Maritime Surveillance 2

Maritime surveillance has become a leading operational application of spaceborne Synthetic Aperture Radar (SAR) with many countries seeking to exploit this capability to identify suspicious or illegal shipping activity. In Canada, RADARSAT-2 is used to provide near real-time surveillance of shipping approaching the east and west coasts. This capability will further expand with the Canadian Space Agency’s three- satellite RADARSAT Constellation Mission (RCM). RCM has been designed to provide daily coverage of approaches to Canada out to 1200 nm. In Europe, both ESA’s Sentinel-1 and national SAR capabilities are used to support maritime surveillance for EMSA and other agencies.

As the commercial operator of RADARSAT-2 and the prime contractor for RCM, MDA has a long history with SAR-based maritime surveillance spanning the full spectrum of required expertise from sensor design through to data analysis. This paper will describe recent MDA development work relating to maritime surveillance, including:

The addition of new ScanSAR modes to RADARSAT-2 that are designed specifically for maritime surveillance.

Improving operational tools for detecting and characterizing vessels in SAR imagery. This includes making greater use of all available SAR data sources, improvements to algorithm accuracy and performance, optimizing operator efficiency and enhancing the information content of outputs.

Validation of SAR performance at detecting and characterizing vessels, in particular understanding the performance of detecting small vessels and estimating vessel length and heading and algorithm improvements arising from this.

Improvements to downstream tools for visual analytics that fuse SAR–based vessel detection reports with other vessel information such as AIS.

Automatic ship reporting systems (AIS, LRIT and VMS) provide a tool to monitor ship traffic. Coastal AIS networks produce a lot of ship position data and enable continuous tracking of the ships in their range, but especially satellite AIS reception and distribution have made it possible to track ships globally and essentially continuously. Still, this only concerns that fraction of the ship traffic that carries an automatic reporting system. AIS, the most widespread self-reporting system, is mandatory on ships of 300 tonnes and up, and on several other classes of ships. Even if it is also carried voluntarily on some smaller ships (Class B AIS), this still leaves many ships not reporting. Satellite SAR is a valuable tool to explore that portion of the ship traffic that is non-reporting. It will however only allow a sparse sampling, as its availability falls far short of a continuous and global tracking capacity, even for Sentinel-1 that already is making available much more data than other satellite SARs.

This paper presents the results of a campaign in the Western Indian Ocean to assess the fraction of non-reporting ship traffic. It was carried out under the PMAR-MASE project (Piracy, Maritime Awareness and Risks for Maritime Security), for which ship reporting data were collected from five different operators of satellite AIS and coastal AIS networks, plus LRIT data from EU-linked ships. The collection persisted during one year, starting 15 Sep 2014. These data provide a very complete picture of the reporting ship traffic in the Area of Interest (AOI), which extends from the East African coast out almost to the Maldives in the east, and from Yemen/Oman in the north to Durban in the south. From this AOI, every day more than 560,000 ship position reports were received, and the derived ship traffic picture shows the presence of 1,350 self-reporting ships in the AOI at any one time (average numbers for June 2015).

In order to probe for the presence of non-reporting ships, SAR images were collected over the AOI from a series of satellites: Sentinel-1 (S-1), RADARSAT-2, TerraSAR-X, COSMO-SkyMed and ALOS-PALSAR-2. The ESA S-1 rolling archive was queried for images in the AOI and a bit wider. This yielded 1,500 images up to 15 Sep 2015. These images, acquired according to the pre-definedobservation plan, are not well distributed over the AOI, being concentrated mostly in a few locations. The S-1 images were analysed with the SUMO ship detection software in fully automatic batch mode. As expected, it was found that to prevent an excessive number of false alarms, the detection thresholds had to be set quite high, especially in the co-pol channels. This set of images yielded around 18,000 detections (4,000 of these south of Durban). After automatic correlation with the self-reported ship positions, interpolated to the time of the SAR images, it was found that 40 % of the detections could not be correlated.

A few of the S-1 images were in addition analysed semi-automatically, first with the automatic ship detector and then verified manually. The images from the other four satellite SARs, in total 60, were treated in a similar manner. Some RADARSAT-2 images, in the MSSR-DVWF mode that is optimised for vessel detection, were analysed by DRDC; the TerraSAR-X images were analysed by DLR at the Maritime Security Lab Neustrelitz. Preliminary results (to be updated in the coming months) show that about 35 % of the targets detected and manually verified in the SAR images were non-reporting. This percentage however is rather variable from one location to another. In particular, many non-reporting ships are seen off the Somali coast. 

Synthetic Aperture Radar (SAR) is a well-established tool for oil spill remote sensing. The SAR imaging of surface slicks depends on slick characteristics, environmental conditions and sensor properties. SAR is currently used operationally for oil spill detection. However, research is increasingly focusing on slick characterization, i.e., discrimination between oil spills and look-alikes and extraction of oil spill information. This topic is frequently addressed using SAR polarimetry. As we are moving into the area of slick characterization, it is important to map the effect the different factors mentioned above have on the multipolarization parameters and on the characterization ability.

In this work, we analyze data from a controlled oil spill experiment, the NOrwegian Radar oil Spill Experiment (NORSE2015), which took place in the North Sea in June 2015. The experiment was a joint UiT The Arctic University of Norway and Jet Propulsion Laboratory (JPL) / National Aeronautics and Space Administration (NASA) experiment done in collaboration with the Norwegian Clean Seas Association for Operating Companies (NOFO). Four substances with varying properties were released onto the sea surface, and remote sensing data were collected by various sensors, including Radarsat-2, TerraSAR-X, RISAT-1 and the NASA Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR). The latter monitored the slicks for several hours after releases, providing regular SAR data acquisitions as the slicks evolved.

We here investigate four UAVSAR scenes, acquired with a temporal separation of about 13 minutes. Due to the short time span between data uptakes, we assume that slick properties change little between the scenes. In addition, in situ wind measurements show stable wind conditions over the observation period. Hence, we can assume that any changes in the slicks between scenes are mainly related to the observation geometry, including incidence angle and the look direction relative to the wind direction. This allows us to investigate the effect of these parameters on the SAR imaging of the slicks.

Each scene covers an incidence angle range from about 22°-67°, but the slicks are located at various incidence angles in the different scenes. The pass direction alternates between ascending and descending, and hence the look direction alternates between close to upwind and close to downwind.

The objective of this work is to investigate how different multipolarization parameters are affected by varying incidence angle and varying look direction relative to the wind in both clean sea and in slick-covered areas.

Preliminary results show that all of the investigated multipolarization features have a clear variation with incidence angle over clean sea. Slick-covered regions show similar variations. The slick-sea difference varies with incidence angle, with a generally larger separation at the lower incidence angles. The features dependencies on look direction relative to the wind vary a lot. The real part of the copolarization cross product, r_co, which has previously been identified as a promising parameter for slick characterization, is found to be among the features with less dependency on wind direction. In this feature, the four slicks in all four scenes have mean values that lie below all four clean sea profiles. We also find that r_co have among the best slick-sea contrasts over the different incidence angles. These findings support the use of this feature as it may be less sensitive to variations in wind direction, and hence could be more easily generalized for operational use.

The final paper will contain a comparison between multipolarization features and their sensitivity to the factors considered here. The UAVSAR time series also allows for the same comparison to be repeated on several data subsets.

More information on data set and preliminary results are provided in the extended abstract uploaded as a pdf.

The new European Sentinel-1 to -5 satellites has been developed by the European Union (EU) and European Space Agency in partnership. The satellites shall provide data for the EU’s Copernicus program well into the next decade for environmental monitoring and security applications. Sentinel-1A (launched 2014) and Sentinel-1B (launch in 2016) are C-band SAR satellites, and data from the satellites will be made available under a free and open data policy.

Norway uses SAR satellites operationally for maritime surveillance, and the Norwegian Defence Research Establishment has therefore done an initial evaluation of Sentinel-1A to determine if the data and associated monitoring services are suitable for national operational requirements.

For the period timeliness remains a key issue. Product availability does not meet timeliness or latency requirements of operational users in the maritime sector. With the recently established Norwegian national ground segment for downloading data, the time delay will be reduced greatly. An analysis of this improvement will be presented.

The use of the EW and IW modes is more or less as expected. Some variations from week to week are unexplained. From a Norwegian maritime surveillance perspective, it would be more advantageous to increase the use of HH/HV dual-polarised data over ocean and coastal areas, and reduce the use of VV/VH. This is especially the case for the IW mode where most of the images are in the VV-/VH-pol combination. The use of the IW mode in the fisheries protection zone around Spitsbergen, as well as along the continental shelf edge from Spitsbergen down to Troms/Finnmark would give better performance for detection of fishing vessels due to better resolution, but a HH-/HV-pol combination is wanted. Statistically, a HH-/HV-pol combination gives better ship detection performance, and also probably better sea ice and iceberg observations.

For dual-polarisation images, cross-polarisation images give better contrast between land and sea as well as between ice and sea.

Ship detection in Sentinel-1A images in the Norwegian interest ocean areas will be presented. The ship detection is done in all available polarisation channels. The contrast between a vessel and the ocean background is calculated in two ways:

1)      The ship to sea contrast - maximum amplitude divided by mean sea background

2)      Target to clutter ratio – radar cross section (RCS) of the vessel divided by the mean sea background

One vessel of particular interest is the oil production vessel Norne FPSO. The vessel is moored to the ocean floor in the Norne field outside the west coast of Norway. Thus, the vessel can be imaged multiple times at the same position. The two contrast measures mentioned above are calculated for Norne FPSO for different polarsations and different incidence angles. The two contrast measures for other vessels at different sizes (small, medium and large) will also be presented.

The co-polarisation channel and the cross-polarisation channels can be combined to increase the two contrast measures for better ship detection results. Results from this improvement will also be presented for the three different vessel size groups.

A further development of a new multipolarization technique for sea surface slick type discrimination [Ivonin et al., 2015] based on the full normalized radar cross-section model [Kudryavtsev et al., 2003] is proposed. It uses the ratio between the damping in the slick resonant and non-resonant parts of the backscattered signal related, correspondingly, to damping of the Bragg resonant capillary-gravity wind ripple and breaking waves. The technique is tested on variety of TerraSAR-X co-polarized (VV,HH) images synthetic aperture radar images obtained for the Gulf of Mexico, the Caspian Sea, and the North Sea. The images contain as slicks of a priori known provenance reported in [Skrunes et al., 2014, 2015]  as well unknown slicks. It has been shown that the suggested multipolarization technique provides a desired discrimination between mineral oil and plant oil slicks. The technique possesses the beneficial property of being independent to variations in incidence angles in a wide range. Oil spills on images for the Gulf of Mexico, the Caspian Sea, and the North Sea show stably the same characteristic in terms of the ratio between the damping in the slick the capillary-gravity wind ripple and the breaking waves calculated using the model [Kudryavtsev et al., 2003]. Results of our multipolarization technique are compared with one of the co-polarized phase difference method and others.

Acknowledgments. This work was supported by the Russian Foundation for Basic Research, project № 14-05-93084.

References

1. Ivonin D.V., A.Yu. Ivanov, C. Brekke C., and S. Skrunes (2015), Calibrated method for discriminating sea surface slicks using Radarsat-2 co-polarized SAR images. In Proc. Geoscience and Remote Sensing Symposium (IGARSS), IEEE International, p. 3739-3742.

2. Kudryavtsev, V., D. Hauser, G. Caudal, and B. Chapron (2003), A semiempirical model of the normalized radar cross-section of the sea surface: 1. Background model, J. Geophys. Res., 108(C3), FET 2-1-FET 2-24.

3. Skrunes, S., C. Brekke, and T. Eltoft (2014), Characterization of Marine Surface Slicks by Radarsat-2 Multipolarization Features, IEEE Trans. Geosci. Remote Sens., 52(9), 5302-5319.

4. Skrunes, S., C. Brekke, T. Eltoft, and V. Kudryavtsev (2015), Comparing near coincident C- and X-band SAR acquisitions of marine oil spills,” IEEE Trans. Geosci. Remote Sens., 53(4), 1958-1975.

5. Velotto, D., M. Migliaccio, F. Nunziata, and S. Lehner (2011), Dual-polarized TerraSAR-X data for oil-spill observation, IEEE Trans. Geosci. Remote Sens., 49(12), 4751-4762.