8 Perspectives on Oil Spill Detection Using Synthetic Aperture Radar
139
levels. Such levels are supposed to describe the probability that an observed dark
feature in the satellite image is related to the presence of an oil spill. The SAR
derived oil spill detection probability estimation has been mathematically explored
as an intrinsic aspect of oil spill classification, which fundamentally computes the
likelihood that the detected dark area and its extracted features are related to oil spill.
In particular, Fiscella et al. (2000) introduced a nearest neighbours classifier based
on the Mahalanobis metric applied to the feature vector, and compared its results to a
statistical classifier. Nirchio et al. (2005) investigated a Fischer discriminant analysis
approach applied to the selected features. Nevertheless, the SAR based probability
estimation should be integrated with additional criteria in order to become a more
effective tool for the End User. It is worth mentioning also that some of these criteria and their correspondent weight within the decision making process, are very
National and Regional dependent.
New trends in confidence level estimation involve different sources of data,
linking the SAR derived information to ship traffic routes, metocean and other
context-specific information. For this reason we are observing a progressive transition from a three-level SAR based Confidence output to a more articulated Alert
System that includes additional criteria. This is summarised by Ferraro et al. (2009).
The application output shall consist of a product that delivers additional information
not only in terms of reliability but also concerning the related alarm extent of the
detection (i.e. potential impact and polluter detection capability).
8.3.4 Ancillary Data
In order to increase the reliability of oil spill classification, environmental data have
been used as additional help to the human operator (Tahvonen and Pyhalahti, 2006;
Muellenhoff et al., 2007).
The ancillary data can be grouped into two main conceptual datasets, Risk of
Pollution and Detection Capability. The former contains information about areas at
risk due to its proximity to potential pollution sources. These data contain information about distance from main traffic lanes, world ports distribution, pipeline runs,
wreckage risk, oil rig locations. Detection Capability Dataset holds information
about specific ocean and atmospheric phenomena that decrease the detection capability of oil spills based on SAR imagery. This dataset includes among others: low
wind conditions, Sea Surface Temperature (SST) derived cold water fronts, air-sea
interactions, current shears, biogenic oily films, etc.
The generated maps of Risk of Pollution and Detection Capability degree could
be used for cross checking with the automated detected outcomes, in order to effectively assign the correspondent level of reliability to oil spill candidates. The data
described here are not related to the specific time of SAR image acquisition. On the
contrary, such maps aim at describing monthly and seasonal trends in the areas of
interest.
These maps could be considered a potential decision making support and
validation tool that could be easily adapted to stakeholders requirements.
139
levels. Such levels are supposed to describe the probability that an observed dark
feature in the satellite image is related to the presence of an oil spill. The SAR
derived oil spill detection probability estimation has been mathematically explored
as an intrinsic aspect of oil spill classification, which fundamentally computes the
likelihood that the detected dark area and its extracted features are related to oil spill.
In particular, Fiscella et al. (2000) introduced a nearest neighbours classifier based
on the Mahalanobis metric applied to the feature vector, and compared its results to a
statistical classifier. Nirchio et al. (2005) investigated a Fischer discriminant analysis
approach applied to the selected features. Nevertheless, the SAR based probability
estimation should be integrated with additional criteria in order to become a more
effective tool for the End User. It is worth mentioning also that some of these criteria and their correspondent weight within the decision making process, are very
National and Regional dependent.
New trends in confidence level estimation involve different sources of data,
linking the SAR derived information to ship traffic routes, metocean and other
context-specific information. For this reason we are observing a progressive transition from a three-level SAR based Confidence output to a more articulated Alert
System that includes additional criteria. This is summarised by Ferraro et al. (2009).
The application output shall consist of a product that delivers additional information
not only in terms of reliability but also concerning the related alarm extent of the
detection (i.e. potential impact and polluter detection capability).
8.3.4 Ancillary Data
In order to increase the reliability of oil spill classification, environmental data have
been used as additional help to the human operator (Tahvonen and Pyhalahti, 2006;
Muellenhoff et al., 2007).
The ancillary data can be grouped into two main conceptual datasets, Risk of
Pollution and Detection Capability. The former contains information about areas at
risk due to its proximity to potential pollution sources. These data contain information about distance from main traffic lanes, world ports distribution, pipeline runs,
wreckage risk, oil rig locations. Detection Capability Dataset holds information
about specific ocean and atmospheric phenomena that decrease the detection capability of oil spills based on SAR imagery. This dataset includes among others: low
wind conditions, Sea Surface Temperature (SST) derived cold water fronts, air-sea
interactions, current shears, biogenic oily films, etc.
The generated maps of Risk of Pollution and Detection Capability degree could
be used for cross checking with the automated detected outcomes, in order to effectively assign the correspondent level of reliability to oil spill candidates. The data
described here are not related to the specific time of SAR image acquisition. On the
contrary, such maps aim at describing monthly and seasonal trends in the areas of
interest.
These maps could be considered a potential decision making support and
validation tool that could be easily adapted to stakeholders requirements.