The state-of-the-art in both operational oceanography, remote sensing, and computational capacity, enables now the possibility of developing near-real time, holistic automated services capable of dramatically improving maritime situational awareness to responding to oil spill emergencies.
Based on the European satellite-based oil spill and vessel detection service – CleanSeaNet (EMSA – European Maritime Safety Agency), which distributes oil pollution detection standardized notification packages in less than 30 minutes, a new automated early warning system (EWS) for near-real time modelling and prediction of the detected oil spills was developed. This EWS provides 48-hour oil spill forecasts + 24-hour backward simulations, delivering results 5–10 minutes after the reception of the oil spill detection notifications. These forecasts are then distributed in multiple formats and platforms (e.g. Google Earth, e-mail).
The oil spill fate and behaviour model used in this EWS is part of MOHID modelling system, and is coupled offline with metocean forecast solutions, taking advantage of autonomous models previously run in multiple institutions. The system is currently able to integrate various metocean forecasting systems, being agnostic about the data sources and applied locations, as long as their outputs comply with commonly adopted formats, including CF compliant files or CMEMS (Copernicus Marine Environment Monitoring Service).
The EWS is currently operational in western Iberia, supporting Portuguese Maritime Authority, and is being expanded to neighbourhood regions (from Spain and Morocco) with high resolution metocean models (MARPOCS project funded by European Union Humanitarian Aid & Civil Protection). Taking advantage of the coupling of MOHID oil spill model and CleanSeaNet, an oil spill hazard assessment is made in the Portuguese continental coast, based on the cumulative analysis of drift model simulations from previously detected spills using metocean model data, for a period between 2011–2016.
Although this EWS doesn’t replace on-demand operational oil spill forecasting systems, it supports maritime authorities with a fast first-guess forecast solution, allowing:
Anticipation of tactical response (including visual inspection of the spill) and mitigation of the pollution episode;
A more effective identification of the pollution source, and in case of suspected illegal spill, earlier actions towards effective prosecution of the polluter;
In the other hand, the hazard assessment generated is a valuable instrument for the development of efficient planning and prevention strategies.
The EWS can be connected to any satellite-based detection service (inside or outside Europe) as long as the detected oil slicks are automatically distributed in a structured and standardized data format similar to CleanSeaNet.
Oil spills remain a serious environmental risk, as the livelihood and the marine ecosystems can be considerably affected in the event of a significant incident (ITOPF, 2011). The incidence of maritime pollution with oil spills is frequently linked to ship’s incidents. Large spills from tankers and oil industry operations have become less frequent in the last few decades. Similarly, the huge number of illegal, small spills are being progressively controlled and reduced. International conventions have provided important preventative measures. In fact, the vast majority of spill incidents are associated to small amounts of released oil (ITOPF, 2015). But the severity of a spill is not only connected to the amount released - for instance, the remoteness of the site or the difficulty of an emergency environmental response are factors that can significantly increase the impact.
In the last years, new ocean and coastal forecasting capabilities have been achieved and resulted in stabilized operational services (e.g. CMEMS - Copernicus Marine Environment Monitoring Service), being now daily accessed and used by several intermediate and final users and downstream services - remote sensing is part of the mentioned operational services. Remote sensing technologies can play an important role in the early detection of these kind of incidents, since Earth observation services are now able to provide realtime and near-real time data, shortly processed, easily accessed, and oriented to multiple integrated maritime services. EMSA’s (European Maritime Safety Agency) CleanSeaNet is a good example of a near-realtime remote detection monitoring service. Launched and maintained since 2007, this is an important pan-European operational system able to provide detected oil spill notifications (based on SAR - synthetic aperture radar images) to the EU coastal states in less than 30 minutes (Figure 1 ).
Following a regional approach, cooperation agreements were established between CleanSeaNet and operational institutions working closely with national maritime authorities in different regional seas, in order to promote the development of oil spill modelling services able to link to CleanSeaNet. A regional sea approach to oil spill modelling can take advantage of opportunities for cost efficiencies from sharing operational and maintenance costs of modelling. The technical interfaces developed for data exchange enable data on oil spills detected by satellite to be sent to the regional models and to be available in their systems, and for appropriate modelling results to be returned to CleanSeaNet. In addition, having a designated model per regional sea, or group of models per regional sea, could reduce the cost of developing, maintaining and operating a model by each EU coastal Member State separately.
In this context, two operational services were implemented, in the Baltic Sea (Seatrack Web – operated by Swedish Meteorological Hydrographic Institute) and more recently, in the North Sea (OSERIT – Oil Spill Evaluation and Response Integrated Tool – operated by Royal Belgian Institute for Natural Science). A demonstration in the Mediterranean Sea (using MEDSLIK model – developed by Cyprus Centre for Oceanography - CYCOFOS) was also developed.
In a similar approach, the work presented in this paper is relative to an initial demonstration of the link between CleanSeaNet and MOHID oil spill modelling services (hereinafter called MOHID-CSN service) in the Atlantic Ocean – on Portuguese continental coast, and its neighbourhood regions in Spain - Andalucia and Galicia (see Figure 2 ).
The MOHID-CSN service is a result of a close and fruitful cooperation between Portuguese Maritime Authority / Response Pollution Service (DGAM-SCPM), Action Modulers, MARETEC-IST (Marine, Environment & Technology Centre from Instituto Superior Técnico), EMSA, and Spanish Maritime Safety Agency (SASEMAR), in the context of multiple European projects implemented in the Atlantic region – ARCOPOL+ (“Improving maritime safety and pollution response through technology transfer, training & innovation”); ARCOPOLplatform (“Atlantic Regions Coastal Pollution Response”); (ARCOPOL project website; both projects funded by European Union INTERREG Atlantic programme), and presently, MARPOCS (“Multinational Response and Preparedness to Oil and Chemical Spills”) - MARPOCS project website) and MARINER (“Enhancing HNS Preparedness through training and exercising”. MARINER project website; both projects funded by DG-ECHO – Humanitarian Aid and Civil Protection from European Commission), always in cooperation with regional and national authorities.
The area of initial implementation is subject to strong coastal pressures in different aspects. The area has a strong economic dependence of the sea, manifested in the most diverse ways, whether it’s fishing, leisure activities, water sports, coastal tourism, or employment associated with the sea economy. This coast is crossed by important international shipping routes with most of the maritime traffic circulating to and from Northern Europe (view Figure 2 ), increasing the responsibility in terms of safety of navigation as well as in preventing and combating marine pollution. Figure 3 shows the CleanSeaNet detected oil spills in the Portuguese Economic Exclusive Zone (EEZ) since the beginning of the service (2008) till September 2016 (approximately 500 spills). It is visible the high concentration of detections (black dots) in the Portuguese Coast, which reflects the dimension of the problem in this area. Some of these detections have been also reported from fishing vessels, or from land sources (e.g. emergency discharges from wastewater treatment plants), however no exact data exists about the number of shore-based spills.
The development of automatic service-to-service (S2S) early warning spill forecasting systems (back and forth in time) connected to existing maritime surveillance automatic services in the referred areas of interest can be seen as a relevant effort to enhance preparedness and response mechanisms when facing oil spills, and contributing to the minimization and prevention of coastal contamination risks. If the developed systems are able to work with common, standardized, versatile and open-source technologies, the transferability of those operational systems to any other areas will be greater. Hence, the conceptual design adopted for the system here developed is totally based in the premise that it must be, as much as possible, transferable to any other geographical areas, possibly using any other remote sensing data sources, as well as other metocean modelling systems.
The development of a system with those conditions requires an open-source, versatile, reliable and efficient oil spill model, as well as an appropriate and scalable operational software framework capable of supporting and digesting multiple input data sources, from different data providers. Thus, the oil spill model adopted is MOHID, and the operational software framework used to manage the data is ACTION Server.
Oil spill model
A lagrangian particle transport model (MOHID) is used to compute oil spills at sea in the automatic simulation of the trajectory of detected oil spills from CleanSeaNet service. The model simulates the trajectory forward and backward in time. In lagrangian models, simulated pollutants are represented by a cloud of discrete particles (or super-particles) advected by wind, currents and waves, and spread due to random turbulent diffusion or mechanical spreading.
The MOHID lagrangian module (Fernandes, 2013, Ascione Kenov, et al., 2014) can be run simultaneously with the hydrodynamic model (currents, water temperature, salinity, etc.), or in “offline” mode – the mode adopted in this work, for operational reasons. The model is able to digest currents, water properties, wave parameters and atmosphere properties from different model providers. Additionally, MOHID lagrangian module allows backtracking modelling, as well as a multi-solution approach (Fernandes, 2013) (generating computational grid on-the-fly, and using the available information from the multiple metocean forecasting solutions available). The user interaction with MOHID lagrangian module is presently available through many different web and desktop platforms (Fernandes, 2013), and the most recent is ACTION Seaport (ACTION Seaport Demo website); (Fernandes, 2016).
The MOHID oil spill fate and behaviour model was initially developed in 2001 (Fernandes, 2001) and has been operationally applied in different incidents (Carracedo et al., 2006; Janeiro et al., 2014), field exercises and studies worldwide, allowing the simulation of major oil transport and weathering processes at sea. The source code of oil spill modelling system was recently updated to include full 3D movement of oil particles, wave-induced currents, and oil-shoreline interaction (Fernandes et al., 2013).
The central point of software framework developed is the operational software ACTION Server. This software works as the “heart” of the whole system, collecting, processing and storing all the input data (numerical models, remote sensing data from CleanSeaNet), interacting and operating MOHID, post-processing the information and then distributing the output data (including alerts and notifications).
ACTION Server is presently a plugin based, server system composed by a set of software components allowing to integrate and process different data sources and multiple numerical models. These software components can be installed as plugins, allowing the system to be scalable, depending on the end user needs. Several plugins are currently integrated to download data automatically from global, regional and local systems (e.g. NOAA GFS; CMEMS Global Forecast Solutions; MOHID, WW3, MM5, and WRF model outputs provided by MARETEC-IST and Meteogalicia; WW3 and SWAN models provided by Puertos del Estado; real time acquisition from the meteorological network Weather Underground).
This software is currently used in production mode as backbone for ACTION Modulers’s flood early warning system (ACTION Flood), sea port operational support system (ACTION Seaport) and operational bathing water forecast system ACTION Beach (ACTION Beach description; ACTION Beach Romania).
Metocean modelling conditions
The oil spill model simulations take into account forecasted environmental field conditions that can influence the fate and behaviour of oil at sea. In the initial implementation presented here, the MOHID-CSN EWS digests metocean forecasted parameters generated by validated numerical models developed by MARETEC-IST. Using ACTION Server, the integration with any other data sources (either implemented by default in ACTION Server, like GFS or CMEMS global forecasting system, or using any other external data provider) is possible and straightforward, as long as the operational model results are published online in a catalogue web server (e.g. FTP, HTTP, THREDDS / OPENDAP), and the model output files are stored in a typical numerical data format (netCDF or HDF).
Currents and water properties (sea surface temperature, sea surface salinity and suspended particulate matter) are obtained from hourly outputs from PCOMS-MOHID model (Mateus et al., 2012, Pinto et al., 2012). PCOMS (Portuguese Coastal Operational Modelling System) is a 3-D hydro-biogeochemical model of the Iberian western Atlantic region. Ocean boundary conditions are provided by the Mercator-Ocean PSY2V4 North Atlantic and by tidal levels computed by a 2-D version of MOHID (Neves, 2013; Ascione Kenov et al., 2014), forced by FES (Finite Element Solution) global tide model in the version FES2004, and running on a wider region. PCOMS has a horizontal resolution of 6.6 km and a vertical delineation of 50 layers with increasing resolution from the sea bottom upward, reaching 1 m at the surface (Ascione Kenov et al., 2014).
Atmospheric conditions (wind velocity, surface air temperature, atmospheric pressure and visibility) are obtained from the meteorological forecasting system IST-MM5, using MM5 model (Grell et al.1994) with a 9km spatial resolution. This operational model was initially implemented by Sousa, 2002, and updated in 2005 (Trancoso, 2012). This model is also used as atmospheric forcing of PCOMS-MOHID, and its outputs are also every hour.
The wave parameters (wave period, wave height, wave direction and wave length) are obtained from the Portuguese wave forecasting system implemented at MARETEC-IST, using the hourly outputs from WaveWatchIII model (Tolman, 2009) with a 5km spatial resolution, and wind forcing provided by Global Forecasting System (GFS) from the National Oceanic and Atmospheric Administration (NOAA), with a spatial resolution of 0.5° (Franz et al., 2014).
The system is continuously monitoring EMSA’s FTP server where any CleanSeanet oil spill notification is stored. The oil spill notification is downloaded immediately, uncompressing it and checking the geographical area of the spill. Depending on the area, a different mailing list is used to disseminate the results (for instance, if the oil spill is in Spain, SASEMAR contacts are notified). The whole process between the CleanSeaNet notification stored at FTP Server, to the mail distribution, takes an average time of 5 to 10 minutes.
The operational system prepares then the model input data files, by a) changing the model configuration files, b) generating polygons for the oil slick positions in a MOHID-readable fashion, and c) fetching the needed metocean forecasts to properly simulate the oil spill forward (24 hours) and backward (24 hours) in time. The model simulations usually take less than 5 minutes (this value can be higher if the spill has multiple slicks).
After the end of the model run, post processing actions are taken to convert MOHID model results in multiple formats adequate to be distributed to the oil spill responders. Conversions include common GML / XML formats to be uploaded and readable in EMSA web GIS; a standardized numerical data format (netCDF) for the same purpose; and KML format - to be loaded directly in Google Earth. MOHID native formats are also provided, to allow end-users to directly open it in MOHID Studio desktop GIS.
The last stage of the process is the distribution / uploading of the model results in the multiple platforms: a) EMSA’s FTP server; b) MOHID-CSN EWS central web server; c) email notification sent to the proper responders, with meta-data of the forecasted spill and hyperlinks to visualize results hosted in the central web server (Figure 4 ).
Since no data is available related to the exact oil product, its weathering stage and if oil is present in the water column, different generic and common assumptions are taken in the configuration of the oil spill model. By default, a generic crude oil (API - American Petroleum Institute - gravity of 30, viscosity of 39 cP, oil-water interfacial tension of 20 dynes/cm) is assumed to be released on the surface exactly in the same polygon locations(s) as identified by CleanSeaNet. Each polygon is simulated independently. It is assumed that each identified polygon area corresponds to slick area from the end of the first spreading stage (gravity-inertial phase, that usually lasts only a few minutes; Fay, 1969). The volume or mass is then computed based on that assumption. It is assumed that the oil is fresh and all the typical oil transport and weathering processes are computed (mechanical spreading, evaporation, natural dispersion, emulsification, adsorption to sediments, dissolution, Stokes drift, turbulent diffusion, wind-driven velocity represented by 3% of the surface wind velocity). Assuming that most of the detected oil spills are vessel-borne and with oil products with a density lower than the seawater, it is considered that the initial oil slick is only present at surface (therefore, no initial oil is considered in the subsurface). An internal time step of 60 seconds is used in oil weathering computations; 120 seconds is the time step used in the computation of remaining lagrangian processes. A total of 500 particles are released in every polygon simulation, and the computational mesh used is composed of 500×500 orthogonal cells with 300m of spatial step, centred in the mass centre of the detected oil slick.
MOHID-CSN EWS: Examples and performance
The MOHID-CSN EWS has been tested in operational mode since 15-4-2016, distributing notifications to SCPM-DGAM. During this test period, the system has already proven its usefulness on different occasions – mainly because the automatic system is able to generate near-realtime oil spill forecasts in user-friendly outputs (e.g. Google Earth), accelerating the usability of the model forecasts when compared to the traditional on-demand modelling systems used by response operators. The first example is related to an incident in the centre of Portugal, in 26-4-2016. SCPM-DGAM received the initial oil spill notification from CleanSeaNet at 19h47, relative to a supposed oil spill identified at 19h29 with 1.47 km2, offshore of Aveiro. The MOHID-CSN EWS forecasts were duly distributed by email at 20h02 - in less than 15 minutes after initial CleanSeaNet notification (see Figure 5 ).
At 20:41, an email from MRCC (Maritime Rescue Coordination Centre), reported a correlation with the ship BITTEN THERESA (which had been questioned), and a drift forecast was requested by MRCC to the Hydrographic Institute (IH). At 23:22, IH sent the official forecast, with results similar to the MOHID-CSN notification. A response from the vessel was sent to MRCC at 26-4-2016 10h36, confirming that at 18h30, at the indicated location, tank cleaning had started, reporting that the last load was soybean oil.
In summary, the MOHID-CSN EWS forecasts were able to provide the fastest forecasts to this incident, with the results being comparable to the available on-demand forecasting systems. MOHID-CSN EWS was used as a first guess decision support system, allowing for faster identification of the potential polluter.
The second example is relative to an incident in 14-12-2016 (spill instant: 06:41), related to an operational / emergency discharge in a sensitive coastal area from a wastewater treatment plant (Guia outfall, in Tagus Mouth / Cascais), as a result of unusual water flow income due to pluvial waters in a strong raining event. In this case, the DGAM-SCPM was previously advised of the need for an emergency discharge, and was able to immediately prepare an observation team to follow the oil slick at sea. The MOHID-CSN EWS forecasts (see Figure 6 ) were duly distributed by email in less than 8 minutes after initial CleanSeaNet notification upload, and could be used by DGAM-SCPM to improve the tracking of the slick at sea (narrowing the search area), and to efficiently control if any nearshore resource would be potentially affected. Once the spillage happened (and slick detected), the main advantage here in relation to traditional on-demand systems is that the user wouldn’t need any type of intervention to generate the result.
At the present moment, the same system is currently being tested in operation mode for Andalusian and Galician regions (Spanish coastal regions in the vicinity of Portuguese continental coast), since 10-10-2016. During the testing period, corrections mainly related with operational modelling procedures and logistics are being performed (e.g. improvement of redundancy in terms of metocean forecasts; adjustment and optimization of the compromise between model domain sizes, computational time steps and precision of the results; correction of minor bugs when converting CleanSeaNet original polygons to MOHID model format, among others). As the oil spill detections are usually unpredictable and not so frequent, this testing and optimization phase is longer than the usual time involved in testing operational systems running in a daily-basis.
Oil spill hazard analysis
The coupled MOHID-CSN system has also been applied in a historical analysis covering a long-term period (5 years between September 2011 and September 2016) for the Portuguese continental coast (i.e., using the oil spills detected in the Portuguese continental coast), with the aim of building a quantitative and representative oil spill hazard map of the Portuguese continental coast during the last 5 years. All the oil spills detected inside the area of study during the selected period (155 detected spills after filtering area and time period), were simulated in the same conditions as defined in the operational system (MOHID-CSN EWS) previously described. The geographical position and mass of all the lagrangian particles were recorded every hour and then integrated in time, allowing to obtain a single 5-year map representing the oil concentrations for the whole period. Higher concentrations represent greater probability of oil contamination. A seasonal comparison between spring + summer (21-March till 21-September) and autumn + winter (21-September till 21 of March) was also performed.
In the annual integration (Figure 7a), the pollution off the coast is demonstrated in the vicinity of the typical international shipping routes crossing Portuguese coast (Figure 2 ), particularly, near the Mediterranean Sea, in the boundary with Spain. Higher probabilities of nearshore contamination in the centre and north of Portugal is also registered. The southern boundary of Portugal (Algarve) is also identified as a “hotspot” location, but only in the western side, while the nearshore coast of Alentejo (west coast, in the south) has low probability of contamination – eventually due to a combination of the oil spill locations (usually far from the coast of Alentejo) and favourable metocean conditions along the year (offshore wind and currents). When analysing Figure 7b) and Figure 7c), higher probabilities of contamination can be found in summer, in a general way. Although specific reasons could not be clearly identified in the scope of this work, several hypotheses can be discussed (in the next section) and evaluated in further research work.
The MOHID-CSN EWS developed and being operationally tested in an Atlantic region is in line with the EMSA strategy of contributing to the setting up, implementation and promotion of integrative decision support systems in EU Member States, supported by the intelligent combination of the existing surveillance capacity in EMSA (e.g. CleanSeaNet) with the regional and national operational metocean forecasting systems used by the national maritime authorities. The system here presented is now being distributed to Portuguese and Spanish National Maritime Authorities involved in oil pollution monitoring and surveillance. In the future, the system will be extended to other areas (and corresponding national or regional maritime authorities) in the Atlantic (Morocco, Canary Islands, Madeira Islands, and eventually other regions in Spain).
The MOHID-CSN EWS represents a step-up in the oil spill pollution preparedness & response in the Atlantic coast (handled by the corresponding maritime authorities in articulation with EMSA), proving its ability to reduce the time needed for the identification of the potential polluter (using backtracking modelling capacity), and being able to work as a credible first guess solution. Although not replacing the existing on-demand simulation tools, MOHID-CSN EWS works as an additional source of modelling information, contributing to the reduction of the uncertainties, inherent to the modelling activities.
From a developer point a view, since this operational system is all sustained by a set of flexible software tools (ACTION Server; MOHID oil spill model), there is a true potential of implementing this MOHID-CSN EWS in any region in EU (with CleanSeaNet covering the whole region) or worldwide, using different configurations and data sets: either using metocean forecasts from different data sources, or using other oil slick polygons detected by different Earth observation services (other than CleanSeaNet). This abstraction in relation to the data sets used represents an innovative and strategic development in terms of expanding the adopted methodology to any other areas. The implementation of this automated system using other kinds of software (e.g. using scripts and other oil spill model) is also potentially possible - as long these will not compromise the overall performance and reliability of the EWS.
The oil spill hazard mapping analysis raised some important issues that need to be addressed in future work. In terms of seasonal differences, hypotheses like false positives and illegal oil discharges from greater nearshore fishing activity in summer; false negatives and land-based emergency discharges from wastewater treatment plants in winter should be properly evaluated and studied. In addition, a comparison between this oil spill hazard mapping analysis based on oil slick detections, and a hazard mapping based only in vessel traffic routes could clarify some of the aspects identified in this work, for instance, a better identification of the relative weight of ship-borne pollution in the general vulnerability of oil spill contamination (i.e. if both methods for hazard mapping provide similar results, it means that ship-borne pollution must be the most important source of oil spill contamination).
Also, the verification of new techniques to detect oil spills with recently launched satellite SAR instruments (Copernicus Sentinel 1A / 1B), and the fusion of SAR data with high resolution ocean colour data (using for instance Copernicus Sentinel 3A) will contribute simultaneously to an even more efficient detection of oil spills and to the reduction of false detections.
The authors would like to acknowledge the support and assistance of Anne Marie Hayes and Gianluca Luraschi (EMSA). The authors would also like to thank Rosa Trancoso for the development of the atmospheric model used, and Jorge Palma for its operational maintenance. A special thanks to DGAM-SCPM for all the cooperation, data sharing and promotion of the system development and implementation, as well as supporting the beta-testing. We also acknowledge SASEMAR for its cooperation in beta-testing.
This work has been partially sponsored by projects MARPOCS and MARINER, both co-funded by the European Union in the framework of the Union Civil Protection Mechanism, DG-ECHO, European Commission.