actually usable [12][13].
2.2 Technical considerations
However limitations do not only arise from local sea state conditions nor from oil type and thickness. SAR
instrument characteristics also play a major role in the capacity to detect oil spills on the sea surface: radar
frequency, polarisation, incidence angle, spatial resolution, etc. An attempt is made to briefly present their role
in the detection capabilities of SAR imagery.
At present, all the existing operational satellite SAR missions are operating in C-band (ERS-2, ENVISAT,
RADARSAT-1). The very recently launched Japanese SAR instrument ALOS, operating in L-band, cannot
presently be considered as fully operational. Nevertheless, the radar frequency issue has to be seriously
considered as a number of the next planned satellite SAR missions will operate in different bands (e.g.
TerraSAR-X and COSMO-Skymed in X-band). To assess the role of the radar frequency, some multi-
frequency scatterometer measurements have been conducted. Those studies revealed that the oily/clean sea
contrast ratio increases with the frequency [6][19] (Figure 1). As observed, oil spill detection is expected to
provide better contrast ratio at higher radar frequencies.
Although the scattering mechanism that actually occurs on the sea surface at incidence angles ranging
between 20 and 50 degrees is still a matter of on going discussion (regarding in particular the role of breaking
waves contribution), it is however generally admitted that the Bragg scattering mechanism dominates.
Following this theory, the lower radar frequencies respond to longer ocean surface waves. Hence, the
damping of “responding” waves is expected to be more efficient at high radar frequencies (i.e. shorter radar
wavelengths) such as C or X bands
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 4
Figure 1: Radar contrast as a function of frequency for two kind of fuel [19]
The incidence angle (that is the angle between the transmitted electromagnetic (e.m.) waves and the normal
of the sea surface) also changes the contrast ratio between oily areas and clean areas. The imaging
mechanism varies as a function of incidence angle as well as the typical sea surface wavelengths that interact
with e.m. signal. Since the latter react progressively to the wind forcing, with increasing energy when the
wavelength increases, the contrast ratio will also change correspondingly.
Figure 2: Radar contrast as a function of incidence angle in Ku-Band [19]
The influence of polarisation in the SAR capability to detect oil spills is more difficult to assess. As stated by
the radar theory, SAR backscattered signal is expected to be higher in VV-polarisation that in HH-polarisation.
In practice, the vertical (VV) polarisation for both transmission and reception will be preferred to any other
configuration for oil-spill detection when this is available (e.g. ERS-1/2, ENVISAT).
Eventually, the spatial resolution is also of importance. Typically, SAR images products that offer wide
coverage (e.g. ENVISAT ASAR Wide Swath or Radarsat-1 ScanSAR modes) usually have poor spatial
resolution (between 50-150 meters). Nevertheless, such products have revealed to be sufficient for normal
operational surveillance of intentional en route ship discharges. Indeed, this spatial resolution still allows
detection as well as the possibility to measure the length of en route discharges (but not the width, this one
being of the same dimension as the pixel size). On the opposite, higher resolution products have limited
swath but they will be preferred in case of the monitoring of widespread accidental spills as the area of interest
is known a priori and exhaustive detailed reporting of potential slicks is wanted over this area, as all these
potential slicks will be checked by aerial or sea surface crafts.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 5
2.3 SAR capacities to monitor the ocean surface
As it is essential to present the physical and practical limitations of SAR imagery to detect oil spills, then the
knowledge of the SAR capacity to visualise all the dynamic ocean processes occurring at the time of the
acquisition is of uttermost importance They are related to physical and biological oceanic and/or atmospheric
phenomena or even to human activities. The knowledge of these phenomena is crucial in particular in the
setup of semi-automatic oil spill detection schemes supervised by a trained operator. The information brought
by the analysis of all phenomena captured by the SAR is not only useful to interpret ambiguous SAR
signatures that can lead to positive false alarms (see next paragraph) but also to support subsequent steps of
an oil spill service such as drift modelling.
2.3.1 Examples of atmospheric SAR signatures
The Figure 3(a) presents typical SAR signature of katabatic winds blowing down a sloping terrain and over
the near shore area on an ENVISAT ASAR Wide swath products acquired in the evening (6 Jan. 2004 at
21.17 UTC). These winds are generated because in the evening, the air near the surface cools faster over the
land than over the sea. This turns into a down-hill flow of the cold air which is not always captured by
numerical weather predictions models (NWP). SAR images bring very interesting information on such local
wind flows occurring at specific period during the day (land/sea breeze, katabatic winds). Further south on the
same figure, large regions with very low backscatter signals are associated with very low wind speeds.
The Figure 3(b) shows a number of atmospheric signatures that are frequently observed near atmospheric
fronts. Two distinct regions can clearly be observed on either part of a SW-NE oriented atmospheric front.
The wind is blowing from the N-NW direction on the upper left side of the atmospheric front while it is blowing
from the SW direction on the right side below the atmospheric front. SAR signatures of atmospheric
boundary layer (ABL) rolls are particularly visible in the left part of the image. These features are generally
aligned with the wind direction and can be detected using either Fourier analysis [7] or Radon transforms [18].
On the same figure, atmospheric gravity waves can also be observed running along the atmospheric front
(right side of the upper part of the front). Such waves, the typical wavelength of which range between 5 and 10
km, are frequently observed near atmospheric fronts that are the place of rapid varying wind speed at the sea
surface. Atmospheric gravity waves also manifest themselves in layered atmosphere and are often generated
behind mountains in which case they are called lee waves. Eventually, Figure 3(b) also shows typical SAR
signature of rain events on the upper left side of the atmospheric front. The sea surface roughness is here
affected by the airflow associated with the rain event leading to typical elliptic patterns. All these events are
very often associated with the presence of an atmospheric front. The SAR capacity to provide the exact
location of the atmospheric front can be very useful to adjust NWP models outputs to will be used to model the
drift of the oil spills detected using SAR imagery.
The Figure 3(c) presents SAR signatures of ABL rolls in the NW part of the SAR image that feature the wind
flow of the well-known Tramontane and Mistral winds that frequently occur in the French Mediterranean Zone
de Protection Ecologique (ZPE). Wind funnelling associated with atmospheric gravity lee waves can be
clearly observed between Corsica and Sardinia. Atmospheric lee waves can also be clearly observed in the
NE of Corsica, as well as Wind relief shadowing caused by the Italian Capraia or Elba islands.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 6
(a)
(b)
(c)
Figure 3: Typical atmospheric signatures (a): katabatic winds near the coast, (b): atmospheric front
and rain events, (c): wind funelling between Corsica and Sardinia (from VV-polarized ENVISAT ASAR
Wide Swath products).
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 7
2.3.2 Examples of oceanic SAR signatures
The Figure 4(a1) presents examples of SAR signatures of oceanic fronts on an ERS-2 SAR image acquired in
the Black Sea the 25 July 1999. The modulation of the sea surface roughness at the interface of different
water masses are caused by waves-current interactions. The modulation rate is function of the interactions
and the geometry of the scene with regard to the sensor line-of-sight. The sea surface temperature as
captured by AVHRR instrument is presented on Figure 4(a2). It clearly corroborates the position of the
different water masses outlined by the SAR image.
The Figure 4(b1) presents an example of internal waves such as observed on the West Atlantic coast of
Morocco. In this case, these waves are generated by a local underwater relief promontory. They manifest
themselves as solitary waves (solitons), the wavelengths of which typically range between 200 m and 2 km
(Figure 4(b2)). Again, the imaging mechanism is caused by the wave-current interactions. Other interesting
SAR signatures are also visible on the quick-look of this ERS-1 SAR image presented on Figure 4(b1) such as
ocean slicks related either to oil spills or natural films.
(a1) (a2)
(b1) (b2)
Figure 4: Typical oceanic signatures such as observed on VV-polarized SAR products.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 8
2.3.3 Ship detection on SAR images
SAR images also give the capacity to monitor the local maritime traffic and/or infrastructures such as oil
exploitation offshore platforms. The type and the level of precision of information delivered for each vessel
depend upon the intrinsic SAR image characteristics (spatial resolution, quality of the image georeferencing,
incidence angle and/or relative position with the satellite route?)). This allows to determine:
• Geographical location of the ships
• ship size in case of high resolution SAR products
• ship route if the trailing wake is visible or if the longitudinal axis of the ships can be determined (with
180° ambiguity in the latter case)
• ship speed if the trailing wake can be observed and if the ship route is not parallel to the satellite
track.
Only ships having enough metallic material to return strong backscatter signals to the SAR instrument can be
detected. The minimum detectable size is a function of the ship length and orientation with respect to the
radar line-of-sight direction, the sensor characteristics and the degree of development of the local sea state.
The most commonly observed wake structure on SAR images (and the most persistent) is the turbulent or
trailing wake stretching out a few km directly behind the vessel. In most cases, turbulent wakes will appear as
a dark narrow line in SAR imagery, because the turbulence generated by the ship’s propeller damps the
surface waves, which translates into a reduction of the backscattered radar power [16]. However, under low-
wind conditions the turbulent wake may appear as brighter than the surrounding sea. The ability to measure
the speed of the detected ships from the observed trailing wake may be very useful in some specific cases to
avoid confusion with en route ship oil discharge. For example, high speed ferries (40 knots) in the
Mediterranean Sea, in summer conditions with thermocline, exhibit clear and long linear wakes with low
backscatter. In these particular cases, the low backscatter results from cold water upwelling caused by the
powerful propellers. Some studies have also reported situations that the presence of wakes on an SAR image
enables the localisation of ships that may not have been detected by detection algorithms.
Another important wake structure is the Kelvin wake, which consists of two arms trailing the ship in the form of
a V-shaped pattern. It is formed by cusp waves that cannot be resolved with satellite SAR standard modes but
contribute to the radar imaging of Kelvin arms. When the cusp waves propagate towards or away from the
radar look direction, the Kelvin arms have the strongest signature. Conversely, Kelvin arms have the weakest
SAR signature when the propagation direction of cusp waves is perpendicular to the radar look direction [9].
This explains why only one of the Kelvin arms is often visible on SAR images. The latter cases may lead to
error in the estimation of the route.
Typical SAR signatures of ship wakes are shown on Figure 5(a) such as captured by the European ERS-1
SAR instrument in 1993. The accumulation of ships detected on a large number of ENVISAT ASAR Wide
Swath products is presented on Figure 5(b). This striking figure clearly demonstrates the potentiality of SAR
instruments to monitor shipping routes with precision such as featured by the traffic separation schemes in the
Channel, off Ushant and off Finisterre as well as the main ferry routes in the North Occidental Mediterranean
Sea [13].
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 9
(a) (b)
Figure 5: (a) Examples of SAR signatures of ships with trailing wakes-note that the ship is moved on
the side because of Doppler Effect. (b) Synoptic view of maritime traffic such as detected on a large
number of ENVISAT ASAR Wide Swath images acquired over European waters.
2.4 Identification of ambiguous slicks
One of the major challenges of SAR-based oil-spill detection exercise is to keep the false-alarm rate (FAR)
below an acceptable level for the user; this level may depend on the specific goal of each user. In order to
address this issue and present the possible methodologies to discriminate slicks on the sea surface such as
detected by the SAR imagery, some examples of typical ambiguous SAR signatures will first be presented in
this paragraph.
SAR signatures of oceanic fronts may, for instance, appear very similar to oil spill signatures and thus lead to
misinterpretation from a not-well experienced operator. In such cases, slick on the sea surface may locally
occur at the separation of two water masses due to wave-current interactions. The latter give indeed rive to
areas of convergence and divergence which can modulate the sea surface roughness. Such an example is
presented on Figure 6. The front area which is outlined on the map of SAR roughness (a) causes low
backscatter signal due to local diminution of the sea surface roughness. Here, the use of sea surface
temperature helps to identify this ambiguous SAR signature as an oceanic front.
Local decreasing of the SAR image intensity leading to possible confusion with presence of oil spills may also
be caused by upwelling phenomena. Such an example is presented on Figure 7 as observed on an
ENVISAT ASAR Wide Swath image product acquired over the western Mediterranean Sea the 14 July 2002.
When such phenomena occur, the amount of surface active substances secreted by marine plants and
animals, which float to the sea surface and create there a surface film, are greatly enhanced.
The Figure 8(a) present examples of natural films such as observed by the SAR instrument on-board the
European ERS-2 satellite the 24 June 2000 in the Black Sea. Such films become visible under low wind
conditions as they accumulate themselves in convergence zones. In such cases, natural films (nutriments,
algae, etc) act as ideal tracers of very complex currents dynamics. Although, the confusion with oil spills is not
possible in the case of Figure 8 because of typical circular shape of eddies and the high density of these
slicks, SAR signatures of natural films can easily be misinterpreted in other situations. The knowledge of the
local wind sea surface may help to assess the probability of occurrence of such events.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 10
(a) (b)
Figure 6: Use of the sea surface temperature (SST) information to identify low backscatter divergence
areas resulting from waves-currents interactions at the interface of two water masses (red curve).
Figure 7: Decreasing of the SAR image intensity due to local up-welling corresponding to a submarine
canyon
(a) (b)
Figure 8: (a) SAR signatures of natural films such as observed in the Black Sea by the European ERS-
2 satellite the 24 June 2000. Eddies become visible under low wind conditions because of the
accumulation of natural films at the sea surface. These eddies are confirmed by the SST image
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 11
(a) (b)
Figure 9: (a) SAR signature of estuary plume such as observed on ENVISAT ASAR Wide Swath image
the 10 January 2004. (b) MODIS observation.
Suspended matters in estuary plumes may also contribute to locally decrease the sea surface roughness. As
a result, SAR signatures of river outputs may potentially look like typical oil spills signature. For this reason,
the use of ocean colour (visible data) can bring complementary useful information to solve ambiguous
situations
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 12
3 HOW TO ANALYSE SAR IMAGE AND HOW RELIABLE IS THE IDENTIFICATION OF OIL
SPILLS?
3.1 Context analysis and objectives
In order to be able to define what is the best methodology to analyse SAR images, one need first to specify
the requirements and objectives to be achieved according to the user’s monitoring goals and the context in
which the SAR oil spill detection is to be done. In the introduction, a distinction was made between accidental
oil disaster and oil pollution due to ship intentional operations the former obviously require different
monitoring options in comparison with the usual routine monitoring of polluting vessel.
In this section, the priorities and resulting decision criteria will be presented as highlighted by the context of
the SAR image analysis.
The issue of selecting the most appropriate SAR image mode products as well as the most appropriate
processing by the operator will be addressed in the next part devoted to operational use of SAR imagery
3.1.1 Monitoring accidental disasters
Such situations are essentially exceptional and unpredictable. They require rapid response to picture at best
the total extent of oil spilled during the accident. As such, SAR data programming, ordering and acquisition
must be done in the shortest delay. To achieve this goal, authorised users have the possibility to activate the
international Charter “Space and Major Disasters”1. The latter aims at providing a unified system of space
data acquisition and delivery to countries that are affected by natural or man-made disasters.
The requirement then is not to minimise the false positive rate in the SAR oil spill detection scheme but to
maximise the detection probability and provide exhaustive and detailed monitoring of sea areas that
are potentially affected by oil pollution.
Since the area of interest is known, the use of high resolution SAR products is possible and will be preferred
to optimise the capacity to detect small slicks.
3.1.2 Routine surveillance of oil pollution due to ship intentional discharges
Here the objective is to perform recurrent and systematic analysis of SAR images acquired over a large area
to routinely monitor maritime circulation. Recent or on-going oil discharge due to passing vessel should be
reported in the shortest delay.
If the oil discharge is on-going at the acquisition time (i.e. a slick signature clearly appears in the wake of a
ship), there is a possibility to target an aircraft to confirm the pollution and identify and catch the polluter red
handed. This implies of course very short delay between SAR data acquisition by the satellite over the waters
of interest and the delivery of the oil spill detection report to the authority in charge (typically less than 1 hour),
as well as the aircraft availability.
In practice, this ideal scenario is not much realistic because of the low probability that the on-going discharge
will last until the arrival of the aircraft, even if the timeliness of the whole chain is achieved. However, it may be
possible sometimes to associate oil pollution with a ship (or platform). Ship detection must then be
systematically performed on the SAR image that is used for oil spill detection. As a result, this may provide
precise geographical position of the ships presents within the area covered by the SAR image. This
information can be further coupled with AIS ships reports to obtain complementary information on the ships
characteristics (name, type, destination, etc).
In addition to the timeliness requirement, the false positive alarm rate must also be kept as low as
possible to keep the cost of aerial surveillance at an acceptable level. It is essential to demonstrate the
1
The Charter came into force on November 1, 2000 and involves Canada, France, Japan, EC, NOAA....?.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 13
benefit of using satellite SAR imagery as a complementary early-warning observing system complementing
and increasing the efficiency of the existing surveillance means.
3.1.3 Statistical analysis of oil pollution by means of SAR imagery
Although such an analysis does not enter into the operational category, it allows the assessment of the
situation (occurrence of en route ship oil discharges per year) over determined areas according to the effort of
surveillance and law enforcement strategy. Thus it is important to remind the main considerations that should
prevail when carryout out such a study.
First of all, statistical analysis means that oil spill detection is performed on a wide dataset of SAR images
acquired over a given area and time period. In order to achieve meaningful statistics, the oil spill detection
criterion must be clear and constant during the whole study period. Thus, in order to avoid confusion
and not to bias the results, one should mostly focus on typical SAR oil spill signatures that are only slightly
distorted by wind and currents, that is to say relatively recent pollution due to passing vessels.
In addition, statistics should take into account the sea area of each SAR image that is indeed exploitable for
oil spill detection and not the total area of each SAR image. As previously, mentioned, the oil spill detection
using SAR imagery in C-band is known to be reliable when the local sea surface 10 meter wind ranges
between 5 and 25 knots. This should improve the consistency of the detection results and enable
comparisons of ship discharges statistics over targeted areas.
3.2 Methodology
3.2.1 Analysis of the slicks candidates
Although several methods have been developed to allow automatic detection of oil polluted areas within low
backscatter regions in the SAR images, none of these have revealed to be able to solve all ambiguous
situations and avoid positive false alarm with a sufficiently high degree of confidence. Hence, it is today
generally agreed that oil spill detection service should be based on semi-automatic detection process
complemented by the expertise of a trained operator.
Once a slick candidate is detected by the dedicated tool and presented to the operator for visual
discrimination and classification, the two main questions to be answered are the following: What is the shape
and contour of the slick signature? What is the probability that this slick is caused by oil discharges or by
natural phenomenon given the general context (meteo-oceanic conditions and proximity of a ship or a
maritime route)?
At first, the analysis of the shape of the slick signature allows to derive preliminary information on the type
and/or the age of the slick. On-going or very recent oil pollution usually appears as a highly linear thin dark
pattern stretching out directly behind the vessel. Often, the slick can have one sharp edge and one more
diffuse due to wind-driven drift.
In course of time, the shape of oil spills changes with the local wind and current dynamics (Figure 10)
while wind and waves will influence the dispersion and thus the contrast of the SAR signature. Hence, the
knowledge of the past local wind and current conditions is likely to bring valuable information in an attempt to
fill the gap between the spill event occurrence and the SAR observation. As an example, Espedal and Wahl
[5] demonstrated that precise knowledge of local wind-vector information may potentially help to describe the
wind impact of the shape of the detected spill (together with the underlying surface current). In addition wind
force evolution will influence the possibility to detect the slick or not
Indeed the analysis of SAR images for oil spill detection will strongly benefit from the precise knowledge of the
current and conditions and the on-going human activities in the surroundings. Typically, the detection of a
vessel in the prolongation of a suspicious slick will highly influence the final decision of the operator.
Alternatively, the probability to classify suspicious slicks as oil spill will increase in the proximity of well known
maritime routes.
After all, the quality level of oil spill detection service is also increased owing to the experience accumulated
by the operator on the SAR instrument itself but also on the local oceanic/atmospheric climatology as well
as human and economic activities on the area at stake.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 14
Figure 10: (a) On-going oil discharge such as captured by ENVISAT ASAR Wide Swath image acquired
the 16 September 2003 at 20:03 UTC. (b) Same oil spill observed by RADARSAT-1 on a ScanSAR
Narrow product the 17 September 2003 at 16:13 UTC
3.2.2 How can the use of ancillary data improve the detection?
In section 2.3, the important SAR capacity to observe various oceanic and atmospheric phenomena has been
introduced. Sometimes however, SAR imagery does not suffice in practice to distinguish between man-made
oil spills and ambiguous slicks caused by low-wind conditions, up-welling effects, current fronts, and natural
films. The use of ancillary valuable information helps to reduce the False alarm rate (FAR) in the case of
ambiguous slicks [8].
For instance, local wind estimation provides good confidence level of the detection scheme outputs. Such
an estimation can be directly inferred from SAR imagery with a spatial resolution as high as 500 m, for it
provides a reliable indication of wind force and direction, such as those given by numerical wind-model
outputs or scatterometer measurements. This estimation helps to assess the reliability of the detection.
Beyond the detection process, having simultaneous and precise information on the local wind can be very
helpful to initiate correct drift predictions (precise position of atmospheric fronts, presence of air/sea breeze or
katabatic winds, etc). This has revealed to be particularly useful over the occidental Mediterranean Sea where
the wind fronts and the local wind extension are not precisely known.
Sea surface temperature (SST) can be very useful to identify oceanic fronts characterised by local
wave/current interaction that locally modulates the sea surface roughness.
Ocean colour has also shown its capacity to identify plumes from estuaries and coastal rivers [8].
One should be aware that level 2 SST or Ocean colour products are limited by cloudy situations. However
recent developments allow providing global map of relevant information (SST or ocean colour) merging
several sensors, having different spatial resolution, sensitivity or time sampling. As an example, the
Medspiration project now provide L4 SST products using optimal interpolation methods over the
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 15
Mediterranean Sea with a 2 km x 2 km spatial resolution. Alternatively, the GLOBECOLOR project will provide
similar ocean colour high level products.
The use of precise bathymetry may also locally help to identify ambiguous SAR signatures due to local up-
welling effects along the continental shelf.
The knowledge of the main maritime traffic routes is also a critical issue in a semi-automatic detection
scheme validated by a trained operator. The latter can then focus the analysis to these sensitive areas in
priority.
3.3 Classification of the detected slicks
The capacity of SAR images to detect oil spills has been introduced as well as potential risks of
misinterpretation leading to false alarms. Based on these considerations, a methodology has been proposed
to analyse SAR images. However, the classification problem (that is the ensemble of possible decision or
classes) has not been discussed so far. Today, classification is usually limited to 2 (high, low)2 or 3 classes
(high, medium, low).
However, the quantitative characterisation of detected possible oil slicks in terms of probability of being oil is
almost no longer used for in fact it was not of much help for the end-users to take appropriate decisions and
was much dependant on the operator own experience.
2
In the case of the oil spill demonstration that was conducted over the French ZPE in 2004 [13], classification was limited
to 2 outputs classes: oil (no ambiguity) and feature of interest (SAR signature of slicks that a trained operator could neither
discard nor associate for sure to oil spills).
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 16
4 OPERATIONAL USE OF SAR IMAGES
4.1 User requirements
Though near-real-time coverage over the whole globe is clearly not possible with the existing satellites and
receiving stations, it is however possible to do it in most European seas and then it is essential to reduce, the
processing time delay in order to deliver information on detected possible spills as quickly as possible, at least
with a delay not exceeding 60 min. Such a delay will preserve some chance to visually identify the source of
pollution. However the exact position of the potential culprit is well determined and it can be combined with the
Automatic Identification System (AIS) data. This represents an important step towards culprit identification and
would certainly have an important dissuasive impact for it could be used for further action according to the
neighbouring country law.
The minimisation of the false alarm rate and the maximisation of detection probability also stand as important
user requirements. Notwithstanding that these two objectives are not compatible with regard to the detection
theory; the present situation could be significantly improved. Indeed, efficient use of ancillary information (such
as sea surface wind, sea surface temperature, bathymetry) make it possible to improve the detection process:
semi-automatic oil discharges detection validated by trained operator [13]. This would at least help to reduce
the rate of positive false alarms (i.e. situations where low SAR backscatter areas are misinterpreted as oil spill
signatures).
4.2 Oil Spill detection portfolio
The basic-service deliverables include a map of detected oil slicks together with additional information on the
suspicious areas, such as
• Geographical position (latitude, longitude)
• UTC time
• Geometric parameters such as length (for recent linear oil spills) or area, perimeter [17] (in case of
accidental disaster.
In addition, other useful complementary information should also be provided, such as previous observations
from various validated sources
• Potential source identification (ship, platform)
• Description of in situ meteo-oceanic conditions either directly derived from SAR imagery (high-
resolution wind fields) or inferred from ancillary satellite data (e.g., ocean colour, sea-surface
temperature).
4.3 Description of the Oil spill detection scheme
4.3.1 On the choice of appropriate SAR image modes
The selection of appropriate SAR image mode product must be driven by the general context in which the oil
spill detection is performed on the SAR image.
In case of accidental oil spill disaster, one would prefer to provide exhaustive reporting of SAR signatures of
sea surface slicks that may be associated with oil. In such situations the higher spatial resolution could be
preferred in order to more precisely report oil slicks. Extended spatial coverage is not necessarily required as
the location of the accident and the extension of the slicks are supposed to be more or less known in advance.
Hence wide swath mode products could rather be disregarded and high resolution products could be preferred
such as:
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 17
• ENVISAT ASAR image products (VV or HH polarisation)
o Image Mode Precision product (IMP). This 4-looks ground range detected product has a
pixel size of 12.5 meters. The Swath length is approximately 100 km in the along track
direction. The swath width which depends on the selected swath number (i.e. incidence
angle, IS1-7) ranges between 55 and 105 km. Also available as Single Look Complex (SLC)
products.3
• RADARSAT-1 SAR image products (HH polarisation)
o Standard beam mode (S1-7). This 4-looks ground range detected product has a pixel size of
12.5 meters. The swath coverage is about 100 x 100 km. Also available as SLC product.
o Wide beam mode (W1-3). This 4-looks ground range detected product has a pixel size of
12.5 meters. The swath coverage is about 150 x 150 km. Also available as SLC product.
o Fine beam mode (F1-5*). This Single look ground range detected product has a pixel size of
6.25 meters. The swath coverage is about 50 x 50 km. Also available as SLC product.
In case of routine surveillance of oil discharges resulting from en route ship discharges, the objective to
be achieved is to detect oil discharges on a wide coverage. For this reason, Wide Swath coverage products
will be preferred to the prejudice of the spatial resolution. It has been largely demonstrated however that such
products enable to correctly detect illicit oil discharge. Although it is not possible to measure the width extent
of such oil spills due to coarse pixel size (ranging between 25 and 75 meters), it is however possible to
measure the length. The available products are:
• ENVISAT ASAR Wide Swath products (VV or HH polarisation). This ground range detected product
has pixel size of 75 meters. The swath coverage is approximately 400 km x 400 km.
• RADARSAT-1 products.
o ScanSAR Narrow beam mode (SNA-B). This ground range detected product has a pixel
size of 25 meters. The swath coverage is about 300 x 300 km.
o ScanSAR Wide beam mode (SWA-B). This ground range detected product has a pixel size
of 50 meters. The swath coverage is about 500 x 500 km.
4.3.2 Methodology
So many approaches have been proposed in the literature to support automatic detection of SAR signatures
of ocean surface slicks (region/contour approach, wavelets, multi-scale, fractal, etc) that it would be vain to
present these methods exhaustively in this document. Among these methods that generally considered SAR
images as the unique data to be analysed, none them have in fact succeeded in providing reliable
classification of SAR slick candidates. The optimal Bayesian classification approach proposed by Solberg et
al. is probably one of the most interesting and efficient method.
It a matter of fact that our understanding of SAR imaging mechanisms of the various physical processes that
occur on the sea surface have significantly improved with the increasing number of SAR missions and the
synergetic use of different sensors. It is also true that observation capacities of several sea surface
parameters keep increasing: wind (scatterometry, altimetry, radiometry, SAR) , sea surface height (altimetry),
sea surface temperature (radiometry, IR) , ocean colour (visible) and soon salinity. As a result, more and more
remote sensing products become available. Level 2 swath information products which are sometimes limited
by clouds and illumination conditions are now efficiently merged together (and possibly combined to model
outputs) to bring the most sophisticated higher level synthetic products with increasing spatial resolution and
time sampling. Thus, it is likely that data fusion methods will be more seriously considered in the near future.
Meanwhile, it is agreed that only semi-automatic SAR oil spill detection process validated by a well trained
operator can reasonably meet user requirements.
SAR images should be scanned sequentially tile by tile in full resolution to detect and delineate dark patches
on the SAR image intensity that are associated with physical slicks on the sea surface.
3
This SAR product is more complex than the IMP product (slant range projection, phase information, etc). This product
has extended capacities with regard to IMP products because of the phase information (swell field retrieval field, sea
surface displacement field from SAR Doppler centroïd analysis).
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 18
Perhaps one of the most important things at this level, although nearly never addressed in practice, is the
correction for the well known attenuation of the intensity in the line-of-sight direction with increasing
range (i.e. when the incidence angle increases). This attenuation, well predicted by the radar equation and the
scattering theory, is all but negligible on wide swath coverage products that are used for routine surveillance
of voluntary oil discharges.
Delineation of slicks can be either supported by automatic segmentation process or simply manually
performed by the operator. Each slick candidate will then be visually inspected in light of complementary
information delivered by ancillary data.
At the end of the process an oil spill detection report is produced summarising the detection results, providing
details on the slicks of interests (location, time, size, source, etc), including general information on the current
sea state (wind and other oceanic/atmospheric phenomena) and human activities (ships detection).
Notification of the oil report output must eventually be sent to the end-user using appropriate dissemination
methods (email, SMS, fax, etc)
4.3.3 Where to collect ancillary information
A large number of data sources are now available to provide complementary information to support SAR
images analysis. They can be of crucial interest to solve ambiguous situations where the sole use of the SAR
roughness information would not have been sufficient. As such, ancillary data truly contribute to reduce the
false positive alarm rate.
The type and characteristics of these ancillary data are presented in Table 1.
Time
Type Source Pixel size Location
sampling
NASA Twice a 25 km http://manati.orbit.nesdis.noaa.gov/quikscat/
Scatterometry
QUIKSCAT day 12.5 km http://manati.orbit.nesdis.noaa.gov/hires/
Sea Surface
MEDSPIRATION daily 2km ftp://ftp.ifremer.fr/ifremer/medspiration/data/l4uhrsstfnd/eurd
temperature
300m
Once a (systematic)
ENVISAT MERIS http://ewfs.eo.esa.int/
day 1200 m (on
Visible
request)
TERRA/AQUA 2x once a
250 m http://rapidfire.sci.gsfc.nasa.gov/realtime/
MODIS day
Table 1: Type and characteristics of some ancillary information to support oil spill detection on SAR
imagery
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 19
REFERENCE
[1]
[2] W. Alpers and H. Hühnerfuss, “Radar signatures of oil films floating on the sea surface and the
Marangoni effect,” J. Geophys. Res., vol. 93, no. C4, pp. 3642–3648, Apr. 1988
[3] W. Alpers and H. Hühnerfuss, “The damping of ocean waves by surface films: A new look at an old
problem,” J. Geophys. Res., vol. 94, no. C5, pp. 6251–6265, May 1989
[4] C. Brown, M. Fingas, and R. Hawkins, “Synthetic aperture radar sensors: Viable for marine oil spill
response,” in Proc. 26th Arctic and Marine Oil Spill Program (AMOP) Tech. Seminar, Victoria, Canada,
Jun.10–12, 2003.
[5] H. Espedal and T. Wahl, “Satellite SAR oil spill detection using wind history information,” Int. J.
Remote Sens., vol. 20, no. 1, pp. 49–65, Jan. 1999.
[6] M. Gade,W. Alpers, H. Hühnerfuss, V.Wismann, and P. Lange, “On the reduction of the radar
backscatter by oceanic surface films: Scatterometer measurements and their theoretical interpretation,”
Int. J. Remote Sens., vol. 66, no. 1, pp. 52–70, Oct. 1998.
[7] Gerling T., Structure of the surface wind field from Seasat SAR, J. Geophys. Res., vol. 91, pp. 2308-
2320, 1985
[8] F. Girard-Ardhuin, F. Collard, G. Mercier, and R. Garello, “Oil slick classification by SAR imagery
using synergetic data,” in US Baltic Int. Symp., Klaipëda, Lithuania, 2004
[9] I. Hennings, R. Romeiser, W. Alpers, and A. Viola, “Radar imaging of Kelvin arms of ship wakes,” Int.
J. Remote Sens., vol. 20, no. 13, pp. 2519–2543, Sep. 1999.
[10] H. Hühnerfuss, W. Alpers, O. Fast, P. Lange, A. Loffet, K. Richter, R. Scriel, N. Skou, and F. Witte,
“The Discrimination Between Crude Oil Spills and Monomolecular Sea Slicks by Airborne Sensors,”,
1987.
[11] H. Hühnerfuss, A. Gericke, W. Alpers, R. Theis, V. Wismann, and P. Lange, “Classification of sea
slicks by multifrequency radar techniques: New chemical insights and their geophysical implications,” J.
Geophys. Res., vol. 99, no. C5, pp. 9835–9845, May 1994.
[12] V. Kerbaol and F. Collard, “SAR-Derived Coastal and Marine Applications: From Research to
Operational Products”, IEEE J. Oceanic Engin., vol. 30, no 3, pp. 472-486, july 2005
[13] V. Kerbaol, F. Collard, P. Leilde, and F. Parthiot, Improved oil spill detection service over the French
ZPE: developments and results, SEASAR 2006, ESA/ESRIN, Frascati, 23-26 Jan. 2006
[14] J. Lichtenegger, “ERS-1 SAR images for oil spill surveillance,” Earth Obs. Q., vol. 44, no. 3, pp. 7–10,
Jun. 1994.
[15] On the Monitoring of Illicit Oil Discharges — A Reconnaissance Study in the Mediterranean Sea, P.
Pavlakis, D. Tarchi, and A. Sieber. (2001).
http://europa.eu.int/comm/environnement/civil/marin/reports_publications/jrc_illicit_study.pdf [Online]
[16] R. Peltzer,W. Garrett, and P. Smith, “A remote sensing study of a surface ship wake,” Int. J. Remote
Sens., vol. 8, no. 5, pp. 689–704, 1987
[17] A. H. S. Solberg, G. Storvik, R. Solberg, and E. Volden, “Automatic detection of oil spills in ERS SAR
images,” IEEE Trans. Geosci. Remote Sens. E, vol. 37, no. 4, pp. 1916–1924, Jul. 1999
[18] Wackerman C., W.G. Pichel, P. Clement-Colón, A Projection Method for Automatic Estimation of
Wind Vectors with RADARSAT SAR Imagery , ESA SP-Series, SP-565, pp. 55-60, 2004
[19] V. Wismann, R. Theis, W. Alpers, and H. Hühnerfuss, “The damping of short gravity capillary waves
by experimental sea slicks measured by a multifrequency microwave scatterometer,” in Proc. OCEAN,
vol. II, Victoria, Canada, 1993, pp. 342–347.
Use and analyse of satellite SAR images for oil spills detection --- Cedre - July 2007 20
Related docs
