You are on page 1of 20

OIL SPILL ALONG THE TURKISH STRAITS

SEA AREA; ACCIDENTS, ENVIRONMENTAL


POLLUTION, SOCIO-ECONOMIC IMPACTS
AND PROTECTION

Edited by
Selma ÜNLÜ – İstanbul University
Bedri ALPAR – İstanbul University
Bayram ÖZTÜRK – İstanbul University

Publication No: 47

İstanbul 2018
Ünlü, S., Alpar, B., Öztürk, B. (Eds) (2018). Oil Spill along the Turkish Straits Sea Area; Accidents, Environmental Pollution,
Socio-Economic Impacts and Protection. Turkish Marine Research Foundation (TUDAV), Publication No: 47 İstanbul, Turkey.

OIL SPILL DETECTION USING REMOTE SENSING TECHNOLOGIES-


SYNTHETIC APERTURE RADAR (SAR)

İbrahim PAPILA 1*, Elif SERTEL 1,2, Şinasi KAYA 1,2 and Cem GAZIOĞLU 3
1
Istanbul Technical University, Research and Application Centre for Satellite
Communications and Remote Sensing (CSCRS), Sariyer, Istanbul, Turkey
2
Istanbul Technical University, Department of Geomatics Engineering, Sariyer,
Istanbul, Turkey
3
Istanbul University, Institute of Marine Sciences and Management, Department of
Marine Environment, BERKARDA Remote Sensing and GIS Laboratory, Fatih,
Istanbul, Turkey
* papila@cscrs.itu.edu.tr

1. Introduction
Marine environment and ecosystems have been threatened by oil spills occurring
on the sea surfaces which are mostly result of either accidents or cautious oil discharges
from the ships. Integration of satellite based data and in situ observations are widely used
to monitor and detect oil spills (Solberg, 2012). Visible and infrared-based remote sensing
observations are significantly affected by atmospheric conditions; moreover, there are
also spectral similarities between oil and water in specific parts of the visible region of
the electromagnetic spectrum. Although oil presents slightly larger reflectance than water
in the visible and near infrared region; it does not exhibit characteristics absorption or
reflectance features; such as, thin oil layers seem in similar colour. These limitations
should be overcome by using different remote sensing systems to accurately detect and
map the oil spills (Fingas and Brown, 2018; Gazioğlu, 2018; Leifer et al., 2012). Although
hyperspectral data and analyses are thought to be promising in determining oil pollution,
processes that examine specific spectral regions are known to not increase discrimination.
Laser fluorosensors are another alternative remote sensing technology to detect oils
considering the fact that after the interaction of oil compounds with ultraviolet light; some
light energy is absorbed and extra energy is released as visible light (Brown, 2017).
However, in addition to being used as remote sensing technology, it is necessary to define
numerous features related to the marine environment. The fact that optical systems and
laser fluorosensors, under any circumstances, fail to perform in a suitable, acceptable and
efficient manner, allows the Synthetic Aperture Radar (SAR) systems to be preferred to
provide high standards under all conditions (Doğan et al., 1998). Active remote sensing
systems such as SAR have been commonly used to detect oil spill with their advantages
of covering wide areas, operating in all-weather conditions either day or night (Şeker et
al., 2013). SAR images could complement aircraft observations to cover larger areas in a
short time and cost-effectively. Slicks on the sea surface are effective in dampening
capillary waves (cm wavelength) therefore having smoother surface conditions in contrast
to approximate sea surface conditions. In order to prevent oil spills and cope with oil spill
based disasters, it is important to have an oil spill surveillance system. Location and extent
of oil spills could be determined promptly and accurately using remote sensing
technologies which in turn provide valuable information to monitor oil spills and
minimize their damages to ocean ecosystem.

140
Both oil spills and slick impostors (e.g., variable surface wind stress, rain cells,
turbulent wakes caused by ships, fresh water slicks) might be detected as a dark formation
on the SAR images (Fingas and Brown, 2014). In most cases it is impossible to
differentiate between oil slicks and look-alike phenomena just using SAR intensity
values. Several researches have been focused on distinguishing the look-alikes from the
oil spill (Genovez et al., 2017). Oil spills mixed with look-alikes in classification
procedure while using oil spill detection algorithms are categorized as below (Solberg
and Solberg, 1996):

• Thin, piecewise-linear slicks


• Low-contrast slicks in homogeneous sea
• Slicks on a very heterogeneous background

In general, sea state and/or wind conditions are important parameters impacting
the ability of SAR to detect oil spills. Moderate wind speed required to cause surface
roughness variation between sea surface and slick. Typically, 3 m s-1 to 10m s-1 wind
speeds are the most accepted limits. The acceptable range of wind speed depends on oil
type and the age but more over it is strongly related with the incidence angle and radar
frequencies (Bern et al., 1992). X-band SAR data is provided better results and preferred
over C and L-band for the detection of oil spills. The contrast amount between oil and
water is changing with respect to the wavelength of the SAR system. The biggest contrast
is obtained in X-Band, moderate in C-Band and the smallest in L-Band. However, a recent
research illustrated that low-noise L-Band system could be able to detect oil slicks
successfully. Incidence angles ranging between 20 ˚ and 45˚ for X or C-Band systems
would provide an optimum measurement configuration for oil spill cases (Marzialetti and
Laneve, 2016). In calm and light breeze conditions in which wind speed is smaller than
3m/s and wave height is smaller than 0.3 m, it is not possible to differentiate oil from
water. Ideal conditions to identify oil spill is known as gentle strong breeze conditions, in
which 3m/s<wind-speed<10m/s and 0.3m<wave height<3 m. It is not also possible to
identify oil spills in storm conditions with very huge wave speeds.

Different wave polarizations exist in SAR systems based on vertical (V) and
horizontal (H) electromagnetic wave propagations. Some studies had suggested VV
polarization data could provide the best results for the detection of oil spills (Masuko et
al., 1995 and Girard-Ardhuin et al., 2003). Incidence angle is significantly impacting the
HH polarized data compared to VV polarization. Therefore, VV is preferable in most oil
spill detection case whereas HH could be appropriate in very light wind conditions and
for the accurate detection of ships. (Lehrer 2014, Migliaccio et al., 2015). Some of the
available SAR satellite systems are shown on Table 1.

141
Table 1. SAR Satellite Systems
Satellite Operation Band Polarisation Swath Resolution
(km) (m)
TerraSAR-X 2007 - X Dual-pol 10-100 1 -40
COSMO 2007 - X Quad-pol 10-200 1 -100
Skymed 1-2
RADARSAT- 2007 - C Quad-pol 25-500 1 – 100
2
RISAT-1-2 2012 - C Quad-pol 10-223 1 – 50
KOMPSAT-6 2013 - L Quad-pol 5-100 0.5 - 20
ALOS-2 2014 - L Compact 25-350 1 - 100
SENTINEL 2014 - C Dual-pol 20-400 5 - 80
1A-1B
PAZ 2015 - X Dual-pol 10-100 1 - 15
NovaSAR-S 2015 - S Tri-pol 15-750 6 - 30
SAOCOM 2015 - L Quad-pol 20-350 10 - 100
Radarsat 2018? C Compact 14-500 1 - 50
Constellation

2. Radar Image Processing


Several researches have been conducted to analyse different image processing
algorithms to improve the accuracy of supervised and semi-supervised techniques for
object detection and classification. Generally processing steps include speckle removal
Tripathi et al. (2017), edge detection Zhang et al. (2010), texture analysis Zhou et al.
(2009) - feature extraction Schvartzman et al. (2016), Mera et al. (2017) and finally
classification algorithms Vijayakumar et al. (2016) and Taravat et al. (2013).
a. Speckle Removal
Radar images are subject to receive strong speckle noise as a result of coherent
processing of radar echoes. The presence of speckle noise impacts the quality of the
spatial details and makes the classification step more difficult. Thus an effective speckle
removal filtering is required as a pre-processing step before applying any other image
processing method. The objective of speckle removal filter is simple:
• To suppress the multiplicative noise in homogenous regions
• To preserve the edges, image details and the texture features
Many of the speckle removal filters (Lee, Kuan and Frost) which are adaptive
filters operate by smoothing over a fixed window to minimize the mean square error based
on the statistical characteristics of SAR image and the speckle. Studies show that the
statistical filters can reduce the speckle better but the detail of the image is greatly lost.
Some hybrid methods are also suggested which combines Frost speckle filter with
Relaxed Median Filter (Radhika and Padmayathy, 2011).
Most commonly used filters in literature are briefly mentioned below (Huang
and Genderen, 1996) and (Argenti et al., 2013):

142
Box Filter: Typical low-pass filter, which removes the noise with high frequency
spectrum as well as smooth the details (Fuk-kwok and Held, 1983).
Median Filter: Non-linear filter derived from maximum likelihood estimation
principle; assuming that the signal is contaminated by additive noise with Laplace
distribution (Richards 1993).
Lee Filter: It is a local-statistic filter. The goal is minimizing the minimum mean
square error (MMSE) by linearizing the multiplicative noise using the local statistics.
Depending on the variance of the noise, the smoothing operation is applied if the variance
is low but it is not suitable for high variance. This filter preserves details in low and in
high contrast at the same time (Lee 1980, 1983).
Frost Filter: Frost filter parameters use local area parameters that depends on
the local coefficient of variation defined as the ratio of local standard deviation to the
local mean of the noisy image. It convolves the pixel values to replace the central pixel
with the values of weighted sum of neighbourhood. (Frost et al., 1982)
Kuan Filter: It implements the Linear MMSE filter solution starting with a
nonstationary scene where the signal has nonstationary mean and variance. Unlikely with
Lee there is no approximation involved (Kuan et al., 1985, 1987).
Enhanced Frost and Enhanced Lee: In order to preserve the texture information,
Lopes et al. (1990b) applied the enhancement. If the local coefficient of variation is below
a lower threshold, the pixel value is calculated by taking the average of the filter window.
If it is above the higher threshold then the pixel value keeps the same. In the case of
having, the local variance between two thresholds then impulse response is used as a
convolutional kernel to determine the pixel value.
Lee-sigma filter: This filter assumed that data has Gaussian distribution. The
pixels within the standard deviation range are replaced with the average of all pixels in
the moving window (Lee, 1983).
Gamma Map Filter: In order to minimize the loss of texture information
Maximum A Posteriori (MAP) filter is proposed by Lopes et al. (1990a) where the
Gamma distribution is used to estimate the speckle noise characteristics. The original
value is also assumed to lie between the local average of the window and its actual value.
Choosing the optimum speckle filter size has also huge impact on speckle
reduction. As the window size of speckle filter increases speckle in the image reduces,
but as window size increases details in the image might be lost (Argenti et al., 2013). Best
filter window size is considered in Dasari and Anjaneyulu (2017) based on three indices:
i) Speckle suppression index ii) Equivalent Number of Looks, and iii) Speckle
Suppression and Mean Preservation Index and all of these indices are derived from by
using mean standard deviation and variance of the image. In Samad and Mansor (2002)
it was showed that Lee filter with size of 7x7 is the best filter to detect dark slick in the
Radarsat-1 Standard 2 Beam mode with HH polarization image while removing the
speckle it preserves edge and the texture of the dark slick very well. The same study
showed that using the Kuan filter size of 11x11 is the appropriate filter to improve ships
detection in Radarsat images.

143
b. Edge Detection
Generally, this step is applied to differentiate the dark spots from the
neighbouring seawater. Edge detection is used to outline the edges between water and oil.
Different segmentation methods have been used to identify edges of the oil slicks. Markov
random field (MRF) based segmentation Lankoande et al. (2005) and Multiscale MRF
segmentation Jiao and Wen (2009) based on the statistical properties offer a good
performance on image segmentation. Local and global thresholding algorithms are used
in Solberg et al. 2007, Garima et al. (2015) to detect the dark spots. Dutta et al. (2018)
proposed amalgamate approach to combine over-segmented regions to form meaningful
dark objects. Some other techniques such as Adaptive Thresholding, Fuzzy clustering,
Region Growing, Multiscale Segmentation and Mathematical Morphology can also be
used in this step.
Auto Associative Neural Networks (AANN) can be used to extract polarimetric
and textural features from SAR images. It provides the optimal input for the segmentation
by increasing the contrast between the possible dark object and the background (Frate et
al. 2017).
The segmentation step using Artificial Neural Network (ANN) outperformed
edge detection segmentation and adaptive thresholding approaches (Singha et al., 2013).
It is also indicated that using separate neural networks for both the segmentation and
classification stages perform better accuracy over single input neural networks (Singha et
al., 2013).
c. Feature Extraction
Surface characteristics are investigated in feature extraction step using different
approaches. Such features like geometric shape, contrast with surrounding areas and
contextual information are used to characterize the dark spots. To distinguish the oil slicks
from look-alikes those features have to be extracted and used as inputs. Oil has a uniform
and smooth texture whereas surrounding sea surface is a less uniform a smooth in terms
of texture. Texture information provides useful input to separate Look-alikes from oil
spills. Some research have been conducted on shape based analysis which is suggested as
useful approach to separate oil spills from look-a-likes due to the unique shape
characteristics of oil spills. Most commonly used methods in feature extraction are
distributed in four categories (Genovez et al., 2017):

• Statistical: Backscattering parameters


• Textural: Grey Level Co-occurance matrix (GLCM)
• Geometrical: Geometry of the polygons
• Polarimetric: Polarization mode
In order to detect the oil slicks, different number and types of features are used
among the researchers : Solberg and Theophilopoulos (1997) used 11 features, Fiscella
et al. (2000) used 14, Keramitsoglou et al. (2006) used 14, recently Chehresa et al. (2016)
used 74, while Mera et al. (2017) used 52 features. Most commonly used features from
the majority of those studies are presented on Table 2 (Genovez et al., 2017).

144
Table 2. Commonly used features.
Statistical Textural Geometrical
Mean Second Moment Area Rectangular Fit
Std deviation Contrast Length Compactness
Median Correlation Width Degree of
Skeleton
Maximum Dissimilarity Border Length
Branching
Minimum Entropy Asymmetry Perimeter
(Polygon)
Asymmetry Homogeneity (GLCM) Density Shape Factor
Kurtosis Mean (GLCM) Shape index Complexity
Mode Std deviation (GLCM) Elliptic Fit

The ranking of the 10 most important features are presented in Genovez et al.
(2017) as: 1. Median, 2. Mean, 3. Mode, 4. Minimum, 5.Maximum, 6. Mean (GLCM), 7.
Rectangular Fit, 8. Homogeneity (GLCM), 9. Asymmetry (Polygon) and 10. Density.
d. Classification
The final stage of oil spill detection is the classification in which oil slicks are
identified using the features extracted in the previous step. Several classifiers have been
applied for the detection of oil spills such as Mahalanobis Distance Classifier, Maximum
Likelihood, Fuzzy Logic, Generalized Additive Model, Support Vector Machine,
Artificial Neural Network, and Tree-based techniques (Xu et al., 2014):
Mahalanobis Classification: This algorithm is a direction sensitive and it
assumes all class covariance are equal. Based on the distance threshold all the pixels are
classified to the closest region. Different patterns can be distinguished by using the
correlation between variables (Topouzelis et al., 2009).
Maximum Likelihood: It is the most commonly used algorithm for pixel-level
processing, however segmentation result can also be used with a region-based approach.
It requires a training data set to maximize the likelihood function by choosing the model
parameters (Ahmad and Quegan, 2012).
Fuzzy Logic: This method estimates the contribution of each class in the pixel.
All pixels are belonging to a class with a membership degree that uses non-linear
techniques. In order to predict the dark areas as oil, a model has to be created with the
inputs that give known outcomes (Garcia-Pineda et al., 2008; Robson et al., 2006).
Penalized Linear Discriminant Analysis: Penalized version of LDA is proposed
to solve high dimensional discriminant problem, since it maintains the advantage of LDA
while adding shrinkage to the discriminant vectors. It is less time-consuming comparing
to others. Penalized LDA can simultaneously select predictors and estimate the change of

145
mean classification error rate. By using the posterior probabilities of different classes, the
model can predict the class membership (Fiscella et al., 2000).
Generalized Adaptive Model: GAM is an extension of generalized linear models.
It provides a modeling approach that combines powerful statistical learning with
interpretability, smooth functions, and flexibility. It is used for capturing nonlinear,
unspecified relationships between original predictor variables and the response variable
(Hastie, 2006).
Support Vector Machine: The SVM technique is supervised learning model that
seeks the optimal separating hyperplane between binary classes. The goal is to find the
support vectors, by using minimum number of training samples. It is used effectively in
geomorphological modelling (Moguerza and Muñoz, 2006; Mountrakis et al., 2011;
Brekke and Solberg, 2008).
Artificial Neural Networks: They are efficient tools for modelling the
relationship between complex input and output responses. Class memberships can be
directly estimated from posterior probabilities. Multilayer perceptron (MLP) is one of the
most commonly used classifier among the ANNs (Zhang, 2000; Topouzelis et al., 2007;
Frate, et al., 2000).
Tree Based Ensemble Techniques – Bagging: In this classifier, in order to
improve the prediction all the individual classification trees are combined. Based on the
binary decisions, the input dataset is split into subsets by the Classification trees. Bagging
is capable of training the separate trees based on samples of dataset. Majority voting and
averaging among the trees determines the predicted classes (Breiman, 1996; Knudby et
al., 2010).
Tree Based Ensemble Techniques – Bundling: In order to build classification
trees, Bundling allows the integration of the predictions of arbitrary classifiers trained as
an additional predictor variable, which is more efficient than bagging (Hothorn and
Lausen, 2005).
Tree Based Ensemble Techniques – Boosting: In this technique accuracy is
improved by mixing the output of tree-based classifiers. Using the weighted voting
among the trees, makes it possible to predict the labels and allow the evolution of trees
over time (Friedman, 2001).
3. Case Study I - Supervised
A supervised object-oriented classification is applied to discriminate oil slicks
by using segmentation techniques and fuzzy classification method. The model has been
developed in the eCognition Software environment. The detailed presentation of the
methodology can be found in the Topouzelis et al., (2007) and Karathanassi et al., (2007).
The model involves filtering, segmentation, dark formation detection, feature extraction
and the fuzzy classification steps. The performance of the simplified version of the
methodology is demonstrated on TerrSAR-X sample dataset from Gulf of Mexico, USA
(Figure 3). The data acquired in single polarization VV on 10 July 2010 for the Deepwater
Horizon oil spill. The data is collected in Stripmap mode with the resolution of 3m.

146
Figure 3. TerraSAR-X, Stripmap mode, VV, sample data.
a. Object Based Classification Model (Topouzelis et al., 2007)

The model based on capturing the statistical properties of the dark areas.
Segmentation is performed at two different scales; coarse and small scales. Coarse scale
is used for to estimate local contrast threshold for each large segment. Each small
segment’s statistical values are compared with the threshold of the big segment. Dark
areas on the image are detected by using threshold algorithm in two stages. In the first
stage, objects with high contrast and low brightness values relative to neighbours were
distinguished. At the second stage threshold that is defined by using mean and the
standard deviation of the large segment is compared with the mean value of each small
object by using empirical statistical formula Karathanassi et al., (2007). Once the dark
areas are detected several features can be derived. A set of 10 features was introduced in
(Stathakis et al., 2006). Following the feature extraction step, the classification step is
performed by using fuzzy logic algorithm. Fuzzy rules are used to express the
characteristic of the oil spills according to the features given by the condition. The model
steps are shown in Figure 4.

147
Figure 4. Flow diagram of the used model.

b. Model implementation on TerraSAR-X Dataset

The original data is divided into three images, one for training and the two for
the testing of the model. Each image contains a certain number of dark objects. Filtering
was done by using Sigma Lee filter with 7x7 window size to remove the speckle noise
from each image. 8-bit filtered TerraSAR images are imported to the eCognition
Software. It was also proposed in Karathanassi et al. (2007) to use different kind/size of
filter and imported them as a different image layer to the model.
Multiresolution segmentation is performed at two different scales in order to produce
large and small objects. To detect the dark areas the two stages thresholding algorithm is
performed. The first stage uses the high contrast and low brightness to neighbour feature
values to characterize the dark areas. Second stage uses the formula Karathanassi et al.
(2007) given below:
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵 + 𝑆𝑆𝑆𝑆. 𝐷𝐷𝐷𝐷𝐷𝐷𝐵𝐵𝐵𝐵𝐵𝐵 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵
+ − 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 ≥ 0
2 𝑆𝑆𝑆𝑆. 𝐷𝐷𝐷𝐷𝐷𝐷𝐵𝐵𝐵𝐵𝐵𝐵
Where 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 is the mean value of the fine detailed segmentation, 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵
is the mean of the coarse scale objects and 𝑆𝑆𝑆𝑆. 𝐷𝐷𝐷𝐷𝐷𝐷𝐵𝐵𝐵𝐵𝐵𝐵 is the standard deviation of the
coarse scale objects.
Two features out of a set of 10 features Stathakis et al. (2006) are considered:
first one is the object power to mean ratio which can be formulated as the ratio between
standard deviation and the mean values of the object. Second, one is the mean Haralick
texture; it is calculated as the average of the grey level co-occurrence matrices of the sub
objects. Fuzzy rules are applied to every feature selected to perform the classification and
oil spill detection. To increase the performance of the model for distinguishing the oil
spill from look-alike, other features proposed at Stathakis et al. (2006) should be
considered.
c. Classification Results

Training and testing images classification results are shown at Figure 5 and
Figure 6. Detected dark objects which aren’t classified as oil spills are shown in red while
the possible oil spills are shown in green at the result image.

148
Figure 5. Training data set and the oil slick result.

Figure 6. Testing data 1 and the oil slick result.


4. Case Study II – Semi-supervised
In 2017 Ennore oil spill was occurred outside Kamarajar port in Ennore India.
The spill occurred on 28 January 2017 when empty tanker collided with a loaded oil
tanker. Sentinel 1A in Interferometric Wide (IW) acquisition mode data is acquired in
dual polarization VV VH on 29 January 2017. High-resolution (40mx40m) Level-1
Ground Range Detected (GRDH) product is chosen for oil spill detection to be used in
ESA SNAP Platform.

SNAP uses Adaptive Thresholding Algorithm for oil spill detection. The dark
spots are detected using an adaptive method. First the local mean backscatter level is
estimated in a large user defined window size, if the value is lower than the detecting
threshold (user defined) then the area is detected as dark spot. Contiguous detected pixels
are clustered into a single cluster. Clusters, which are smaller than user selected minimum
size, are eliminated. For dark spot detection, following parameters can be defined by user:

• Source Bands: User can select one or more bands to produce multi-look image
• Background Window Size: To compute the local mean backscatter level, the
window size in pixels has to be set
• Threshold Shift (dB): The detecting threshold for black spots, which is compared
with the local mean backscatter level within the window.

149
To improve the accuracy, different types of speckle filters can also be applied to
the data before the oil-spill detection step. Original amplitude image (dB) of VV polarized
data is shown at Figure 7.
Sigma Lee filter with 7x7-window size is applied to the original image to remove
the speckle noise. Following the speckle removal filter step oil spill detection algorithm
is applied. Parameters are set to 256 pixels for window size and 0.1 sq. km for minimum
cluster size. Different threshold levels are tested to detect possible oil spills. The result is
shown at Figure 8 and Table 3.

Applying a speckle filter gives better accuracy for detecting the oil slicks. Taking
different parameters for the threshold algorithm would also affect the performance of the
algorithm.

Table 3. Different Thresholds and oil probability.


Oil Spill Mask Threshold (dB) Decision
Green 1.0 dB Very Low Probably Oil
Dark Blue 2.0 dB Low Probably Oil
Light Purple 2.5 dB Probably Oil
Purple 3.0 dB High Probably oil

Figure 7. Original amplitude image (dB) from GRDH VV polarized S1A data.

150
Figure 8. Window size is set to 256. Four oil spill masks for four different
Thresholds are shown on the filtered amplitude image (dB).
5. Conclusion
Satellite based SAR systems provide a cost effective solution to monitor oil
spills over large water bodies with their advantage of being operable in all weather
conditions and during day and night. Data obtained from optical sensor can provide
complementary information for oil spill monitoring specifically under clear weather
conditions. In addition, hyperspectral sensors have great potential for detailed
identification of materials and better estimation for oil spill monitoring. Laser
fluorosensors mostly mounted on aerial platforms could be also used for the identification
of oil spills. However, it is not possible to cover large areas with areal platforms which is
a drawback for these systems. On the other hand, the effectiveness of such systems is very
high in rapid intervention-based approaches and they enable the intervention teams and
equipment to be used in a very effective way to prevent pollution.
It is not possible to provide all the information needed for oil spill surveillance
by using just a single sensor. Therefore, integrated usage of different remote sensing
sensors and platforms is a good approach to create a valuable solution for oil spill
monitoring and detection. In this study, SAR based satellite systems are examined in
detail to illustrate their potential for oil spill detection. The impact of the spills depends
on the type and amount of oil, location, term, water depth, meteorological conditions and
oceanographic features. Remote sensing technologies could aid to distinguish spills
before they cause extensive damage. In the case of larger accidents, remote sensing
images allow easy review of the extent of oil spills. Successful study had to use contextual
information like wind, bathymetry and presence of algae. Biogenic film, grease ice, wind
front areas, wind sheltering, rain cells and shear zones can cause false alarms and resulted
in look-a-likes. Polarimetric SAR is an alternative data source to better identify oil slick
and biogenic slicks. Oil type and thickness information is also important for an accurate
detection. Speckle noise can affect radiometric quality. Good radiometric resolution is

151
also another parameter that should be considered for reliable oil slick detection. Satellites
with higher spatial resolution and more frequent revisit capabilities are expected to
improve in terms of better oil slick detection, identification, oil type recognition and
volume estimation.

References
Ahmad A., and S. Quegan, 2012. Analysis of maximum likelihood classification on
multispectral data. Applied Mathematical Sciences, 6(129), 6425-6436.
Argenti. F., A. Lapini, T. Bianchi, and L. Alparone, 2013. A Tutorial on Speckle
Reduction in Synthetic Aperture Radar Images. IEEE Geoscience and Remote
Sensing Magazine, 1 (3), September pp 6-35.
Bannerman, K., 2003. Oil Spill and Seep Detection, MDA Presentation.
Bern, T.-I., S. Moen, T. Wahl, T. Anderssen, R. Olsen, and J.A. Johannessen, 1992. Oil
spill detection using satellite based SAR.” Completion report for Phase 0 and 1.
Technical report, OCEANOR report no. OCNR92071, Trondheim.
Breiman, L., 1996. Bagging predictors. Machine Learning, 24, 123–140.
Brekke C. and A.H.S. Solberg, 2005. Oil spill detection by satellite remote sensing.
Remote Sensing Environment, 95(1), 1-13.
Brekke, C., and A.H.S. Solberg, 2008. Classifiers and confidence estimation for oil spill
detection in ENVISAT ASAR images. IEEE Geoscience and Remote Sensing
Letters, 1, 65–69.
Brown, C.E., 2017. Laser fluorosensors. In Oil Spill Science and Technology, 2nd ed.
(Ed: M. Fingas), Gulf Publishing Company, Chapter 7; 402–418, Cambridge,
MA, USA.
Chehresa, S., A. Amirkhani, G. Rezairad, and M. Mosavi, 2016. Optimum features
selection for oil spill detection in SAR image. Journal of the Indian Society of
Remote Sensing, 44 (5), 775-787.
Dasari, K. and L. Anjaneyulu, 2017. Importance of Speckle filter window Size and its
impact on Speckle reduction in SAR images. International Journal of Advances
in Microwave Technology (IJAMT), 2(2), 98-102.
Doğan, E., C. Gazioğlu, E., Okuş, İ.E. Gönenç, H. Yüce, and Z.Y. Yücel, 1998.
Examination of the pollution in the West Black Sea by remote sensing
technologies. Journal of Black Sea/Mediterranean Environment, 4(2), 89-101.
Dutta S., M., E.V.S.S. Joseph, and A.V.V, Kumari Prasad, 2018. Automated approach
for extraction of oil spill from SAR imagery. Journal of the Indian Society of
Remote Sensing, 46, doi: 10.1007/s12524-017-0728-0.
Fingas, M. and C. Brown, 2014. Review of oil spill remote sensing. Marine Pollution
Bulletin 83, 9–23.
Fingas, M. and C.E. Brown, 2018. A review of oil spill remote sensing. Sensors, 18, 2-
18.
Fiscella, B., A. Giancaspro, F. Nirchio, P. Pavese, and P. Trivero, 2000. Oil spill detection
using marine SAR images” International Journal of Remote Sensing, 21(18),
3561-3566.

152
Frate, F. D., A. Petrocchi, J. Lichtenegger, and G. Calabresi, 2000. Neural networks for
oil spill detection using ERS-SAR data. IEEE Transactions on Geoscience and
Remote Sensing, 38 (5), 2282–2287.
Frate, F. D., D. Latini; and V. Scappit, 2017. On neural networks algorithms for oil spill
detection when applied to C- and X-and SAR, 2017 IEEE International
Geoscience and Remote Sensing Symposium (IGARSS). 5249-5251.
Friedman, J. H. (2001). Greedy function approximation: A gradient boosting machine.
Annals of Statistics, 29 (5), 1189–1232.
Frost V.S., Stiles J.A., Shanmugan K.S. and Holtzman J.C. 1982. A model for radar
images and its application to adaptive digital filtering on multiplicative noise.
IEEE Transactions On Pattern Analysis and Machine Intelligence PAMI-4(2),
157-165.
Fuk-kwok, C.C. and D. N Held, 1983. Comparison of Several Techniques to Obtain
Multiple-Look SAR Imagery, IEEE Transactions on Geoscience and Remote
Sensing, 21(3), 666-671.
Garcia-Pineda, B., I.R. MacDonald, and B. Zimmer, 2008. Synthetic aperture radar image
processing using the supervised textural-neural network classification
Algorithms IGARSS IV (1), IV1265–IV1268, 4779960.
Garima, V., B. Anupama, and G. Divya, 2015. Detection of oil spills using feature
extraction and threshold based segmentation techniques, 2015 2nd International
Conference on Signal Processing and Integrated Networks (SPIN), 579-583.
Gazioğlu, C., 2018. Biodiversity, coastal protection, promotion and applicability
investigation of the ocean health index for Turkish seas. International Journal of
Environment and Geoinformatics (IJEGEO). 5(3), 353-367.
Genovez, P., N. Ebecken, C. Freitas, C. Bentz, and R. Freitas, 2017. Intelligent hybrid
system for dark spot detection using SAR data. Expert Systems with
Applications: An International Journal, 81, 384–397.
Girard-Ardhuin, F., G. Mercier, and R. Garello, 2003. Oil slick detection by SAR
imagery: potential and limitation” Proceedings OCEANS'03, 1, 164–169.
Hastie, T. 2006. GAM: Generalized additive models. R package version 0.98.
http://cran.r-project.org/
Hothorn, T., and Lausen, B., 2005. Bundling classifiers by bagging trees. Computational
Statistics and Data Analysis, 49, 1068–1078.
Huang Y., and J. van Genderen, 1996. Evaluation of several speckle filtering techniques
for ERS-1 and 2 imagery, International Archives of Photogrammetry and
Remote Sensing, 21 (B2) ,164-169.
Jiao X., and X. B. Wen, 2009. SAR image segmentation based on markov random field
model and multiscale technology, Sixth International Conference on Fuzzy
Systems and Knowledge Discovery, Tianjin, 442-446.
Karathanassi V., K. Topouzelis, P. Pavlakis, and D. Rokos, 2007. An object‐oriented
methodology to detect oil spills, International Journal of Remote Sensing,
27(23), 5235-5251, DOI: 10.1080/01431160600693575
Keramitsoglou I., C. Cartalis, and C. Kiranoudis, 2006. Automatic identification of oil
spills on satellite images, Environmental Modelling and Software, 21, 640-652.

153
Knudby, A., E. Ledrew, and A. Brenning, 2010. Predictive mapping of reef fish species
richness, diversity and biomass in Zanzibar using IKONOS imagery and
machine-learning techniques Remote Sensing of Environment, 114(6), 1230–
1241.
Kuan D.T., A.A. Sawchuk, T.C. Strand, and Chavel, P. 1985. Adaptive noise smoothing
filter for images with signal-dependent noise, IEEE Transactions on Pattern
Analysis and Machine Intelligence, PAMI-7 (2), 165-177.
Kuan D.T., A.A. Sawchuk, T.C. Strand, and P. Chavel, 1987. Adaptive restoration of
images with speckle. IEEE Transactions on Acoustic and Signal Processing, 35,
373–383.
Lankoande O., M. M. Hayat and B. Santhanam, 2005. Segmentation of SAR images
based on Markov random field model, IEEE International Conference on
Systems, Man and Cybernetics, 3,2956-2961.
Lehrer, S., 2014. Ship surveillance with high resolution Terrasar-X satellite in African
waters, Remote Sensing of the African Seas, New York, NY, USA: Springer-
Verlag, pp. 285–313
Leifer, I., B. Lehr, D. Simecek-Beatty, E. Bradley, R. Clark, P. Dennison, Y. Hu, S.
Matheson, C. Jones, and B. Holt, 2012. State of the art satellite and airborne oil
spill remote sensing: Application to the BP Deepwater Horizon oil spill. Remote
Sensing Environment, 124, 185–209.
Lopes, A., E. Nezry, and H. Laur, 1990a. Maximum a posteriori speckle filtering and first
order texture models in SAR Images, Processing of IGARSS, pp.2409-2412.
Lopes, A., H. Laur, and E. Nezry, 1990b. Statistical Distribution and Texture in Multi-
look and Complex SAR Images, Processing of IGARSS 1990, pp.2427-2430.
Marzialetti, P. and G. Laneve, 2016. Oil spill monitoring on water surfaces by radar L, C
and X band SAR imagery: A comparison of relevant characteristics. In
Proceedings of the 2015 IEEE International Geoscience and Remote Sensing
Symposium (IGARSS), Beijing, China; pp. 7715–7717.
Masuko, H., T. Kobayashi, K. Okamoto, and W. Alpers, 1995. Observation of artificial
slicks with SIR-C/X-SAR around Japan. IEEE Proceedings of the International
Geoscience and Remote Sensing Symposium IGARSS'95, 1, 227-229.
Mera, D., V. Bolon-Canedo, J.M. Cotos, and A. Alonso-Betanzos, 2017. On the use of
feature selection to improve the detection of sea oil spills in SAR images,
Computers & Geosciences, 100 (2017), pp. 166-178,
10.1016/j.cageo.2016.12.013.
Migliaccio, M., F. Nunziata and A. Buono, 2015. SAR polarimetry for sea oil slick
observation. International Journal of Remote Sensing, 36(12), 3243-3273.
Moguerza, J. M., and A. Muñoz, 2006. Support vector machines with applications.
Statistical Science, 21(3), 322–336.
Mountrakis, G., Im, J., and C.Ogole, 2011. Support vector machines in remote sensing:
A review. ISPRS Journal of Photogrammetry and Remote Sensing, 66, 247–259.
Radhika, V. and G. Padmavathy, 2011. De-speckling of oil spill SAR images using fusion
technique. International Journal of Computer Applications, 21 (1), 26-32.

154
Robson, M., J. Secker, and P.W. Vachon, 2006. Evaluation of recognition for assisted
target detection and recognition in SAR imagery. IGARSS 1454241189, 145–
148.
Richards, J. A., 1993. Remote Sensing Digital Image Analysis: an Introduction, 2nd
edition (New York: Springer).
Samad, R. and S.B. Mansor, 2002. Detection of Oil Spill Pollution Using RADARSAT
SAR Imagery. In: CD Proc. of 23rd Asian Conference on Remote Sensing
Birendra International Convention Centre in Kathmandu, Nepal, November 25-
29. Asian Remote Sensing.
Schisted Solberg, A.H., 2012. Remote sensing of ocean oil-spill pollution. Proceedings
of the IEEE, 100 (10), 2931-2945.
Schvartzman, I., S. Havivi, S. Maman, S.R. Rotman, and D.G. Blumberg, 2016. Large oil
spill classification using SAR images based on spatial histogram. ISPRS
International Archives of the Photogrammetry Remote Sensing and Spatial
Information Sciences 41, 1183–1186.
Şeker, D.Z., Ş. Direk, N. Musaoğlu, and C. Gazioğlu, 2013. Determination of effects of
coastal deformation caused by waves and storms at Black Sea coast of Turkey
utilizing InSAR technique, American Geophysical Union, Fall Meeting
Abstracts, 1629p.
Singha ,S., T.J. Bellerby, and O. Trieschmann, 2013. Satellite oil spill detection using
artificial neural networks. IEEE Journal of Selected Topics in Applied Earth
Observations and Remote Sensing 6(6), 2355-2363.
Solberg, A.H.S., 2012. Remote sensing of ocean oil-spill pollution. Proceedings of the
IEEE, 100 (10), 2931-2945.
Solberg, A.H.S., and R. Solberg, 1996. A large-scale evaluation of features for automatic
detection of oil spills in ERS SAR images. Proc. IGARSS’96, 3, 1484–1486.
Solberg, A.H.S., C. Brekke, and P. Ove Husoy, 2007. Oil spill detection in radarsat and
envisat SAR images. Geoscience and Remote Sensing, IEEE Transactions On
45(3), 746-755.
Solberg, R., and N.A. Theophilopoulos, 1997. Envisys–A solution for automatic oil spill
detection in the Mediterranean. Proceedings of 4th thematic conference on
remote sensing for marine and coastal environments, Environmental Research
Institute of Michigan, Ann Arbor, Michigan, 3-12.
Stathakis, D., K Topouzelis, and V. Karathanassi, 2006. Large-scale feature selection
using evolved neural networks. Image and Signal Processing for Remote
Sensing XII 6365, 636513:1-9
Taravat, A., F. Del Frate, and F. Weibull, 2013. Multiplicative Model and machine
learning models for full-automatic dark-spot detection from SAR images. ISPRS
Archives 40, 421–424.
Topouzelis, K., V. Karathanassi, P. Pavlakis, and D. Rokos, 2007. A new object-oriented
methodology to detect oil spills using ENVISAT images. Proceedings of
‘ENVISAT Symposium 2007’, 23–27 April 2007, Montreux, Switzerland.
Topouzelis, K., V. Karathanassi, P. Pavlakis, and D. Rokos, 2009. Potentiality of Feed-
Forward Neural Networks for Classifying Dark Formations to Oil Spills and
Look-alikes. Geocarto International, 24, 179-191.

155
Tripathi, A., V. Bhateja, A. Sharma, and A. Kuan, 2017. Modified anisotropic diffusion
approach for speckle filtering. Advances in Intelligent systems and Computing,
458, 537–545.
Vijayakumar, S., P. Swarnalatha, and S. Rukmini, 2016. A neural network classification
approach for oil spill detection on SAR images. IIOAB Journal 7, 225–235.
Xu, L., J. Li, and A. Brenning, 2014. A comparative study of different classification
techniques for marine oil spill identification using RADARSAT-1 imagery.
Remote Sensing of Environment, 141, 14-23 02/2014.
Zhang, G. P., 2000. Neural networks for classification: A survey. IEEE Transactions on
Systems, Man, and Cybernetics Part C: Applications and Reviews, 30, 451–462.
Zhang, Y., H. Li, X. Wang, and W. Dan, 2010. Edge extraction of marine oil spill in SAR
images. In: Proceedings of the 2010 International Conference on Challenges in
Environmental Science and Computer Engineering (CESCE), Wuhan, China, 6–
7 March 2010, 439–442.
Zhou, X., G. Ai, and X. Zheng, 2009. Oil spill detection based on the texture
characteristics of SAR Images. In: Proceedings of the 30th Asian Conference on
Remote Sensing, Beijing, China, 18–23 October 2009, 1161–1165.

156

You might also like