1 Introduction

1.1 What Is Seatrack Web?

This chapter deals with Seatrack Web, an online forecasting system developed for predicting the movements of primarily oil spills at sea, but which can also be used for objects lost overboard from ships, algal blooms, dissolved substances, and various other substances or objects. It is an online tool that can not only be used in real-time forecasts during emergencies but also for investigations after the fact; so-called hindcasts. End users are typically authorities and organizations responsible for oil spill and pollution combating, marine forecasting, and marine environmental monitoring.

The system was originally developed in collaboration between the Swedish Meteorological and Hydrological Institute (SMHI) and the Danish Maritime Safety Administration (DaMSA) for use in the Baltic Sea and the eastern part of the North Sea, as the result of a HELCOM recommendation (see below). However, Seatrack Web has now been set up for several different locations on different scales, ranging from a small fjord on the Swedish west coast to the Black Sea. As a result, the continued development of the system and the underlying code is now carried out not only at SMHI and DaMSA but also at the Federal Maritime and Hydrographic Agency in Germany (BSH) and at the Marine Hydrophysical Institute in Ukraine (MHI), the latter on behalf of the Black Sea Commission within the framework of the MONINFO Project. In addition, the program code has also been supplemented with weathering algorithms developed and supplied by the Norwegian independent research organization SINTEF. However, as the focus of this book is the Baltic Sea, this chapter will describe the original Seatrack Web system implemented for the Baltic Sea and the eastern part of the North Sea – also referred to as HELCOM Seatrack Web – although the other operational systems are very similar.

The remaining parts of this section will deal with the history of Seatrack Web and the reasons for its existence. In Sect. 2, an overview of the three major parts of the Seatrack Web system is presented. Each part is then described in more detail in Sects. 3–5. This technical part of the chapter is followed by a section on the usage of and the results produced by Seatrack Web (Sect. 6). In Sect. 7, future developments and possibilities are discussed and the chapter ends with some general conclusions in Sect. 8.

1.2 Seatrack Web and HELCOM

Following the designation of the Baltic Sea area as a “special area” under the Annex I of the International Convention for the Prevention of Pollution from Ships (MARPOL 73/78), any discharge into the Baltic Sea of oil, or diluted mixtures containing oil in any form including crude oil, fuel oil, oil sludge, or refined products, is prohibited. This applies to oily water from the machinery compartments of any ship, as well as from ballast or cargo tanks in oil tankers. Despite the strict legal regime, almost 600 illegal discharges were observed in the Baltic Sea in the interval 2003–2004. The real number of discharges is considered to be even higher. Most of the observed illegal discharges are smaller than 1 m3 but around 8% are larger, sometimes exceeding 100 m3.

The illegally discharged oil has a number of negative effects including the killing of seabirds and the pollution of shores and beaches.

In a vast majority of cases of detected illegal discharges into the Baltic Sea, the polluters remain unknown. In 2006, out of the total number of 236 confirmed illegal discharges, the polluters were identified in only 18 cases. Therefore, strong enforcement of anti-pollution regulations within the HELCOM area is necessary and Seatrack Web can be an important tool in this respect.

Seatrack Web was developed following HELCOM Recommendation 12/6 of February 20, 1991, and is regarded as the common HELCOM modeling and drift forecasting system for oils and chemicals. At the HELCOM Response meeting of October 2002, it was recognized that the system at this time needed upgrading to meet the requirements of a new HELCOM recommendation (Recommendation 24/7) addressing further development and use of drift forecasting for oils and other harmful substances in the Baltic. In addition, various users of Seatrack Web had particular requests, such as to include more oils relevant for the Baltic Sea area, to add predictions of viscosity changes for use in clean-up operations, to model the interaction between oil and ice, extended GIS functionality, etc.

Together with further input given during meetings and workshops, as well as from the expert group developing the system, an extensive specification for a new version of Seatrack Web was established. An extensive upgrading of the system commenced in 2004 and resulted in a new version, Seatrack Web 2.0. Since then the system has been under continuous development and improvement, and this work is ongoing.

2 System Overview

The Seatrack Web system consists of three main parts:

  1. 1.

    Input data and forcing

  2. 2.

    The drift model

  3. 3.

    The client/server web application

The input data and forcing consists of two types of data: forecasts of meteorological and hydrodynamic conditions produced by forecast models, and other input data, such as maps, ship tracks from Automatic Identification System (AIS), satellite data, etc. The forecasts are necessary, whereas the other input data are not but greatly increase the usefulness of Seatrack Web. The forecasts are produced regularly as part of a separate operational forecasting process and made available to the drift model, i.e., they are already present when required and need not be produced on demand. The other input data are similarly delivered regularly and available at all times, or part of the system setup.

The drift model is the part of the system that computes the actual movement and fate of an oil spill, an object or another substance. It is executed on demand, i.e., when requested by a user for a specific case. It takes input from the user via the client/server web application, runs a simulation based on the meteorological and hydrodynamic forecasts for the period in question, and outputs results which are read and presented to the user by the client/server web application.

The client/server web application consists of the user interface (the client) and a server that handles requests from users, starts simulations using the drift model and makes the results available to the users. The client is in the form of a Java program that runs on the user’s computer and which uses a GIS-based interface. The client and the server communicate via the internet.

The three parts of the system and the data flow is sketched in Fig. 1.

Fig. 1
figure 1

An outline of the Seatrack Web system, showing the three main parts and the data flow

Full size image

HELCOM Seatrack Web is currently in operational use both at DaMSA and SMHI, with separate but almost identical setups of the drift model and client/server web application.

3 Input Data and Forcing

3.1 Meteorological and Hydrodynamic Forcing

The minimum forcing data required to perform a drift forecast is the three-dimensional current field for the forecast period. In HELCOM Seatrack Web, this is provided by the operational ocean forecast model HIROMB (High Resolution Operational Model for the Baltic Sea). This model covers the Baltic and North Seas, and comes in two versions: one with horizontal resolution three nautical miles covering both the Baltic and the North Seas, and one with horizontal resolution one nautical mile covering the Baltic Sea and its entrance areas (see Fig. 2). The model presently has 50 layers in the vertical with a maximum resolution of 4 m. HIROMB also includes an ice model and is run four times a day at SMHI, producing forecasts of current velocities, turbulence, sea levels, salinities, temperatures, and ice conditions.

Fig. 2
figure 2

Map showing the approximate extent of the HIROMB model areas for the three and one nautical miles versions, respectively

Full size image

The hydrodynamic model HIROMB is in turn forced by meteorological forecasts. Currently, two versions of HIROMB are run: one using meteorological forcing from the HIRLAM model (High Resolution Limited Area Model) run at SMHI and producing 48 h forecasts, and one using five day forecasts from ECMWF (European Centre for Medium-Range Weather Forecasts). Hence, HIROMB produces forecasts for both the coming 48 h and the coming five days. The short term forecast is available for both one and three nautical miles resolution, whereas the five day forecast is only available for three nautical miles resolution.

HELCOM Seatrack Web also uses the meteorological forecasts for the wind directly. The wind is used to calculate the near-surface profile of current velocities, which will be further described below. In addition, for objects that extend above the surface it is possible to add a wind drag by specifying what percentage of the wind velocity should be added to the drift velocity of the object.

The forcing input is automatically preprocessed as soon as a new forecast is available, producing binary files in the internal format of the Seatrack Web drift model for fast access. These files are stored for a limited time, such that forcing data is available as far back as approximately one month before the current date (this varies slightly depending on the version of the HIROMB model). Hence, calculations can be made within a time window stretching from about a month before the current date to at most five days into the future. Forcing input from earlier dates can be made available upon request.

The preprocessing also produces input files that are used by the client/server web application for presenting the surface current velocities, the wind velocities, and the ice concentration in the graphical user interface, as well as metadata regarding the availability of forcing data.

3.2 Other Input Data

There is obviously a great deal of other input data used in the Seatrack Web system. Necessary data are the bottom topography (HELCOM Seatrack Web currently uses the bathymetry of the HIROMB model) and a detailed coastline, but many other types of geo-referenced data, such as sensitive areas, potential polluters, reference points, etc., can easily be added to the user interface as it is based on a GIS-platform.

Here we will focus on two types of additional input data that show how the usefulness of the Seatrack Web system has been extended. They are ship tracks based on AIS data and detected oil spills extracted from satellite images.

3.2.1 AIS Data

Since Seatrack Web can be run backward in time, it is possible to backtrack and determine the origin of a detected oil spill. By integrating information from AIS, the results from a backtracking simulation can be superimposed on ship tracks, hence aiding in identifying the polluter.

The AIS data are continuously imported into Seatrack Web and can be displayed as ship tracks. Thus, the information on the whereabouts of vessels is available in real time. The AIS data are stored for one month, i.e., ship tracks can be viewed as far back as one month before the current date.

Incorporating AIS data into Seatrack Web has proved to be a very effective tool, substantially increasing the possibilities to identify ships suspected of illegally discharging oil into the sea. This feature is available to the relevant competent authorities in all HELCOM countries. Training on its use has been performed in Denmark, Estonia, Finland, Lithuania, Poland, Russia, and Sweden.

3.2.2 Satellite Data

Satellite data are currently not used directly in Seatrack Web but as part of a system for monitoring of algal blooms. From satellite images, algal blooms can be detected. Their extent is exported to a text file containing positions which can be loaded into Seatrack Web and plotted on the map. A forecast for the drift of the bloom can then be simulated using the imported positions as the initial positions of particles in Seatrack Web. Alternatively, the user can draw an area around the displayed positions and use this as the initial algal bloom location.

Functionality is being developed for directly importing information on oil spills detected using satellite imagery (see Sect. 7).

4 Drift Model

The Seatrack Web drift model is built around a Lagrangian particle tracking code called PADM (Particle Advection and Dispersion Model). This means that the substance whose drift and fate is being forecasted is represented as a cloud of particles. Hence, in the case of an oil spill, each particle will represent a part of the total mass and volume of the oil. Every particle is tracked individually in three dimensions. The processes affecting the fate of the particles can be divided into two categories:

  1. 1.

    Spreading, i.e., how the particles are moved by the surrounding flow field.

  2. 2.

    Weathering, i.e., changes in the particle properties due to substance-specific processes.

Spreading is not only the passive advection due to the currents but also includes random turbulent motions, vertical dispersion of oil caused by breaking waves, the initial gravity-induced radial spreading of oil slicks and vertical motion due to differences in buoyancy. Thus, the spreading processes are, in the case of oil, influenced by the weathering processes when these alter properties such as the density or viscosity of the oil.

Currently only oil weathering is included but in principle many other biochemical processes, such as the decay of a chemical, could be included.

4.1 Spreading

PADM assumes that the flow field is defined in a structured grid, i.e., a three-dimensional mesh of boxlike cells, each with six faces and eight vertices. The flow vectors are defined on the faces of the cells, whereas scalar properties such as salinity and temperature are defined in the cell centers. Assuming that the x-component of the velocity only varies in the x-direction, the y-component only in the y-direction, etc., it can be shown that the particle will follow a well-defined streamline, as long as the flow field does not change with time and the particle remains in the box [1,2]. This so-called passive advection can be modified by the following spreading processes:

  1. 1.

    Turbulent mixing

  2. 2.

    Stokes’ drift

  3. 3.

    Horizontal gravity-induced surface spreading

  4. 4.

    Vertical dispersion by breaking waves

  5. 5.

    Buoyancy-induced vertical movements

  6. 6.

    Ice drift, wind drag, and boundary interaction

4.1.1 Turbulent Mixing

The turbulent mixing is calculated by adding a random turbulent velocity whose magnitude is determined by the turbulent intensity at the point where the particle is located. The turbulent intensity is defined by the turbulent kinetic energy and its dissipation rate, both of which are calculated by the HIROMB model as part of its turbulent closure scheme. The turbulent particle velocity is calculated from the turbulent intensity using a Markov chain model [1,3]. This means that the turbulent velocity is not completely random but depends on the value from previous times, i.e., the turbulent velocity has a “memory” in proportion to the time scale of the turbulent eddies in the flow field. For long time intervals, the correlation with previous values goes to zero. The model also takes into account gradients in the turbulent intensity.

Note that the turbulent mixing is only relevant for particles that are suspended or dissolved in the water column. It does not account for the more large-scale horizontal spreading of a cohesive oil slick at the surface. Horizontal surface spreading of an oil spill is initially after a discharge caused by gravity-induced spreading (see Sect. 4.1.3) but is later on dominated by small scale variations such as horizontal eddies, Langmuir circulation, wind gusts, etc. One process that is included in PADM is the combined effect of vertical dispersion (see Sect. 4.1.4) and a vertical current shear (see Sect. 4.1.2). The result is that buoyant oil droplets that are dispersed to different depths will experience different current velocities and thus resurface in different positions, producing a horizontal spreading.

4.1.2 Stokes’ Drift

The Stokes’ drift is an important mechanism contributing to the surface wind drift, i.e., the near-surface drift due to the wind. Depending on the wind speed and the wave spectrum, a straightforward calculation of the Stokes’ drift yields a drift speed of somewhere between 1 and 2% of the 10 m wind speed. This should be compared to the popular rule-of-thumb which gives the total surface drift as 3% of the wind speed.

Stokes’ drift is the net drift produced by the fact that the orbital motions caused by deep-water wind waves are not exactly closed, a result of the decrease of the orbital velocities with depth. This process is rarely modeled explicitly in hydrodynamic models, although it may be implicitly included in the bulk formulation of the surface boundary conditions. Furthermore, in large-scale hydrodynamic models it is not possible to resolve all the details near the surface, and thus some kind of parameterization of the near-surface velocity profile is necessary. At a solid boundary, the classic logarithmic law of the wall may be employed, but at the free surface this is unlikely to be appropriate [4]. Hence, in drift modeling it is common to simply use the rule-of-thumb mentioned above as an estimate of the surface drift velocities, sometimes with a modification of the direction to account for the difference between the wind direction and the surface current direction.

In PADM, however, the actual Stokes’ drift is calculated as a function of depth from a two-dimensional wave energy spectrum, which is then added to the mean velocity in the surface layer predicted by the hydrodynamic model [1]. Ideally, the wave energy spectrum should be forecasted using a spectral wave model, but currently a parameterized Donelan–Banner spectrum is used [5]. This spectrum is calculated from the wind and the fetch in each point in the model. Finally the Stokes’ drift is modified by the presence of ice, such that the drift velocity is set to zero for ice concentrations exceeding 70% and reduced linearly for lower ice concentrations.

4.1.3 Horizontal Gravity-Induced Surface Spreading

Horizontal gravity-induced surface spreading is only relevant for oil. It is basically the radial spreading that you would see if you poured oil on a table top. The oil will spread radially, fast at first and then slower and slower, until it reaches a terminal thickness determined by the oil’s viscosity. Fay’s classic formula [6] describes how the area of an oil slick increases with time. This formula has been rewritten in terms of the change in oil thickness [1]. To apply it to a cloud of particles, each particle is modeled as a disc whose thickness varies according to the rewritten version of Fay’s formula. Factoring in changes in volume due to weathering or discharge of oil, the resulting radii of all discs can be calculated. The gravity-induced spreading of the particle cloud is then determined by ensuring that the discs do not overlap as they grow in size and become thinner.

4.1.4 Vertical Dispersion by Breaking Waves

Vertical dispersion is defined as the process of spreading substances from the surface into the water column. For dissolved or suspended substances, this is simply modeled using the small-scale turbulent mixing (see Sect. 4.1.1). Cohesive buoyant oil slicks, however, are unlikely to disperse this way. An oil slick needs to be broken up into droplets which then, if they are buoyant, must be forced down into the water column.

A suggested mechanism for this is the breaking of waves over an oil slick. Hence, in PADM we use the empirical expressions derived by Delvigne and Sweeney [7], which yield the mass of oil to be dispersed for a given time interval as a function of the droplet diameter, the breaking wave energy, the fraction of the surface covered by breaking waves, the oil slick area, and the time interval. The breaking wave energy and the fraction of the surface covered by breaking waves are estimated from the significant wave height and the wind speed, respectively. The significant wave height is in turn calculated from the wave spectrum. Once the total mass to disperse has been determined, particles are selected randomly and dispersed until the total mass to disperse has been reached. Each particle is assigned a droplet diameter from a predetermined range of size classes and then injected to a depth selected randomly from the range zero to the intrusion depth, a parameter proportional to the breaking wave height.

The mass to disperse is modified by a factor related to the ice concentration, such that for high ice concentrations no dispersion occurs since it is assumed that high ice concentrations will strongly dampen the wave field. Once an oil particle has dispersed, it is assumed to represent a cloud of droplets of equal diameter, which then will be advected, mixed, and rise (or sink).

4.1.5 Buoyancy-Induced Vertical Velocities

Particles that are lighter or heavier than the surrounding sea water may rise or sink. This buoyancy-induced vertical velocity can be calculated from the classical Stokes’ formula, but this is only valid for small spherical droplets. For larger diameters, this formula will severely overestimate the vertical velocity. Hence, a two-regime formula developed primarily for oil is used [8] but with the coefficients modified according to a more complex three-regime model [1,9].

4.1.6 Ice Drift, Wind Drag, and Boundary Interaction

In the case of high ice concentrations (>0.7), the hydrodynamic flow field is modified in such a way that the surface current is replaced by the ice drift velocity. Thus, it is assumed that the oil will move with the ice.

For objects that extend above the surface, i.e., are not fully submerged, it is possible to add a drag due to the wind, by specifying a percentage of the wind velocity that will be added componentwise to the horizontal flow velocities.

Boundaries in PADM can be of different categories: the sea surface, the coastline, the bottom, and so-called open boundaries through which exchange with water bodies exterior to the model may occur. For each boundary category, a boundary action can be set. This determines what action should be taken when a particle’s trajectory intersects a boundary. Three types of boundary actions are currently available in PADM: slip, halt, and deactivation.

  • Slip means that a particle cannot pass through a boundary but may move tangentially along it.

  • Halt means that the particle is held at the location where it hit the boundary and its position is no longer updated, unless it is released again. However, other processes such as weathering may continue to act on the particle.

  • Deactivate means that the particle is deleted and no longer takes part in the calculations.

In the current implementation of HELCOM Seatrack Web, different boundary actions have been set depending on the type of substance represented by the particles. For oils, the slip action is used for the sea surface but for all other boundaries deactivation occurs. This means that oil that intercepts the coastline or the bottom is assumed to stick in place and not undergo any more weathering. For other substances, e.g., floating objects, algae, etc., the slip action is used for all boundaries except open boundaries, where instead deactivation occurs.

4.2 Weathering Model

When Seatrack Web is used to forecast oil drift each particle represents a quantity of oil with a common set of properties:

  • Mass of oil

  • Mass of water-in-oil

  • Total mass and volume of oil and water-in-oil

  • Density of oil

  • Bulk viscosity

These properties are variables that change due to two different processes: evaporation and emulsification. Evaporation is only calculated for oil on the surface. If the total mass of a particle reaches zero, all the oil is assumed to have evaporated and the particle is deactivated. It is further assumed that dispersed oil droplets do not form a water-in-oil emulsion. Thus, if oil that has formed an emulsion is dispersed, all the water is immediately removed.

Different types of petroleum products can be simulated. These are broadly categorized as: oil classes, specific oils, and oil lumps.

Oil classes are used when the specific type of oil is not known. The three classes available today are light oils (viscosities less than 100 cSt), medium oils (viscosities in the range 100–1,000 cSt) and heavy oils (viscosity greater than 1,000 cSt). This is useful when simulating a spill where the exact oil product is unknown, but where observations may give some indication of the oil’s viscosity. In practice, the class light oils is represented by light diesel fuel, the class medium oils by what is termed intermediate oil and the class heavy oils by Bunker C.

The specific oils available in HELCOM Seatrack Web are listed in Table 1. Depending on the oil being simulated different weathering models are used, as different empirical constants have been determined for different sets of petroleum products. In HELCOM Seatrack Web, there are two alternative weathering models: one based on a proprietary code supplied by SINTEF [10] and the original Seatrack Web model based on simple empirical formulae [11,12]. Which model is used for which oil is also shown in Table 1.

Table 1 Oils available in HELCOM Seatrack Web, including the weathering model used
Full size table

4.2.1 SINTEF Model

This model is based on tables of empirical data for relevant oil properties which show how these properties change in time. Interpolation into these tables gives the values at a given point in time, and these values are then used to determine the evaporation, emulsification, density, and viscosity.

To calculate the evaporation, empirical data on the evaporated fraction in percent f e at different evaporation exposure times is used. The mass of oil M after a given time of exposure is then simply given by

$$ M = \left( {1 - \frac{{{f_{\rm{e}}}}}{{100}} } \right){M_0}. $$
(1)

Here, M 0 is the initial mass of fresh oil. The evaporation exposure time t evap is, however, not only a function of time, but of several other factors as well. The increase in the exposure time is given by

$$ \Delta {t_{\rm{evap}}} = \Delta t\left( {1 - {C_{\rm{ice}}}} \right)\frac{W}{{{W_{\rm{ref}}}}}\frac{{{h_{\rm{ref}}}}}{h}{T_{\rm{corr}}}. $$
(2)

W ref and h ref are reference values for the wind speed and oil thickness, respectively, for which the property tables have been generated. The temperature-dependent correction factor is given by

$$ {T_{\text{corr}}} = {2^{{{\frac{{T - {T_{\text{ref}}}}}{8}}}}.} $$
(3)

Here, T is the sea surface temperature and T ref a reference temperature for which the property tables have been generated. The maximum exposure time, i.e., when no more evaporation occurs and the oil is completely weathered, is set to 140 days.

To calculate the emulsification empirical data on the mass fraction in percent of water in a water-in-oil emulsion m w for different emulsification exposure times is used. The mass of water in a water-in-oil emulsion M w is then calculated as

$$ {M_{\rm{w}}} = M\frac{{{m_{\rm{w}}}}}{{100 - {m_{\rm{w}}}}}. $$
(4)

The emulsification exposure time t emul is again not only a function of the time but is also calculated using the following expression for the increase in exposure time:

$$ \Delta {t_{\rm{emul}}} = \Delta t\frac{{{{\left( {\left( {1 - C{}_{\rm{ice}}} \right)W + 1} \right)}^2}}}{{{{\left( {{W_{\rm{ref}}} + 1} \right)}^2}}}. $$
(5)

The density of the oil ρ oil is determined from tabulated empirical data for different evaporation exposure times (2). The particle density including water-in-oil emulsion ρ P is calculated according to

$$ {\rho_{\rm{P}}} = \frac{1}{\displaystyle{\frac{{\left( {1 - {m_{\rm{w}}}} \right)}}{{{\rho_{\rm{oil}}}}} + \frac{{{m_{\rm{w}}}}}{\rho }}}, $$
(6)

where ρ is the density of seawater.

The oil viscosity in cP μ oil is also determined from tabulated empirical data for different evaporation exposure times (2). The empirical data also contain values for the viscosity of a stable water-in-oil emulsion μ emul in cP at different emulsification exposure times (5). First, the viscosity considering only evaporation is determined by interpolation into the table of empirical data on μ oil and then adjusted for the actual sea water temperature T according to

$$ {\mu_{\rm{oil}}}(T) = {10^{{{{10}^{\lambda }}}}} $$
(7)

where

$$ \lambda = - 0.0045\left( {T - {T_{\rm{ref}}}} \right) + \log \left( {\log \left( {{\mu_{\rm{oil}}}} \right)} \right). $$
(8)

Here, T ref is the reference temperature for which the property tables have been generated. The particle viscosity including the effect of emulsification μ P is then determined according to

$$ {\mu_{\rm{P}}} = {F_{\rm{emul}}}{\mu_{\rm{oil}}}(T). $$
(9)

The ratio between the viscosities of water-in-oil emulsion and oil F emul is determined by interpolating at the current emulsification exposure time into a new table, generated by dividing the empirical data on μ emul by the data on μ oil. The particle viscosity can be converted to kinematic particle viscosity (unit cSt) using

$$ {\nu_{\rm{P}}} = \frac{{{\mu_{\rm{P}}}}}{{0.001{\rho_{\rm{P}}}}}. $$
(10)

4.2.2 Original Seatrack Web Model

All oils are represented using a two-component model, i.e., they consist of a volatile and a nonvolatile component. The oil properties are defined in a database file and comprise the following set of parameters:

  • Densities of the volatile and nonvolatile components

  • Viscosity

  • The maximum water fraction of emulsified oil

  • The level of evaporation required for emulsification to begin

  • An emulsification rate coefficient

  • The fraction of the oil which is nonvolatile

  • Two rate coefficients for evaporation

  • Three coefficients for calculating the viscosity

The densities and the viscosity are approximate standard values for fresh oils at typical sea water temperatures. At the beginning of a simulation, oils are considered either fresh or completely weathered. In Table 2, the properties of the oils for which the original Seatrack Web weathering model is used are presented.

Table 2 Table showing the properties of the oils employing the original Seatrack Web weathering model. For definition of the symbols see the text. A hyphen (-) implies that the parameter is dimensionless
Full size table

Evaporation is calculated based on simple expressions for the evaporation of the form [11,12]

$$ {f_{\rm{e}}} = \left( {{C_1} + {C_2}T} \right)\ln \left( {\frac{t}{{60}} + 1} \right). $$
(11)

Here, f e is the percentage fraction of the particle mass that has evaporated and C 1 and C 2 are coefficients. Values for the coefficient for different oils are presented in [12]. The last term is simply to avoid a singularity at t = 0.

Equation (11) is a solution to the ordinary differential equation

$$ \frac{{{\rm{d}}{f_{\rm{e}}}}}{{{\rm{d}}t}} = \frac{{\left( {{C_1} + {C_2}T} \right)}}{{60}}{e^{{ \frac{{ - {f_{\rm{e}}}}}{{\left( {{C_1} + {C_2}T} \right)}} }}}. $$
(12)

Here, we will model the fractional evaporation rate \( E = {{{{f_{\rm{e}}}}} \left/ {{100}} \right.} \) as

$$ \frac{{{\rm{d}}E}}{{{\rm{d}}t}} = {C_{\rm{e}}}{e^{{ -\frac{K}{{{C_{\rm{e}}}}}E}}}. $$
(13)

The coefficients C e and K have dimension 1/t. The solution to (13) is

$$ E = \frac{{{C_{\rm{e}}}}}{K}\ln \left( {{e^{{\frac{K}{{{C_{\rm{e}}}}}{E_0}}}} + K\left( {t - {t_0}} \right)} \right). $$
(14)

Here, \( {E_0} = E\left( {{t_0}} \right) \). We can identify the coefficients by equating (12) and (13), yielding

$$ K = \frac{1}{{60}} $$

and

$$ {C_{\rm{e}}} = \frac{K}{{100}}\left( {{C_1} + {C_2}T} \right). $$

We can thus calculate E at time t as a function of temperature and the value at time t 0. To account for the presence of ice, the right hand side of (13) is multiplied by a correction factor, r ice. This yields

$$ E = \frac{{{C_{\rm{e}}}}}{K}\ln \left( {{e^{{\frac{K}{{{C_{\rm{e}}}}}{E_0}}}} + K{r_{\rm{ice}}}\left( {t - {t_0}} \right)} \right). $$
(15)

The remaining oil mass is then calculated according to

$$ M = {M_0}\left( {{E_{{\max }}} - E} \right) + {M_{\rm{n}}}. $$
(16)

Here, E max is the maximum fraction that can evaporate, i.e., the volatile fraction in fresh oil, and M n is the constant mass of the nonvolatile component. The ice correction factor is given by

$$ {r_{\text{ice}}} = \{ \begin{array}{*{20}{c}} 0 \hfill & {{C_{\text{ice}}} \ge 0.8} \hfill \\\displaystyle{\frac{{0.8 - {C_{\text{ice}}}}}{{0.5}}} \hfill & {0.3 \le {C_{\text{ice}}} 0.8} \hfill \\1 \hfill & {{C_{\text{ice}}} 0.3} \end{array} \}. $$
(17)

The mass fraction of water in a water-in-oil emulsion, m w, is defined by

$$ {m_{\rm{w}}} = \min \left( {{m_{{{\rm{w}},\max }}},\frac{{{M_{\rm{w}}}}}{{{M_{\rm{w}}} + M}}} \right). $$
(18)

Here, m w,max is an oil-specific maximum water fraction for water-in-oil emulsion and M w is the mass of water in the water-in-oil emulsion. The rate of change of m w is [6]

$$ \frac{{{\rm{d}}{m_{\rm{w}}}}}{{{\rm{d}}t}} = R\left( {{m_{{{\rm{w}},\max }}} - {m_{\rm{w}}}} \right). $$
(19)

The rate at which the oil forms an emulsion, R, is related to the wind speed W, as the process requires agitation of oil and water. However, emulsion only takes place if a sufficient fraction, E emul, of the volatile components has evaporated. Thus, the emulsion rate R is modeled by

$$ R = \left\{ {\begin{array}{ll} 0 \hfill& {E {E_{\text{emul}}}} \\{{r_{\text{ice}}}{C_R}{W^2}} & {E \ge {E_{\text{emul}}}}\end{array} } \right.. $$
(20)

Here, r ice is the same reduction factor as for evaporation (17) to account for the presence of ice and C R is an oil-specific constant coefficient. Integrating (19) yields the following expression for the water fraction:

$$ {m_{\text{w}}}\left( {t + \Delta t} \right) = {m_{{{\text{w}},\max }}} - {e^{{ - R\Delta t}}}\left( {{m_{{{\text{w}},\max }}} - {m_{\text{w}}}(t)} \right). $$
(21)

For a two-component model, the oil density is simply

$$ {\rho_{\text{oil}}} = \frac{1}{{{m_{\text{n}}}/{\rho_{\text{n}}} + {m_{\text{v}}}/{\rho_{\text{v}}}}}. $$
(22)

Here, m n and m v are the mass fractions and ρ n and ρ v are the densities of the nonvolatile and volatile oil components. The particle density ρ P including the effect of emulsification is calculated using (6).

The particle viscosity in cSt, including the effect of emulsification, is determined from the amount of the volatile fraction that has evaporated (E) and the degree of emulsification (fraction of water-in-oil) according to

$$ {\nu_{\text{P}}} = {\nu_{\text{ref}}}{e^{{aE}}}{e^{{\textstyle \frac{{b{m_{\text{w}}}}}{{1 - c{m_{\text{w}}}}}}}}. $$
(23)

Here, ν ref is the reference oil viscosity (cSt) given in the oil properties database file whereas a, b and c are constant coefficients specific for each oil [6].

5 Client/Server Java Application

Seatrack Web is an online system accessible via the Internet using a client/server Java application. This allows users, after logging in to the system via a web page, to start an oil drift simulation and present the results on their own computers, even though the actual simulations are executed on a remote server. The technical architecture consists of three main components: (1) a database containing users, usage statistics, configuration data, and news for the Seatrack Web start page, (2) a server consisting of a Java Servlet that communicates with the drift model and the database, and (3) the Java client that runs the graphical user interface (GUI) on the user’s computer.

When the user starts a calculation in Seatrack Web, the client establishes a connection to the server, which starts the drift model. After a drift calculation is completed, the model results are transferred back to the client where they can be visualized in various ways in the GUI.

5.1 Java Client and Graphical User Interface

The GUI employs the Java Web Start technology. The Java client application including background maps is automatically downloaded and installed on the user’s hard drive. When the user clicks the start link on the Seatrack Web home page, the version of the Java client is checked against the server version and an update is downloaded if necessary. Otherwise, the cached version of the client is executed. This design saves download time and administration of the client installations.

To create the GIS-based GUI which can display maps and other relevant geographic information, an open source Java library for map applications called OpenMap™ (BBN Technologies) is used. The resulting GUI is thus very similar to other GIS software, with a map and standard functionalities such as zooming, panning, etc. In addition, specific tools and menus have been implemented to allow the user to define an oil spill, set up the simulation parameters, and display the results in various ways (see Sect. 6.2).

An important advantage of this so-called rich client technology is that demanding visualizations such as animations are performed noticeably faster than if the GUI would have been implemented in a web browser. Since the GUI runs on the user’s computer, every input and every change in the visualizations need not be transmitted over the internet but are handled locally once the result files have been downloaded. Of course, the transfer time is dependent on the speed of the user’s internet connection.

5.2 Java Servlet

At the server end, Java Servlet technology is used. The server handles authentication of the user, communication with the client application, and execution of the drift model. When the user requests a simulation, the server performs a number of tasks:

  • Receiving the simulation settings from the client application

  • Preparing the settings file for the drift model

  • Setting up the execution environment for the drift model

  • Executing the drift model

  • Responding to client requests for status updates

  • Transmitting the results to the client

  • Handling and transmitting additional data upon request. This includes AIS and possibly satellite data as well as forcing data (surface currents, wind, and ice concentration) and the availability of forcing data which have been prepared during the preprocessing stage (see Sect. 3.1).

The system naturally must be available 24/7 with minimal downtime. Hence, the Java Servlet as well as the preprocessing scripts and software have been operationalized as a critical system, including automatic monitoring, rapid intervention upon a critical failure, and a separate backup system.

The present technology is not very scalable, running on a single multicore machine. However, some tens of concurrent users can still be accommodated. The primary issue is the number of simulations being executed simultaneously. However, the scalability will be improved in the near future (see Sect. 7).

6 Usage and Results

In this section, we will briefly present the output of the drift modeling as well as a user’s view of the system, including important features and functionalities in the GUI. We will also present a validation using data in the form of airborne SLAR images of an actual oil spill.

6.1 Model Output

The fundamental outputs of a simulation are the tracks of the individual particles. The three-dimensional position of each particle is currently stored every 15 min. In addition, the trajectory of the center position of the particle cloud, calculated as the mean position of all the active particles, is output as well. These results are visualized in the map.

In the result table (see Fig. 6), some overall properties of the particle cloud are presented as functions of time. These include the center position, the speed (in knots) and direction of the current at the center position, and the speed (in m/s) and direction of the wind at the center position.

In the case of oil, a number of overall properties for the entire spill are also calculated and presented in the result table. These are based on the physical and chemical properties of each particle computed by drift model and include the following parameters:

  • The total volume of the spill (m3)

  • The mean viscosity (cSt; weighted by mass)

  • The mean density (kg/m3; weighted by mass)

  • The fraction of the initial mass of oil (%) that has evaporated

  • The fraction of the initial mass of oil (%) and the volume (m3) that is on the surface

  • The fraction of the initial mass of oil (%) and the volume (m3) that is dispersed into the water column

  • The fraction of the initial mass of oil (%) and the volume (m3) that is on the sea bed

  • The fraction of the initial mass of oil (%) and the volume (m3) that is on the shoreline

  • The overall relative water content (%)

The sum of the different fractions of the initial mass of oil is 100%.

6.2 Using Seatrack Web

When a user logs into Seatrack Web, the first thing that happens is that the GUI is adapted for that particular user, meaning that certain features may be activated or deactivated depending on the needs and rights of the privileges of the user in question. An example of this is the AIS ship tracks feature described earlier (see Sect. 3.2).

The first thing to be done is to define the location of the oil spill, either as a point, a line, or a three- or four-sided polygon. This is done by simply clicking in the map, but can also be input manually as positions. Next, a choice can be made between which hydrodynamic forecast should be used: a two-day forecast based on meteorological forcing from the HIRLAM model using either the one or three nautical miles HIROMB models, or a five-day forecast based on meteorological forcing from ECMWF and the three nautical miles HIROMB model. This is shown in Fig. 3. After completing this stage, a new case is defined by reviewing the default simulation input parameters and modifying them where appropriate. The user can select a forward or a backward calculation, modify the time period and the discharge position, select a substance, and define the type of discharge and the amount discharged. In addition, it is possible to create simple user-defined scenarios where the wind and current are not read from the meteorological and hydrodynamic forecasts but prescribed in advance. Finally, the user may choose the Calculation Mode (in effect, the number of particles to be used), add an uncertainty factor, and enter information describing the simulation (see Fig. 4).

Fig. 3
figure 3

The GUI of HELCOM Seatrack Web showing the Model Choice dialogue box and the full extent of the model area

Full size image
Fig. 4
figure 4

The GUI of HELCOM Seatrack Web showing the Calculation Parameters input window. In the map layers showing the bathymetry of the 3 nautical miles model, Baltic Sea protected areas and important bird areas have been activated

Full size image

Once the user has pressed Compute, the model simulation starts on a remote server and the progress is shown on the screen. As soon as the simulation has been completed, the results are downloaded and the final particle positions displayed in the map. It is now possible to visualize the model results in several different ways. These include animating the particle drift, plotting the trajectory of the center of the particle cloud, as well as plotting the trace of the particles, i.e., the particle positions at all times (equivalent to the impacted area). The color of each particle indicates its depth (see Fig. 5).

Fig. 5
figure 5

The GUI of HELCOM Seatrack Web showing the results of a simulation. The colored points represent the oil (color indicating depth of each particle) and the red quadrilateral is the initial spill area. In the map are also shown surface current vectors and a few ship tracks from AIS data. The drop-down menu at the top can be used to select the time step to view

Full size image

Many other types of information can be added to the map. Firstly, the wind and surface current vectors can be plotted and animated, as well as the ice cover. A number of predefined GIS layers can be turned on and off, including the depths in the hydrodynamic model, the coastline of the hydrodynamic model, protected areas, important bird areas, major ship routes, borders, etc. Of course AIS data may be shown and text can be added to the map and edited.

The results of the drift simulation can also be viewed as a table, by selecting Results table under the View menu. This shows the properties of the particle cloud as a function of time and also allows for the generation of xy-plots of the different properties (see Fig. 6).

Fig. 6
figure 6

The GUI of HELCOM Seatrack Web showing the Results Table window and a plot chart

Full size image

Finally, the user may save the simulated case locally to be opened again later and may choose to perform a so-called Continued Case, meaning that the current results are used as the start conditions for a new simulation. In this case, it is also possible to shift the entire particle cloud by modifying the center position of the cloud.

6.3 Examples of Results

Around noon on Saturday 31 May 2003, MV Fu Shan Hai, fully laden with a cargo of 66,000 metric ton of fertilizer loaded in Ventspils, Latvia was struck by a container ship, MV Gdynia, at a location north of Bornholm. Fu Shan Hai soon began to sink and the crew was evacuated within two hours. At 18:50 UTC, Fu Shan Hai sank at 55° 20.7′ N, 14° 45.7′ E to a depth of approximately 65 meters. Fu Shan Hai carried approximately 1,700 m3 of fuel oil, which began to leak into the sea during the night and continued to leak for several days afterward.

The HELCOM Seatrack Web system was used to forecast the drift of the oil spill from Fu Shan Hai. After the spill event, a few SLAR images taken from aircraft during the spill were made available. These have been interpreted, plotted in a map, and compared to the results of the Seatrack Web forecasts. Please note that this was an earlier version of HELCOM Seatrack Web with only a three nautical mile resolution.

In Fig. 7, the observed oil spill and the forecast are shown at around 04:00 UTC on 3 June. The agreement is clearly quite satisfactory. The modeled spill has reached approximately the same distance from the shoreline as the observed spill, the overall trajectories agree and the model also replicates the wider east front of the observed spill.

Fig. 7
figure 7

The Fu Shan Hai spill as forecasted by Seatrack Web (upper panel) and observed via airborne SLAR images (lower panel) on 3 June 2003 at around 04:00 UTC

Full size image

A comparison between the observed and forecasted oil spill on 3 June at around 12:00 UTC are presented in Fig. 8. Again, the agreement is quite satisfactory, both regarding timing and the shape of the spill. Hence, Seatrack Web managed to forecast fairly accurately when and where the oil spill reached the shoreline.

Fig. 8
figure 8

The Fu Shan Hai spill as forecasted by Seatrack Web (upper panel) and observed via airborne SLAR images (lower panel) on 3 June 2003 at around 12:00 UTC

Full size image

7 Future Developments

Perhaps the most important factor determining the quality of the forecasts produced by the Seatrack Web system is the quality of the hydrodynamic and meteorological forecasts that force the drift model. This can be improved in various ways, such as increased resolution, data assimilation, improved understanding of key processes, calibration against observations, etc. However, this is a separate issue that will not be dealt with further here.

Focusing on Seatrack Web, there are basically three areas where developments are in progress:

  1. 1.

    Improving the drift model by improving the description of the processes important to oil spill modeling (e.g., the near-surface advection or the weathering algorithms). This also includes enhancing code clarity to reduce the risk of coding errors and to facilitate cooperative development efforts. Improving the performance can also benefit the quality by permitting more computationally costly algorithms and solutions.

  2. 2.

    Increasing the usability by adding features that are important to responsible authorities, incident commanders, and other decision makers, such as incorporating additional local information, simplifying the simulation set up process, integrating other decision support systems into Seatrack Web or vice versa, tailor the output presentation, etc.

  3. 3.

    Improving the security, reliability, and scalability of the client/server application.

Currently three development efforts are under way or being considered within the first category: (1) to use wave spectra from wave forecast models rather than a parameterized spectrum, (2) to improve the implementation of the turbulent mixing, and (3) determining a realistic model of the horizontal dispersion of surface oil. A future wish list would include more sophisticated weathering algorithms and a mechanistic description of the vertical dispersion of surface oil.

Development in progress within the second category includes: (1) refining the current method for defining the initial spill by allowing the user to define a polygon of arbitrary shape and any number of vertices, (2) implementing a system for importing preprocessed satellite images of spills, and (3) improving the access to AIS data throughout Europe. The European Maritime Safety Agency (EMSA) is currently testing functionality for exporting satellite-based Synthetic Aperture Radar (SAR) images of detected spills at sea. Once this has been completed, these geo-referenced images can be imported into Seatrack Web and displayed in the user interface. The user can then define an initial spill area for a forecast based on an image of the actual spill. More sophisticated methods for defining the spill area – e.g., by automatically creating a polygon in Seatrack Web based on a satellite detection of an oil spill – will also be developed and Seatrack Web will then provide automated forecasts and backtracking simulations to EMSA whenever an oil spill is detected.

The third category entails implementing the server software in a load balancing environment, using Web Service technology for handling the execution of the drift model. This will allow for multiple redundant hardware and Seatrack Web Java Servlets. It would also make it easier for other applications to interact with Seatrack Web. A REST-based Web Service software has been developed and is currently being tested. In the future, the Seatrack Web GUI may be moved into the web browser and the whole system may also be used together with Web Map Services (WMS) for displaying Seatrack Web simulation results in a standardized way for possible third parties.

8 Conclusions

Just as weather forecasting has become an integral and vital part of human society, the ability to forecast accidental oil spills is becoming standard practice all over the world. The need for such tools has been clearly highlighted by the recent Montara and Deep Horizon oil spills. With the projected increase in shipping within the Baltic Sea, it is safe to say that the development and operation of HELCOM Seatrack Web should be considered a necessity and the forethought of HELCOM is to be commended.

It is in the nature of forecasting that it will never be an exact science. However, the model description and results presented here hopefully will bolster confidence in the usefulness of HELCOM Seatrack Web. It is therefore important to spread knowledge about this tool and to continuously integrate its use in oil spill response through training, multilateral agreements, and exercises.

Of course, much can still be done to improve systems such as Seatrack Web. Considering the stakes and potential extra costs of an oil spill response based on inferior data, further development should not be neglected. It is also important to realize that oil spill modeling covers many disciplines, from physical oceanography and hydrodynamics to oil chemistry and ecotoxicological issues, requiring the input from many avenues and research and fruitful collaboration.