Evaluation of Vessel Detection System Use for Monitoring of Fisheries Activities

Transcription

1 Not to be cited without prior reference to the author Evaluation of Vessel Detection System Use for Monitoring of Fisheries Activities Guido Lemoine, Marte Indregard, Christian Cesena, Francois Xavier Thoorens, Harm Greidanus and Hendrik Dörner We present results from the use of a vessel detection system (VDS) which locates vessel positions in remote sensing imagery acquired by orbiting Synthetic Aperture Radar (SAR) satellite sensors. The system, which is developed and maintained in house, has been tested in several vessel traffic situations in European waters and international waters during the last 3 years. Our analysis focuses on 2006 results generated over the Redfish (S. mentella) fishing grounds over the Reykjanes Ridge, south-west of Iceland. The use of VDS in this case study is a practical example of the role it could play in support to surveillance operations and wide area vessel traffic pattern recognition. In particular we look at detection rates, correlation to known vessel Vessel Monitoring System (VMS) positions and explanations for the abundance of detected targets that are unmatched by VMS. The hypothesis that the latter relate to Illegal, Unreported and Illegal (IUU) vessels in the area is confirmed by sightings by the Icelandic Coastguard and Greenpeace. We conclude with a projection of future VDS developments. Keywords: vessel detection system, remote sensing, SAR, VMS, AIS, monitoring and control. Guido Lemoine, European Commission, Joint Research Centre, Institute for the Protection and Security of the Citizen (IPSC), Agricultural and Fisheries Unit, TP 266, Ispra, Italy [tel: , fax: , guido.lemoine@jrc.it] Marte Indregard, Christian Cesena, Francois Xavier Thoorens, Harm Greidanus and Hendrik Dörner, European Commission, Joint Research Centre, Institute for the Protection and Security of the Citizen (IPSC), Agricultural and Fisheries Unit, TP 266, Ispra, Italy [fax: , {marte.indregard, christian.cesena, francois-xavier.thoorens, harm.greidanus, hendrik.doerner}@jrc.it] 1. Introduction The global decline in marine fish stocks is stimulating the discussion for increased efforts on the monitoring and control on fishing activities. The use of the Vessel Monitoring System (VMS) to track the fishing fleet has found wide acceptance in the major fishing nations and has proven to be an effective tool to regulate the presence of licensed and VMS-equipped fishing vessels in restricted areas and time periods. The VMS is an autonomous transponder-based system that reports vessel positions to the flag state at a regular frequency, usually at an hourly or 2-hourly interval. In the European Union (EU) the use of VMS is obligatory on fishing vessels with an overall length of more than 15 meters. VMS is increasingly being adopted by developing countries that have a strong 1

2 interest in the sustainable management of fish stocks in the waters under their jurisdiction and those regulated under various regional fisheries management organisations. Illegal, unregulated and unreported (IUU) fishing activities remain largely unknown to a VMSbased monitoring effort and require surveillance measures that are difficult and expensive to operate, especially for remote areas and extended fishing grounds. Classical surveillance measures deployed for fisheries inspection include vessel and aircraft patrols which are both costly to operate and which have a limited effective area of operation, especially the first category. The use of satellite imaging techniques is currently being contemplated as a complementary technique to the existing Vessel Monitoring System (VMS) to optimise costly surveillance efforts, especially for remote areas of the Exclusive Economic Zones and international waters. The core benefit of a complementary system based on satellite remote sensing lies in its ability to monitor semi-instantaneously a large swath of ocean surface for the presence of vessels. Detected positions can be matched against existing sources of position records, such as those from the VMS, on the basis of which areas can be marked where anomalies are evident. Of prime interest are detected vessel positions for which no known positional records are matched and positional records for which no vessel positions are detected. The latter may reveal anomalous functioning of the VMS. Surveillance platforms can then be redirected to inspect these pre-marked areas for possible IUU activities. The introduction of the satellite Vessel Detection System (VDS) would lead to improved efficiency by increasing the probability for the detection of and intervention in IUU activities (deterrent effect) and a reduction in surveillance costs through optimisation of use of patrol resources. In addition, frequent inter-annual monitoring of a major fishing ground or fisheries campaign in regulated areas can yield estimates and patterns of the abundance of unidentified vessels. Such estimates can be used as correction factors to stock abundance assessments which are based on the use of VMS-derived vessel activity indicators and sparse catch and landing reports. In this paper, we demonstrate an operational VDS that is based on the use of SAR imagery from the Canadian RADARSAT-1 and European ENVISAR ASAR satellite sensors. We demonstrate how we have achieved near real time supply of detected positions over European waters and the North East Atlantic international waters for which the NEAFC control regime is in place. In the next section, we shortly introduce the technical characteristics of the VDS. This is followed by a presentation of results from a 2006 case study over the Reykanes Ridge. We conclude with a summary and discuss the near term outlook for operational use of the system. 2. The vessel detection system The feasibility of using SAR for the detection of fishing vessels positions for matching against VMS records was demonstrated by Kourti et al (2001) and it's applicability in fisheries monitoring and control operations by Kourti et al (2005). The feasibility study was followed up by the implementation of the IMPAST project, which combined the efforts of European partners active in SAR remote sensing with those of potential end-users at the Fisheries Monitoring Centres (FMC) in 4 EU Member States (Spain, Portugal, the United Kingdom and Greece) and Norway and Iceland. Technical expertise from FMC service providers and specialised communication and information technology partners was also incorporated. The prototype system was further developed with 2

3 internal JRC funding. A detailed technical description of the various components is given in Lemoine et al (2005a). The key characteristics of the VDS are as follows: CM 2006/N:02 The system is composed of functional modules that communicate via standard interfaces over the internet using mostly XML for data exchange. The components are proprietary for the functions that are performed in the stand-alone processes at the SAR image supplier (e.g. SAR processing) and FMC (e.g. reception and archiving of VMS data) but implemented in Open Source (OS) for those functions that incorporate the added value of the VDS, i.e. the detection process, the correlation infrastructure and the visualisation and analysis environment. The system can process SAR imagery from both RADARSAT-1 and ENVISAT ASAR for all imaging modes, including the alternating polarisation mode of ENVISAT. SAR image products have been processed for 3 ground stations that cover European waters (Kongsberg Satellite Services ground station in Tromsø, Norway, QinetiQ s ground station in West Freugh, United Kingdom and ITU-SAGRES ground station in Istanbul, Turkey). Integration into the data processing flow of other ground stations is not expected to be difficult. Average near real time throughput performance from at-sensor SAR image acquisition to supply of correlated VMS-VDS records is between minutes, depending on image mode (which determines SAR processing speed and file size) and ground station set-up. The steps for SAR processing and image file transfer to the detection module take up 80 to 90% of the total. We have recently tested a set-up in which the detection module is placed at the image supplier (Kongsberg) achieving a record throughput time of 9 minutes. The target detection kernel can be chosen from pluggable algorithms for point target delineation. The current version allows the choice of two different algorithms based on the classical Constant False Alarm Rate (CFAR) approach, one a semi-empirical template matching method and one based on the K-distribution (Greidanus, 2004). The detection step takes 2-3 minutes for full image frames (2.0+ GHz CPU with 1 Gb RAM). Data exchange interfaces to VMS records at the FMC are based on secure web services (SOAP over SSL) or secure upload routines. In an operational scenario, in which VMS are commonly polled, VMS records from various FMCs are usually available in the time span between image acquisition and VDS record supply. The system has already built in interfaces to other positions records such as those from AIS (in coastal areas) or surveillance platforms, which are integrated into the correlation set-up. The system includes an OS web-mapping interface hosted in the University of Minnesota Map Server that allows full interactive access to the data sets, including warped image data. The interface is designed to support analysis of VMS and VDS matching results with an option to edit target attributes. The system back-end is the OS geospatial PostgreSQL/Postgis data base. Depending on the context of the VDS use, results are typically forwarded by automatic or to dedicated web services that process the information. As an example, we are currently discussing with the Swedish coastguard how the VDS results can be integrated into the onboard display system of the patrol air plane in live surveillance settings. Alternatively, all relevant data can be integrated into the JRC back end and made accessible to the parties involved via secure web mapping interfaces. We tend to separate VDS applications in two broad scenarios, (1) those aimed at support to surveillance operations and (2) those for assessment of vessel traffic patterns. The main distinction 3

4 between these two scenarios is related to the operational set-up. In the first case, near real time supply of VDS results to surveillance platforms is very important, while spatial and temporal sampling is a more important issue in the second. Both scenarios also define operational parameters such as required detection quality, total area coverage, and post-analysis requirements. We have experimented with the VDS in surveillance support in the Baltic Sea, the Channel, and the Mediterranean. Assessment studies have been carried out in the Mediterranean and the Western Waters (ICES area V and VI). For the latter a total coverage of 2.5 million km 2 was monitored in August Operational results We illustrate the use of VDS with operational results we have generated recently over the pelagic redfish (S. Mentella) fishing ground in the north-eastern Irminger Sea, covering the Reykjanes Ridge straddling the south-western Icelandic EEZ (ICES Sub-area XIV). A detailed description of the redfish fisheries is given in (ICES, 2006). Applicability of VDS has been tested over this area every year since 2003, stemming from the active participation of the several EU Member States' FMCs, the Icelandic Coastguard and the NEAFC secretariat in the early development of the system. The area is also of interest due to the frequent presence of Illegal, Unreported and Unregulated (IUU) fishing vessels in the area. This problem was highlighted recently by a Greenpeace (Greenpeace, 2006) report detailing flagrant violations of the NEAFC convention rules by IUU vessels that have been blacklisted, but which are openly active in fisheries in the area. Our work in previous years has drawn the attention of the North Western Working Group (NWWG) of the ICES Advisory Committee on Fishery Management (ACFM), who has recognised the potential role of VDS in obtaining quantitative information on IUU presence for correction of total catch and landing estimates (ICES, 2006, part 10). Such corrections are particularly important as trends in redfish catch and landing reports and stock estimates are decidedly negative. At the same time, the agreed total allowable catch (TAC) for regulated redfish fisheries in 2006 are at 99,000 tons well over the ACFM advice of 41,000 tons. In comparison to previous years, we have seen a marked improvement in the availability of VMS records in We have received VMS records from all major flag states active in the fisheries. Some of these are very well timed, usually within 5 minutes of the image acquisition times. An overview of the study area with the colour coded VMS is shown in Figure 1. The blue outlines in this figure show the limits of each acquisition. The red boundary is the Icelandic EEZ. Grey contours are isobaths. 4

5 Figure 1: Overview of the 2006 locations of the VDS imagery. VMS received for correlation are given as colour coded squares. The red boundary is the Icelandic EEZ. All SAR acquisitions analysed for this study are from Radarsat-1 ScanSAR Narrow mode. This mode produces images of 300 by 300 km 2 with a resolution of 50 m (pixel spacing of 25 m). The ScanSAR Narrow resolution is sufficient to detect the vessels in the study area, which are typically above 40 m in overall length. A detailed discussion of factors influencing the detection quality of the VDS is given in (Lemoine et al 2005b). The Radarsat-1 ScanSAR wide swath is required to monitor the area in a single acquisition. We have also received ENVISAT ASAR APP images, which have better resolution and allow imaging in two polarisations. The swath width of these images is only in the order of 100 by 100 km 2, making them more suitable for detailed assessments. Analysis results from the ENVISAT imagery are pending. In Table 1 we list the total number of VMS (column 2) for each of the image acquisitions (column 1) which are used for correlation to VDS targets (column 3). We also list the results produced by automatic correlation. The automatic correlation (column 4) pairs a VMS position with the nearest VDS (both in time and distance). This process is complicated in the Redfish experiments for two reasons: vessels tend to be closely grouped and not all VMS records have been obtained by polling. Even after interpolation, these positions tend to have a relatively large uncertainty. The automatic correlation produces multiple matches between several VMS and one VDS in such cases. Resolving these multiple matches leads to distinct correlated pairs (column 5 in Table 1). Unmatched VDS are listed in column 6. The same problem leading to multiple matches causes some VDS to remain 5

6 unmatched inside clusters of VMS and VDS positions, but most unmatched VDS are genuine vessels for which no matching VMS are available. The overall number of VMS and VDS vary with the location of the image acquisition. For instance, the image of July 8 (08:09 UTC) narrowly excludes a group of 17 VMS positions just to the west of the image frame. The overall vessel presence slightly decreases towards the end of the campaign, but is still considerably higher than observed in previous years, when vessel presence dropped steeply after July 1. A special case is the acquisition of June 28 (19:31 UTC, underlined in Table 1), which was made under very rough sea state conditions. The Icelandic Coastguard reported wind speeds of higher than 40 knots in the area. High sea state severely limits the quality of VDS detection results, as the backscattering signal from the vessels gets saturated by that of the rough sea surface. The number of VDS targets is accordingly low. One of unmatched VDS targets correlates with the position of a IUU vessel reported by the Icelandic Coastguard around the time of the image (the other 4 reported IUU vessels are not detected, however). Sightings by the ICG on other days are not as closely timed to the image acquisitions but confirm the frequent spotting of blacklisted IUU trawlers. Furthermore, a number of vessels in the area that are not carrying VMS are involved in fuel servicing of the fishing fleet and transshipment of catches. Detection and correlation results in Table 1 are typical for the experiments carried out in previous years (Lemoine et al, 2005b). In 2006, however, we have a more complete set of high quality VMS records for all fleets in the area. While this is a significant improvement, we still detect a relatively high number of unmatched VDS, confirming the observations of previous years where this was also the case. Table 1: VMS and VDS statistics for the Redfish 2006 experiment. Acquisition VMS VDS Correlations Distinct Unmatched VDS : (86%) 31(72%) 23 (43%) : (98%) 40 (87%) 15 (27%) : (78%) 32 (63%) 34 (52%) : (95%) 32 (76%) 15 (32%) : (100%) 28 (72%) 22 (44%) : (70%) 14 (33%) 3 (18%) : (100%) 27 (79%) 16 (37%) : (89%) 13 (72%) 22 (65%) : (87%) 22 (73%) 14 (39%) Total 346 (303) 403 (386) 307 (277) 239 (225) 164 (161) A qualitative assessment of VDS results is illustrated in Figure 2, which shows a time series of VMS and VDS results over the campaign period. In early June, all vessels are lined up along the Icelandic EEZ. In the last week of June, two distinct groups become apparent. The vessel flag state compositions of these 2 groups is rather distinct. Also note that the abundance of unmatched VDS is much more significant in the south-eastern subgroup (the one remaining aligned with the EEZ). This is confirmed with sighting from the Icelandic Coastguard taken in the month of June which place most of the IUU reports in this south-eastern group. 6

7 Figure 2: Time sequence of vessel traffic in the Redfish area as captured from VMS and VDS records. Top left is the acquisition of June, 11; top right June, 24; bottom left July 1 and bottom right July 11. In Figure 4 we present a zoom-in of the acquisition of June, 11 (cf. top left image in Figure 2) of the area along the Icelandic EEZ. The pictured situation is remarkable in two ways: the fleet appears to operate in a concerted regular spatial pattern with aligned groups of 3 trawlers and almost all VMS reports are matched with VDS detections. In fact, the average distance between matched VMS and VDS pairs is better than 300 m, which is very good. For comparison, we have reproduced an oblique aerial photograph from the Greenpeace report (Greenpeace, 2006), taken on 12 May, 2006, in Figure 3. The photo shows trawlers in the same area in a similar formation as detected in Figure 4. The vessel in front is the IUU vessel Eva (flying a Georgian flag) and one of the 8 IUU vessels that have been sighted as fishing in the formation. 7

8 Figure 3: Oblique aerial view of fleet formation in the Redfish experimental area taken on 12 May 2006 (Greenpeace, 2006). The vessel in front is the IUU vessel Eva. Greenpeace/Martin Norman. Figure 4: A zoom of the June 11 acquisition in the area straddling the Icelandic EEZ. The regular pattern of the fishing vessels is evident in both the VDS (circles) and the matching VMS positions (triangles).. 4. Summary and conclusions 8

9 We have demonstrated the practical use of VDS in a scenario in which we were able to support authorities in charge of surveillance operations with vessel traffic patterns that are useful to assess the occurrence of unrecognised vessels in a remote area of the EEZ. A number of these were supplied within 20 minutes of the image acquisitions from which they were derived. Our system has been designed in such a way that it can support VDS use in different operational scenarios, in particular near real time support to surveillance and large area monitoring of vessel traffic pattern over major fishing grounds. We believe that VDS can assist in the quantification of corrections to effort figures that are based on sampled information from regulated fisheries in order to estimate stock situations with more precision. This would require operational VDS to be taken up on a routine operational basis, for instance, as one of the future tasks of the European Union's new Community Fisheries Control Agency. Awareness of VDS capabilities and limitations is widespread amongst EU Fisheries Monitoring Centres, the majority of which have been actively involved in system development and (ongoing) demonstration. Staff from EU and international FMCs have attended dedicated VDS training at our facilities in early Various new projects will allow us to demonstrate VDS in remote area scenarios (Western African waters, Indian Ocean) in support of newly developing Monitoring and Control Systems in these regions. Renewed attention for issues related to maritime security are driving many new developments in maritime traffic monitoring and control. The open architecture of our system has allowed us to participate in a number of new projects aimed at maritime surveillance. Our system is already prepared for the use of diverse sets of positions reports, such as those derived from future spaceborne AIS monitoring, LRIT, etc. The near future will see the deployment of the second generation of spaceborne SAR sensors that will enhance and augment VDS capacities in 2 distinct ways. The TerraSAR-X and Radarsat-2 satellite sensors, both to be launched in 2007, will have enhanced spatial resolution (at the cost of area coverage, however) and include improved options for polarimetry. The Indian RiSAT (2008) is an enhancement of the Radarsat-1 instrument, though not as advanced as Radarsat-2. Other SAR missions (e.g. CosmoSkyMed, SAOCOM) will extend capabilities to multi-platform constellations, which will greatly improve area revisit time and bring capabilities to civilian SAR data use that were previously restricted to security applications (e,g. queueing).wider availability of a more diverse range of SAR products will also bring the much needed market pressures to lower prices of the base image products for vessel detection. 5. Acknowledgement The original VDS was developed in projects supported by the European Commission s R&D Framework Programme 5 under contracts Q5RS IMPAST and EVG2-CT (DECLIMS).. We kindly acknowledge the managers and operators at the various Fisheries Monitoring Centres who have contributed VMS records to support our work. 6. References Greenpeace, 2006, Caught RED-handed: Daylight Robbery on the High Seas, Greenpeace Case Study on IUU Fishing #3, 12 p., May

10 H. Greidanus, Applicability of the K-distribution to RADARSAT Maritime Imagery, 2004, In: Proc. Int. Geosci. Rem. Sens. Symp. (IGARSS 04), 2004, Anchorage, Alaska, September ICES, 2006, Report of the North-Western Working Group (NWWG), 25 April - 4 May 2006, ICES Headquarters. ICES CM 2006/ACFM: pp. N. Kourti, I. Shepherd, G. Schwartz, P. Pavlakis, 2001, Integrating Spaceborne SAR imagery into Operational systems for Fisheries Monitoring, Can. J. Rem. Sens. 27, No. 4, pp , N. Kourti, I. Shepherd, H. Greidanus, M. Alvarez, E. Aresu, T. Bauna, J. Chesworth, G. Lemoine, G. Schwartz, 2005: Integrating Remote Sensing in Fisheries Control, Fisheries Management and Ecology 12, G.G. Lemoine, 2005a, Vessel Detection System, a Blueprint for an Operational System, Technical Note I.05.14, European Commission, Joint Research Centre, p. 37, February G.G. Lemoine, H. Greidanus, I.M. Shepherd and N. Kourti, 2005b, Developments insatellite Fisheries Monitoring and Control, 8 th International Conference on Remote Sensing for Marine and Coastal Environments, Halifax, Canada, May