Ocean Engineering 144 (2017) 340350

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Oil drift modeling in pack ice Sensitivity to oil-in-ice parameters

glund a, Lars Axell a, Mikko Lensu b, Olof Liungman c,
Lars Arneborg a, *, Anders Ho
d
Johan Mattsson
a
Swedish Meteorological and Hydrological Institute, Research Department, 42671 V
astra Frolunda, Sweden
b
Finnish Meteorological Institute, 00560 Helsinki, Finland
c
Tyrens AB, 20519 Malm o, Sweden
d
Defence Centre for Operational Oceanography, 2750 Ballerup, Denmark

A R T I C L E I N F O A B S T R A C T

Keywords: An improved oil-in-ice parameterization has been implemented in the Seatrack Web oil spill model, tested using
Ice covered waters data from the Runner 4 oil spill accident, which occurred in high ice concentrations in the Gulf of Finland on 5
Oil spill March 2006. The model is able to describe the observed transport and spreading of oil reasonably well. The
Oil spill modeling sensitivity of the results towards oil-in-ice parameters and hydrodynamic forcing models has been investigated.
Ice modeling Both the mean oil trajectory and the oil spreading are sensitive to a threshold velocity for the withdrawal of oil
from below ice oes, and the oil spreading is highly sensitive to the size of the oes. The trajectories for the ice
drift and for the water current drift set the limits of the oil drift, and these are in turn highly dependent on the
hydrodynamic forcing and the assimilation of ice conditions. In future development of oil spill modeling in ice it is
therefore important to focus not only on the ice parameterization but also on the ability to model ice drift, ice oe
sizes, and the currents below the ice.

1. Introduction
Oil spills in ice covered regions provide an increasing threat to Arctic
and sub-Arctic ecosystems, as vessel trafc in ice covered regions in-
creases and exploration of oil and gas reserves in the Arctic becomes
feasible (Drozdowski et al., 2011; Lee et al., 2015; Afenyo et al., 2016b).
Oil spill trajectory modeling is an important tool for risk management of
oil-related activities, i.e., to estimate environmental risks, plan oil spill
response measures, and to provide operational information about the
location of actual spills (Afenyo et al., 2016b). However, there are large
gaps in our knowledge about the behavior of oil when it enters cold,
ice-covered waters, and thus research on crude oil spills in Arctic waters,
improved risk assessments, improved models of oil-in-ice effects, and
improved quantication of model uncertainties have been identied as
high-priority research areas by a Royal Society of Canada expert panel
(Lee et al., 2015).
Here we will focus on oil spills in the Baltic Sea, which is a heavily
trafcked semi-enclosed sea in northern Europe. The Baltic Sea is
partially ice covered during the winter season, and will continue to be so
in a future climate, even though the ice will become thinner and more
mobile, the ice season will become shorter, and inter-annual variations
will be large (H oglund et al., 2017). The risk of vessel accidents in the
northern Baltic has been studied in Valdez Banda et al. (2015, 2016) and
Goerlandt et al. (2017), and the outow of oil from damaged vessels has
been studied in Kollo et al. (2017) and Sergejeva et al. (2017). As a next
step in a full risk analysis it is important to determine the fate of the oil
spill, and the present work on oil spill modeling in ice-covered water aims
to support this endeavor.
Oil spill trajectory modeling calculates how an oil spill is transported,
spread out, and weathered under the inuence of winds, waves, currents,
turbulence, gravity, surface tension, viscosity, and, if present, ice
(Drozdowski et al., 2011; Afenyo et al., 2016a,b). Early trajectory models
were based mainly on wind data from atmospheric models (ASCE, 1996),
but present model systems also include wave conditions from wave
models, and current, turbulence, and ice conditions from oceanographic
models to predict the trajectories. One such system is Seatrack Web
(Liungman and Mattsson, 2011; Ambj orn et al., 2014; Ambj orn and
Mattsson, 2006), which is a web-based oil spill trajectory model used by
about 80 stakeholders and agencies around the Baltic Sea. It is
co-developed by SMHI in Sweden, the Danish Maritime Safety Admin-
istration (presently the Defence Centre for Operational Oceanography,
FCOO), Bundesamt fur Seeshiffart und Hafen (BSH), and the Finnish

* Corresponding author. Swedish Meteorological and Hydrological Institute, Research Department, Sven Kallfelts gata 15, 42671 Vastra Fr
olunda, Sweden.
E-mail address: lars.arneborg@smhi.se (L. Arneborg).

https://doi.org/10.1016/j.oceaneng.2017.09.041
Received 28 June 2017; Received in revised form 25 August 2017; Accepted 24 September 2017
Available online 4 October 2017
0029-8018/ 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
L. Arneborg et al.
Environment Institute (FMI). In the present work we study how an
improved parameterization of the interaction between oil and ice in-
uences the Seatrack Web model results.
Present knowledge about how the interaction with ice modies the
transport and weathering of an oil spill is based on a limited number of
laboratory experiments and experimental releases in the 70s - 90s (see
Fingas and Hollebone, 2003 for a review), but more recent eld experi-
ments have also been performed (Brandvik and Faksness, 2009; Faksness
et al., 2016). A recent review of oil spill modeling in ice covered waters is
given in Afenyo et al. (2016b). The presence of cold conditions and ice
signicantly complicates an already complicated problem by modifying
the properties of the oil, as well as providing a number of additional
states in which the oil can be transported (see Fig. 5 in Afenyo et al.,
2016b), e.g., under the ice, on top of ice mixed with snow or in meltwater
pools, encapsulated in the ice, propagating up brine channels, etc.
The ice in the Baltic Sea consists only of rst year ice, and the heavy
vessel trafc and ice breaker activity in the region modies the ice
conditions when compared to, e.g., remote Arctic regions. Hence, some
Arctic results may not be directly applicable in the Baltic. In addition,
there are few datasets available for model validation in the Baltic Sea. Of
the few reported accidental spills in ice, we have only found one case
where a large fraction of spilled oil was tracked for a longer period. On 5
March 2006 the Dominican-registered cargo ship Runner 4 collided with
the Malta-registered cargo ship Svjatoi Apostol Andrey in heavy ice
conditions in the Gulf of Finland, and sank (Wang et al., 2008). The ship
carried 102 tonnes of heavy fuel oil, 35 tonnes of light fuel oil, and 600 L
of lubricant. Between 5th and 15th March there are few observations of
the oil patch, but between the 15th and 19th March large oil patches
were observed and combated north-east to north-west of Tallinn, before
the oil-ice mixture moved into the shallow waters around Tallinn (Fig. 1).
Ice drift modeling presented in Wang et al. (2008) was unable to explain
the drift from the collision site to Tallinn. The same problem was expe-
rienced with Seatrack Web simulations of the oil spill, but improved
descriptions of air-ice and ice-water drag, as well as assimilation of sat-
ellite derived ice products, in an updated version of Seatrack Web yielded
a stronger ice drift that enabled the simulated ice and oil trajectories to
reach Tallinn (Raudsepp et al., 2010). The spreading of the oil patch in
the model was much smaller than what had been observed, but by
assuming a continuous oil release between the 5th and 10th March a
reasonable spreading of the oil patch was obtained. Stanovoy et al.
(2012) proposed a more advanced oil spreading model for use in the Gulf
of Finland, where oil is allowed to spread under ice, and where the
advection, diffusion and spreading of ice on the surface is dependent of
ice concentration, c. They did, however, not present any validation of the
model against real spills, e.g., the Runner 4 case.
Here we will present the results of implementing a more advanced oil-
in-ice parameterization in the Seatrack Web code and how this inuences
Fig. 1. Sketch of oil spreading between 15 and 18 March as observed from oil combattting vessel Hylje (reproduced from Wang et al., 2008 with permission from the author).
341
Ocean Engineering 144 (2017) 340350
the advection and spreading in the Runner 4 case. The main aims are to
identify some important parameters in the oil-in-ice parameterization,
and to study the sensitivity of the results to these parameters as well as to
the forcing elds. The implementation largely follows the suggestions of
Venkatesh et al. (1990) where the oil behavior changes between three ice
concentration regimes;

c < 30% where oil moves as if there is no ice,

30% < c < 80% where oil spreading is inuenced by ice oes, some of
the oil may be situated below ice oes, and oil at the surface may
either be moving without being inuenced by the ice or with the ice
drift, and

c > 80% where a large part of the oil is under the ice and the oil in
leads is prevented from spreading further.
The remaining parts of this paper are structured as follows. First we
will present the details of the oil trajectory model, including the old and
the new description of oil in ice. In section 3 we present the runs per-
formed with the model, and the atmospheric and ocean forcing elds
used to produce the results. The results are presented in Section 4 and
discussed in Section 5. Finally, Section 6 summarizes the results and
states the conclusions.
2. Description of Seatrack Web
Seatrack Web (Liungman and Mattsson, 2011) is an oil spill trajectory
system that can be run with a user-friendly web-based interface. Seatrack
Web consists of a forcing interface part, a user interface part, and a tra-
jectory calculation part. The trajectory calculation part of Seatrack Web is
called PADM (PArticle Dispersion Model), and will be described in detail
below. The forcing interface part provides input of currents, winds,
temperatures, ice conditions, etc., from atmospheric and oceanographic
models. Interfaces are available for several models. In the Baltic, the
system is run in operational mode with input from regional operational
models. Until January 2017 the Seatrack Web user could choose between
three different operational congurations of Hiromb (Funkquist and
Kleine, 2007; Axell, 2013). The rst, BS01, was a one nautical mile res-
olution setup covering the Baltic Sea with an open boundary in Skagerak
and forced by the operational HIRLAM atmospheric model (Dahlgren
et al., 2014). The second conguration, NS03, was a three nautical mile
resolution setup covering the Baltic Sea and the North Sea, also forced
with the operational HIRLAM. These two setups produced 60 h forecasts,
four times per day. The third conguration, sometimes referred to as
NS03 long, had a three nautical mile resolution covering the Baltic Sea
and the North Sea and was forced by forecasts from ECMWF producing
10 day forecasts once per day. Since February 2017 the forcing input is
provided by the operational two nautical mile version of Nemo-Nordic
L. Arneborg et al. Ocean Engineering 144 (2017) 340350

forced by forecasts from ECMWF, producing10 day forecasts twice per
day. In this work we will use the three nautical mile conguration of the
Hiromb model, as described in more detail below, as well as archived
data from the one nautical mile BS01 conguration.
2.1. PADM overview
PADM is a exible Lagrangian particle advection and dispersion
model that can be used to calculate particle trajectories based on the
output of a hydrodynamic (HD) model including an ice model, an
optional wave model, and the atmospheric data that drives the HD
model. The model is rather exible and can be used with a variety of HD
models and substances, ranging from passive tracers, oating objects and
plankton, to oil. Here we will give a brief description of the oil part of the
model, focusing primarily on the advection and spreading, but with brief
descriptions of other processes as well, e.g., dispersion and oil weath-
ering. For a more detailed presentation the reader is referred to Liung-
man and Mattsson (2011). The new oil-in-ice parameterization is
described in Section 2.2.
2.1.1. Oil drift and spreading without ice
Oil is represented in the model by a cloud of particles that have a
number of properties, including position, volume, mass, thickness, etc.
These are advected by the three-dimensional ow eld predicted by the
hydrodynamic (HD) model, as well as the Stokes drift, dispersion, and
random displacements caused by turbulence, as described in more detail
below. When the oil particles are located on the surface, each particle is
represented as a circular disc of oil of a certain thickness and radius. The
radius increases and the thickness decreases with time to represent the
gravitational spreading of the oil. In addition, oil may be dispersed into
the water column as a cloud of small droplets due to the action of
breaking waves on a surface slick, and oil below the surface may move
vertically due to buoyancy forces. This is described further below.
2.1.1.1. Advection. The particle positions, xp, are updated based on the
particle velocity vectors, up, using a rst order forward difference scheme


uS z U s exp  As DKz Ps : (3)
here, U s is the surface Stokes' drift horizontal velocity, D is the particle
depth, Kz is the zero-crossing wave number, and As and Ps are tting
parameters. The parameterized Stokes' drift velocity is tted to the actual
Stokes' drift velocity calculated from the full wave spectrum using a least-
squares method.
The three-dimensional turbulent velocity components ut are deter-
mined from the turbulent kinetic energy, its vertical gradient and dissi-
pation as calculated by the hydrodynamic model, using a Markov chain
algorithm taken from Rahm and Svensson (1986).
For more details on the representation of Stokes drift and turbulence
in the model, see Liungman and Mattsson (2011).
2.1.1.3. Gravitational spreading. The spreading of an oil slick under the
inuence of gravity and viscosity is implemented according to Fay (1969)
where the horizontal area of the slick at time, t, is given as
!1=3
V 2 g0
At 2:1 p t 1=2 : (4)
=
here V is the total volume of the oil slick, g is the buoyancy acceleration
of the oil relative to water, and and are the viscosity and density of
water, respectively. According to Venkatesh et al. (1990), later experi-
ments have shown better agreement with data if the viscosity of water in
Eq. (4) is replaced by that of oil, oil. One may now obtain the following
equation for the rate of increase in oil slick area for a slick of con-
stant volume
dA 1 k 2 V 4=3
; (5)
dt 2 A
where
!1=3
g0
k 2:1 p : (6)
oil =

xp t t xp t up tt; (1) The rate of thickness change of the oil slick, h V/A, can then be
derived as
where t is the time step. The particle velocity consists of contributions
from the ambient current, uHD, the Stokes drift, uS, turbulent uctuations, dh h2 dA k2
 h3 2=3 : (7)
ut, and oil-specic contributions, uo, such as gravitational spreading and dt V dt 2V
buoyant sinking or rising If the oil slick volume changes, e.g. due to a continuous spill, or loss of
oil below ice oes, see below, the changes are assumed to be distributed
up uHD uS ut uo : (2)
over the whole oil slick and increase/decrease the thickness uniformly. In
The current velocity, uHD, is interpolated in space and time to the addition, each particle may change its individual thickness due to oil
present particle position. weathering processes, as shown in the next section. The total rate of
The coast is specied independently of the HD grid, and if an oil change in oil particle thickness can then be written as
particle passes the coastline it is assumed to cease motion and hold its  
dhp 1 dV k2 dh
position from then on. The same happens if a particle crosses the hori-  h3 2 : (8)
zontal or vertical bottom of the model domain. dt A dt 2V 3 dt ow

where ow indicates the contribution from oil weathering.

2.1.1.2. Stokes drift and turbulence. As few hydrodynamic models
From the thickness of each particle, the corresponding radius can be
include Stokes drift or subgrid turbulent motions, the inuence of these
calculated assuming a circular disc shape, and if two particles are closer
processes on the particle motion must be determined and added sepa-
together than the sum of their radiuses, i.e., if they overlap, they are
rately, Eq. (2). Note that both of these processes are treated as local, i.e.,
given opposite velocity vectors just large enough to remove this over-
they are subgrid processes whose magnitudes are calculated at the cur-
lapping during one time step. This procedure ensures that the particle
rent particle position, regardless of the hydrodynamic grid resolution.
cloud spreads out in a manner similar to a continuous oil slick, Eq. (4).
Thus they will vary in space within a grid cell.
The thinning, Eq. (8), continues until the particles reach a minimum
The Stokes' drift is calculated from a wave spectrum that may vary in
oil thickness, which is given by the empirical expression (Venkatesh
time and space. This wave spectrum may, e.g., originate from a spectral
et al., 1990)
wave model or be estimated based on the wind. As it is computationally
very expensive to calculate the Stokes' drift from the full wave spectrum oil
for each particle and time step, PADM takes as input a parameterized hp;min 106 : (9)
125
Stokes drift of the form

342
L. Arneborg et al.
2.1.1.4. Dispersion and buoyancy. Breaking surface waves can break up a
surface slick into small droplets that are displaced down into the water
column in a process that is called dispersion. In the model, dispersion is
represented by randomly picking a number of particles, assigning a
droplet size to each, and displacing them down into the water column.
The amount of oil, and thus the number of particles, to disperse depends
on the wave height, the wind speed, the oil viscosity, the slick thickness,
and the ice concentration according to the formula of Delvigne and
Sweeney (1988). The droplet sizes are selected such that they cover a
reasonable size distribution (105 to 2  103 m), whereas the dispersion
depth is chosen as a random value between the surface and a maximum
intrusion depth that depends on the wave height. For a more detailed
description, see Liungman and Mattsson (2011).
Oil below the surface will sink or rise depending on its buoyancy. The
sinking or rising velocity depends on the buoyancy, the water viscosity,
and the droplet diameter (Soares dos Santos and Daniel, 2000). For
dispersed oil the droplet diameter is given by the dispersion routine. For
oil released at depth, the droplet diameter is chosen randomly from a
normal distribution with mean 0.5 mm, standard deviation 0.2 mm, and
minimum diameter 0.01 mm, that was estimated by tuning model results
against an actual sub-surface spill. For particles mixed down by turbu-
lence, the diameter is equal to the thickness of the surface spill.
2.1.2. Oil weathering
Oil is weathered over time due to evaporation, emulsication,
dissolution, biodegradation, and photooxidation (Afenyo et al., 2016b).
In the present version of PADM, the two most important weathering
processes for short-term simulations are included: evaporation and
emulsication. These in turn modify the mass, the density and the vis-
cosity of the oil, as well as the mass of water in the case of a
water-in-oil emulsion.
PADM can use two types of weathering algorithms. The rst uses a
two-component approach where the oil is assumed to consist of a volatile
and a non-volatile component, where the evaporation of the volatile
component is calculated according to logarithmic expressions (Fingas,
1999; Lindgren et al., 2001) and the rate of emulsication is a function of
the wind speed (Lehr, 2001).
The second type uses proprietary empirical formulae developed by
SINTEF based on a large set of experimental data.
Both types of weathering algorithms rely on databases of chemical
and physical properties for all oils to be considered. For further details
see Liungman and Mattsson (2011).
2.1.3. Old oil-in-ice implementation
In the old oil-in-ice implementation, the particles module requires ice
concentrations and ice drift velocities as output from an ice model
coupled to the HD model. The particles move more or less regardless of
the ice as long as the ice concentration is less than 0.7. For larger ice
concentrations, the oil follows the ice drift, uice, i.e.

uHD uS ut uo ; c < 0:7
up : (10)
uice uo ; c > 0:7
Note that the oil can still spread out gravitationally even though it is
advected with the ice drift.
The Stokes drift decreases with increasing ice concentration accord-
ing to
8
< 0:7  cu ; c < 0:7
S0
uS 0:7 (11)
:
0; c > 0:7
where c is the ice concentration, and uS0 is the Stokes drift without ice.
The amount of dispersed oil also decreases with increasing ice concen-
tration, but with a linear decrease between the ice concentrations 0.3 and
0.8, i.e. for ice concentrations below 0.3 the dispersion is unaffected, and
343
Ocean Engineering 144 (2017) 340350
for ice concentrations above 0.8 there is no dispersion. The emulsica-
tion and evaporation decrease with increasing ice concentration in the
same way as the dispersion.
The oil slick area is decreased by a factor (1-c) which means that the
factor k in Eq. (5) becomes
!1=3
g0
k 1  c2:1 p : (12)
oil =
Ice is not considered in the relation between horizontal area, volume
and thickness, which means that the calculated thickness represents a
mean thickness if the particles were spread out over the whole surface,
including the ice-covered surfaces. In practice this means that the surface
area of oil between ice oes decreases with a factor (1 - c)2, and that the
oil thickness increases with a similar amount relative to ice-free condi-
tions. This also means that the minimum oil thickness is a factor (1 - c)1
larger than Eq. (9).
2.2. New oil-in-ice implementation
In the new ice implementation, ice thicknesses are required input in
addition to the ice concentrations and ice drift velocities required by the
old implementation. Oil may be situated at the free surface between ice
oes, where it moves either with the wave and current drift or with the
ice drift, or it may be situated below the ice, where it moves with the ice
drift or with the current, as described further below.
2.2.1. Oil at the free surface
When oil is situated at the surface, its movements will be dependent
on the ice concentration in a manner similar to the old implementation,
but with a gentler transition from small to large ice concentrations. In
addition, the relation between oil thickness and horizontal area is
described in a more consistent manner.
At low ice concentrations (c < 0.3), all oil particles move almost as
without ice. The only difference is that the area contaminated by oil (see
Fig. 2) is larger due to the presence of the ice. The thickness evolution is
still described by Eqs. (8), (9) and (12), but now each particle is assumed
to cover the area
Vp
Ap : (13)
1  chp
Hence, they spread out over a larger area compared to the old
implementation, such that the actual oil area spreading is consistent with
Eqs. (5) and (12). The Stokes drift is still reduced according to Eq. (11).
For ice concentrations above 0.3, some of the oil particles will start to
move with the ice drift velocity, while the remaining particles move in
the same manner as for c < 0.3. The fraction of oil particles moving with
the ice drift increases linearly towards one at the ice concentration
c 0.8. This is implemented by calculating the probability that each
Fig. 2. Denition of contamination area for oil in ice covered region (after Venkatesh
et al., 1990).
L. Arneborg et al. Ocean Engineering 144 (2017) 340350

particle will shift its state between moving freely or moving with the ice (c > 0.8), oil spreading mainly takes place below the ice, and therefore
drift. If, as an example, the ice concentration increases from c1 to c2, some the spreading routine for these particles takes into account the presence
of the remaining free particles will attach to the ice. If the ice concen- of ice-free water.
tration decreases, none of the free particles will attach to the ice. Thus, Oil particles under ice follow the ice drift velocity as long as the
the probability of an oil particle that moves freely to become attached to difference between the ice drift velocity and the water velocity below is
the ice is given by not too large. When the velocity difference between ice and water ex-
8 ceeds a certain threshold, oil will be peeled off from the ice surface and
< f c1  f c2 ; c2 > c1
> instead move with the water velocity, i.e.
Pfreeattached f c1 ; (14) 
>
: uice uo ; juHD  uice j < uthres
0; c2 < c1 up;ui : (20)
uHD uo ; juHD  uice j > uthres
where f(c) is the fraction of particles that move freely at concentration c. The threshold velocity, typically of magnitude uthres 0.2 m/s (Cox
Here and below the arrow indicates that the oil particle changes from one and Schultz, 1981), is dependent on the buoyancy of the oil, the inter-
state to another, in this case from moving freely to becoming attached to facial tension between oil and water, and the shape of the ice cavities that
the ice. Similarly, when the ice concentrations decreases from c1 to c2, contain the oil (Cox and Schultz, 1981). In addition, the oil velocity will
some attached particles will become free, whereas the attached particles depend on the friction between oil an ice and between oil and water, and
will remain attached when the ice concentration increases. The proba- show a more continuous behavior than that in Eq. (20). However, this
bility that an oil particle that is attached to the ice will start to move simplistic expression, with a constant threshold velocity given as input to
freely is therefore the model, is chosen due to the general lack of knowledge about the
8 under-ice geometry, and because it gives us the opportunity to test the
>
< 0; c2 > c1 sensitivity to this parameter.
Pattachedfree f c2  f c1 : (15)
>
: ; c2 < c1
1  f c1 2.2.3. Transfer between states
Oil particles at the free surface may transfer between moving freely or
At high ice concentrations (c > 0.8), oil is assumed to be so conned with the ice drift velocity, as described above. In addition, they may end
by the ice oes that it cannot spread out anymore. The thickness of oil up under the ice, and oil under the ice may end up at the free surface. A
between oes is then assumed to be large enough to ll up any cavities particle situated below the ice (in contact with the ice or in the free water
there. When the density of oil, oil, is larger than that of ice, ice, the column) may move to a position on or under the ice-free water surface if
maximum oil thickness between ice oes is then given by the horizontal particle velocity is different from that of the ice drift. The
ice probability that this will happen depends on the movement of the par-
hp;max hice ; (16) ticle relative to the ice during a time step, i.e., s (jup uicej)t, and a
oil
typical length scale of the ice oes, Loe.
and the remaining oil will spread out under the ice. When oil is lighter For an innitesimal time step, the probability to remain under the
than ice, the maximum oil thickness is ice is

w  ice Lfloe  s
hp;max hice ; (17) Pbelow icebelow ice : (21)
w  oil Lfloe

where w is the density of water. The remaining oil will spread out on top For a number of innitesimal time steps, or a larger time step, a better
of the ice. representation would be
 
s
2.2.2. Oil under ice Pbelow icebelow ice exp  : (22)
Lfloe
When oil rises towards the surface in ice covered waters, it is assumed
that the probability of an oil particle ending up under ice is equal to the The probability of shifting from a position below ice to a position
ice concentration. The amount of oil that will be trapped under ice oes below the ice-free surface is therefore given by
depends on the standard deviation of the ice thickness as a measure of the  
ice cavities that can trap oil. According to Venkatesh et al. (1990) the s
Pbelow icenot below ice 1  exp  : (23)
mean oil thickness under thick ice (>0.5 m) is Lfloe

hp;ui 0:021hice ; (18) Conversely, the probability to go from a position below the ice-free
surface to a position below the ice is
where ui indicates under ice. Ice in the Baltic is generally thinner than 8  
0.5 m and the relative ice thickness standard deviation is also generally >
< 1  exp  cs
; zp <  hice
smaller than for multiyear ice. However, lacking alternatives to Eq. (18), Pnot below icebelow ice 1  cLfloe ;
>
:
we retain this as the default relation. When spreading out the particles 0; zp >  hice
under ice, the area of each particle is
(24)
8
>
>
Vp
< hp;ui ;
> c < 0:8 where the ice concentration dependency comes in because the length
Ap;ui : (19) scale of the leads relates to that of the oes by the factor (1 - c)/c. Note
>
> Vp that oil particles at the surface will not be able to move under the ice by
>
: ; c > 0:8
chp;ui this process. Only oil particles situated deeper in the water column than
the ice draft will be able to move below the ice. The model includes a
The reason for the differentiation between ice concentrations below
variable, Loe, that is not typically given as output from ice models. This is
and above 0.8 is as follows. At low and medium ice concentrations, where
instead specied as an input parameter, and the sensitivity of the results
oil spreading is mainly governed by oil on the free surface, the particles
to the magnitude of this parameter is studied further below.
under ice interact directly with each other. At high ice concentrations

344
L. Arneborg et al. Ocean Engineering 144 (2017) 340350

For low and medium ice concentrations, oil on the surface will spread
out and the oil slick will typically be much thinner than the ice thickness.
For high ice concentrations, spreading of the oil is restricted, and the oil
in the gaps between ice oes may become so thick that some of it changes
from being at the surface to being under ice. The model keeps track of the
volumes below ice, Vui, and at the ice-free surface, Vsurf. At high ice
concentrations it is assumed that the lateral expansion of the oil is
controlled by the oil lling up the gaps between ice oes with a thickness
given by Eq. (16), as well as spreading out below the ice oes with an oil
thickness given by Eq. (18). The area contaminated by oil (Fig. 2)
then becomes
V assimilation was employed for the simulations; see Axell and Liu (2016).
Atot ; (25)
1  c ice
oil
hice chui
where the oil thickness below ice, hui, is given by Eq. (18). In reality, the
fractions of oil at the surface and below ice may not be as assumed in Eq.
(25). The thickness of oil at the surface is then given by
Vsurf
hsurf : (26)
1  cAtot
where Atot is given by Eq. (25). When the oil thickness is larger than the
maximum oil thickness, a fraction of the particles will change state from
being at the free surface to being under ice, and the remaining surface
particles will reduce their thickness according to the rst term in Eq. (8)
and spread out. The probability of a change in state is
 
hsurf  hp;max
Pnot below icebelow ice min ;0 (27)
hsurf
parameter for the upper ocean that depends on the amount of ice at the
surface, (ii) an improved wind stress parameterization that takes into
account atmospheric stability, (iii) an improved air-ice stress, (iv) an
improved ice-ocean stress, (v) a simple parameterization of fast ice,
depending on depth and ice concentration, and (vi) improved
thermodynamics.
The HIROMB NS03 setup has a horizontal resolution of three nautical
miles and a vertical resolution of 4 m near the surface. The atmospheric
model used as forcing was the High-Resolution Limited Area Model
(HIRLAM) with an 11 km resolution from the Euro4M atmospheric
reanalysis project (Dahlgren et al., 2014). To keep the results close to
observations, Three-Dimensional Ensemble Variational (3D EnVar) data
The observations used in the assimilation were sea surface temperature
(SST), ice concentration and level ice thickness. The assimilation was
performed at 00 UTC every day, and input les to PADM are available
every hour.
The ice thickness data came from the IceMap system, which has been
used to deliver operational, daily charts of ice concentration and ice
thickness at SMHI for several decades. The charts are hand drawn using
available satellite images and in-situ observations from ice breakers and
other ships, to produce a gridded data set of 4.4 km resolution. Though
the IceMap product also includes ice concentration, it was decided to use
ice concentration data from the Advanced Microwave Scanning Radi-
ometer for the Earth Observing System (AMSR-E instrument) on board
the Aqua satellite, to obtain a more accurate variability within the ice
eld compared to IceMap.
To test the sensitivity to the HD forcing we also present results forced
by the original 2006 operational HIROMB BS01 model version 3.0.

where hp,max is given by Eq. (16). 3.2. Runs

When the volume of oil under ice increases, the oil will spread out
according to the gravity spreading routine, so the minimum distance
between particles corresponds to the particle area given in Eq. (19).
3. Simulations
3.1. HD and ice models
The HD model used is the High-Resolution Operational Model for the
Baltic (HIROMB). It is a three-dimensional, fully baroclinic circulation
model which includes an ice model. HIROMB was the operational ocean
forecast model at SMHI during 19952017 but has now been replaced by
a model called NEMO-Nordic. For a description of HIROMB and its ice
model, see Funkquist and Kleine (2007) for HIROMB version 2.0 and
Axell (2013) for HIROMB version 3.0. The version mainly used in these
simulations (HIROMB NS03) was 4.6, which compared to version 3.0
includes some improvements that are important for the ice dynamics.
These include (Axell personal communication, 2017): (i) a roughness
The case that was modelled in this investigation is the Runner 4 oil
spill accident in the Gulf of Finland, described in the introduction.
The base run (R4_Base) was performed using HIROMB_NS03 forcing,
assimilated against the IceMap and AMSR-E ice products. Oil particles
were released at the collision point 59 52.920 N 26 19.840 E at 70 m depth
during a 1 h period starting at 04:00 on the 5th of March 2006. The
trajectory model was then integrated forward in time with a 60 s time
step until 25th May 2006, which is after most of the available observa-
tions of the oil patch. The simulated oil type was a heavy oil, as was the
main part of the real spill. The oil density was 980.1 m3 s1 and the
viscosity was 4.74 103 m2 s1. The ice oe size was 1000 m and the
velocity threshold for oil below ice was 0.1 m s1.
3.2.1. Sensitivity study
A number of runs with different oil-in-ice parameters and forcing
were performed to test the sensitivity to changes in ice parameters and
forcing models, Table 1:

Table 1
Runs.

Name HD model Ice assimilation Ice Param. uthres (m/s) Loe (m)

R4_Original HIROMB NS03 AMSR-E IceMap Old

R4_NoIceDrift HIROMB NS03 AMSR-E IceMap Only advection
with the current.
R4_Base HIROMB NS03 AMSR-E IceMap New 0.10 1000
R4_U10_L100 HIROMB NS03 AMSR-E IceMap New 0.10 100
R4_U10_L10000 HIROMB NS03 AMSR-E IceMap New 0.10 10000
R4_U5_L1000 HIROMB NS03 AMSR-E IceMap New 0.05 1000
R4_U20_L1000 HIROMB NS03 AMSR-E IceMap New 0.20 1000
R4_IceMap_orig HIROMB NS03 IceMap Old
R4_IceMap HIROMB NS03 IceMap New 0.10 1000
R4_BS01_orig HIROMB BS01 Old
2006 operational
R4_BS01 HIROMB BS01 New 0.10 1000
2006 operational

345
L. Arneborg et al. Ocean Engineering 144 (2017) 340350

Two runs were performed to test the sensitivity to the basic ice
parameterization; one with the old oil-in-ice parameterization
(R4_Original), and one where the oil moves with the current without
being inuenced by the ice drift (R4_NoIceDrift). The forcing and oil
release is the same as in the base run.

Two runs were performed to test the sensitivity to ice oe size
(R4_U10_L100 and R4_U10_L10000) with ice oe sizes 10 times
smaller and larger than the base run, respectively.

Two runs were performed to test the sensitivity to the velocity
threshold in Eq. (20) (R4_U5_L1000 and R4_U20_L1000) with veloc-
ity thresholds half and double the base run, respectively.

Two runs were performed to test the sensitivity to the ice assimilation
product (R4_IceMap and R4_IceMap_orig), both using assimilation
against the IceMap concentration and thickness products for the new
and old oil-in-ice parameterization, respectively.

Finally, two runs were performed to test the inuence of the forcing
model (R4_BS01 and R4_BS01_orig), using HIROMB_BS01 instead of
HIROMB_NS03 forcing data for the new and old ice
parameterizations.
4. Results
Fig. 3 shows ice and wind conditions as well as the oil cloud drift and
spreading for the Runner 4 case with the original and the new ice drift
parameterization. Until 17 March the ice moves mainly westward in
response to the easterly winds. After that, the winds become westerly to
northerly, and the ice drifts towards southeast and is packed against the
Estonian coast. The initial release of oil occurs under conditions of high
ice concentration (>0.9).
In the original parameterization, the oil moves with the ice drift with
very little spreading, and ends up on the Estonian coast quite far west of
the observed main oil slick that ended up around Tallinn.
In the new parameterization, the area contaminated by oil spreads out
more, and moves more slowly westward but further northward in the
initial phase with easterly winds. The oil ends up outside Tallinn, as
observed, but a few days later. In the results based on the original
parameterization, the oil remains in areas of high ice concentrations
(>0.8, Fig. 4) for most of the period. In the case of the new parameteri-
zation, most of the oil is initially situated below ice (Figs. 3 and 4) and

Fig. 3. Oil cloud spreading using old and new ice parameterization at different times. Note the different scale in the upper 3 panels. The patch calculated with the old parameterization is
shown with the yellow marker, wheras the particles calculated with the new parameterizations are shown with blue and magenta dots representing particles under ice and at the surface,
respectively. The spill site is indicated with the red star. Background colors show the ice concentration with dark blue corresponding to c 0 and white corresponding to c 1. Contours
show ice thickness in meters. Green arrows show wind velocities, and black arrows show ice drift velocities.

346
L. Arneborg et al.
Fig. 4. Particle statistics for run R4_Base. Upper panel: Fraction of particles moving freely
(dotted), at the surface in contact with ice (full), in the water column below ice (dot-
dashed), and under ice (dashed). Lower panel: Ice concentrations experienced by particles.
moves with the ice drift, but late on the 5th of March the difference
between the ice drift and the current velocity exceeds the threshold of
0.1 m/s and the oil below ice starts moving with the current. This causes
particles below the ice to end up in leads which quickly ll and deliver oil
particles back to the pool below the ice in a continuous exchange. The
velocity difference between the current and ice causes the oil patch to
spread out, as seen on the 6th of March at 12:00 (Fig. 3).
When the ice concentration decreases slightly, for example on the 8th
of March, the fraction of oil at the surface quickly increases, and when the
ice closes in again the oil in the leads is forced back under the ice (Fig. 4).
With a velocity difference of about 0.1 m s1 and a ow size of 1000 m,
the time scale for this adjustment of the oil fractions at the surface and
below the ice to the ice concentration is about 3 h. However, this ex-
change is only taking place if the oil moves relative to the ice, i.e., if the
velocity threshold is exceeded. For example, on the 11th of March the
fraction of particles below the ice remains high, even though the ice
concentration allows more oil in the leads. On the 12th of March the
velocity threshold is exceeded, and the fraction of oil in leads increases.
After the 12th the ice concentration is generally lower than in the initial
phase of spreading, and the fraction of oil in leads is generally larger than
the fraction below the ice.
The oil at the surface moves faster to the west than the oil below ice
(Fig. 3), and on the 14th of March the oil spill has broken into two
patches. On the 16th it has broken further into three patches. The total
extent of the oil contamination is large, in the order of tens of km, similar
to what was observed, but has moved too far north-west, which is seen by
comparing Figs. 1 and 3. However, the oil patch ends up outside Tallinn
as in the observations.
Note that the ice concentrations are larger than 0.8 for the majority of
particles during most of the period, and therefore dispersion and
weathering due to evaporation and emulsication do not inuence the
oil patch in the present version of the model.
4.1. Sensitivity to oil-in-ice parameters
The mean trajectories of various runs for different values of the ice
parameterization parameters are shown in Fig. 5. The trajectory of the
run without ice drift, i.e., where the oil drifts with the surface current,
ends up much closer to the initial release point than the run with the
original ice parameterization. In the latter, the oil basically drifts with the
ice drift, and the result therefore illustrates that the mean ice drift out of
the Gulf of Finland is much larger than the mean surface water drift. The
trajectory of the base run is approximately midway between the run
without ice drift and the run using the original parameterization, which
shows that the current and ice drift are about equally important in
that run.
347
Ocean Engineering 144 (2017) 340350
Fig. 5. Paths of the mean position of the oil patch for various oil parameterizations, see
Table 1. Markers are shown at release time and at 00:00 on the 10th, the 15th, the 20th
and the 25th of March 2006.
The mean trajectory is highly sensitive to the velocity threshold
(Fig. 5). For the velocity threshold uthres 0.2 m/s, the oil moves more-or-
less with the ice drift as in the old parameterization, and moves further
west than observations indicate. If, on the other hand, the oil never
moves with the ice drift, the trajectory is more northward and less
westward directly after the release, and continues this way until after
17th of March, when the eastward movement is also smaller than the
eastward ice movement. The base run with uthres 0.1 m/s and
Loe 1000 m is a combination of these two extremes, and is positioned
about half way between them. The run with uthres 0.05 m/s is more
similar to the base run than to the run without ice drift, which shows that
the response to the velocity threshold is rather non-linear, and that a too
small threshold may work better than a too large threshold, at least in this
particular case. Note that zero velocity threshold is not the same as the
run without ice drift, since oil at the surface will still move with the
ice drift.
The mean trajectory is less sensitive to the oe size parameter than to
the velocity threshold. The difference between the mean trajectories for
the runs with oe sizes 100 and 1000 m is marginal, whereas in the runs
with the largest oe size, 10000 m, the oil ends up somewhat further to
the east.
The oe size is much more important for the oil spreading than for the
mean trajectory, Fig. 6. The horizontal area contaminated by oil is highly
dependent on the oe size. For large oe sizes, the residence time of an
oil particle below the ice or at the surface is larger than for small oe
sizes. For small oe sizes, there is a continuous replacement of oil from
below the ice to the surface and vice versa, whereas particles will remain
longer time at the surface or below the ice for large oes. For large oes,
some particles (at the surface) will move with the ice drift for an extended
period, and others (below the ice) will move with the current velocity, if
the velocity difference exceeds the threshold, and this velocity difference
between the oil particles leads to increased spreading.
The oil spreading is also sensitive to the velocity threshold (not
shown), with very small spreading when the velocity threshold is large,
as in the old parameterization where the oil particles are bound to follow
the ice trajectory, and larger spreading for smaller threshold velocities.
However, the oil spreading is much more sensitive to the oe size.
L. Arneborg et al.
Fig. 6. Particles distributions at 2016-03-20 04:00 for the runs R4_U10_L100 and R4_U10_L10000. For further description, see Fig. 4.
4.2. Sensitivity to forcing model
Fig. 7 shows the mean oil trajectories for the various HD forcing
models using old and new oil-in-ice parameterizations with
uthres 0.1 m/s and Loe 1000 m. The HD forcing can be seen to play an
important role both in the new and old parameterizations, as one would
expect. The ice drift is much smaller in the BS01 run than in the NS03
runs, which probably has to do with the different drag parameterizations
used in the 2006 operational model (Raudsepp et al., 2010). Fig. 8 shows
the nal distributions for the new ice parameterization forced with
HIROMB_BS01 and HIROMB_NS03 assimilated against IceMap. The
much smaller oil spreading in the BS01 run is probably a consequence of
the smaller ice drift, which implies that the velocity threshold is almost
never exceeded. The oil more or less follows the ice drift. This is also seen
in Fig. 7 where the trajectories of runs R4_BS01_orig and R4_BS01 are
very similar.
The IceMap assimilation run has larger ice drift than the AMSR-E
assimilation run, which is reected in the oil trajectories being shifted
further westward in the IceMap runs than in the AMSRE-E runs.
Fig. 7. Paths of the mean position of the oil patch for various HD forcings. Markers are
shown at release time and at 00:00 on the 10th, the 15th, the 20th and the 25th of
March 2006.
348
Ocean Engineering 144 (2017) 340350
5. Discussion
The velocity threshold as well as the velocity difference between the
ice and the water are functions of the ice roughness. For rough ice, oil
may get trapped between ridges and the drag between water and ice is
large, which means that the velocity difference is smaller and thus would
generally mean a tendency for oil to remain under rough ice. For smooth
ice, as is generally the case in the Baltic, the ice-water drag is smaller, and
there are less roughness elements to trap the oil, which will make the oil
more mobile. There are theoretical and laboratory-based models of the
relative oil-ice movement (e.g., Cox and Schultz, 1981) that include ice
roughness and friction between oil and water and between oil and ice.
The layered hydraulics of the oil-water system that is governed by the
interfacial Froude number is probably also highly relevant. The situation
in the eld, however, is much more complex than in the laboratory or in
theoretical approaches, so with the limited amount of validation data
present we have chosen to use a simpler approach.
The ice oe size has a fairly small inuence on the mean trajectory but
a large inuence on the oil spreading. For large oes, the oil can move
over larger distances before it is trapped in a lead. In addition, it takes
longer time before the leads are lled and oil starts to move back under
the ice. Therefore, in case the velocity threshold is exceeded, the oil
under ice and the oil in leads will move in different directions for a longer
time than in an area with small ice oes, where oil is continuously
exchanged between leads and ice oes.
The statistical description in the model does not calculate what
happens in a specic lead, and therefore oil that ends up in a lead in one
end of the oil patch may cause oil in another end of the oil patch to move
from a lead to under the ice. This will tend to cause removal of oil in leads
in the back end of the patch, whereas it may be more realistic that oil
will be left behind in those leads. The small spreading for small ice oes
seen in Fig. 6 may therefore partly be a model artifact, and it remains to
be shown in reality whether there is such a dependency on ice oe size.
On the other hand, the model does not include oil pumping when leads
close, which would also cause oil in leads to move under the ice. It is
therefore difcult to say how realistic or unrealistic the ice oe size
dependence of the oil spreading is.
There are a number of processes that are not included in this rst step
in improving the oil-in-ice parameterization in Seatrack Web, e.g.,
encapsulation of oil in ice, oil pumping onto and under ice due to closing
leads, oil on ice, interactions of oil and snow, trapping of oil in ice ridges,
biodegradation, etc. However, we have shown that there is a huge
sensitivity of the model to the presently included parameters, and there is
not enough validation data to strictly test whether the model works
better with one parameter set than another. In the end, it will be the HD
forcing that sets the limits for the accuracy of the oil drift. It is reasonable
to expect that the oil will move somewhere between the ice drift
L. Arneborg et al.
Fig. 8. Particles distributions at 2016-03-20 03:00 for the runs R4_IceMap and R4_BS01. For further description, see Fig. 4.
trajectory and the current drift trajectory, and if there is a large vari-
ability between those trajectories in different HD models using different
atmospheric forcing, ice models, assimilation parameters, etc., there will
be a large variability in predicted oil trajectories, independent of the ice
parameterization used. The path towards improved oil trajectory
modeling in ice must therefore go both via improved current and ice drift
models and via improved oil-in-ice parameterizations. The focus in the
HD model validation should not only be on the ice drift, but also on the
currents below the ice under various ice conditions. In addition, the
present results motivate the need for ice oe size predictions in
ice models.
The main challenge in the development of oil trajectory modeling in
ice is the lack of validation data. In the Baltic there is only one accident
where a reasonable amount of oil has been tracked for some time. Here
we have focused on that case, and there is a fundamental problem when
tuning a model against a data set and using the same data set for vali-
dation. For example, the ice concentration was rather high during the
whole Runner 4 spill period, which means that we do not know how well
the model performs for lower ice concentrations. Hence, we do not claim
to have developed a validated model for oil trajectory modeling in ice in
the Baltic. We have developed an ice model that gives reasonable results
for the Runner 4 case, which can be tested and validated against future oil
spills in ice and thus provide more information on the parameter values
and the needs for further improvements.
6. Summary and conclusions
An improved oil-in-ice parameterization has been implemented in the
oil trajectory model Seatrack Web, and the model has been tested against
the Runner 4 oil spill accident in the Gulf of Finland in 2006 with the
main aims of identifying important oil-ice interaction parameters and
investigating the sensitivity of the model results to these parameters and
the forcing elds.
Compared with the old oil-in-ice parameterization where oil followed
ice drift for ice concentrations larger than 70% (0.7), the new parame-
terization has a continuous transition from low to high ice concentra-
tions, where more and more oil at the surface becomes inuenced by the
ice as the concentration increases. In addition, oil can be situated below
ice oes, and be exchanged between that position and the leads/free
surface. Oil under ice moves with the ice drift if the velocity difference
between the ice and water below is less than a threshold velocity, uthres.
For larger velocity differences, the water velocity advects the oil.
The model is able to predict the observed transport of oil from the
release point of the Runner 4 accident to Tallinn as well as the spreading
of the oil patch for reasonable values of the oil-in-ice parameters.
The results are sensitive to the oil-in-ice parameters. The velocity
threshold and the ice oe size, Loe, are two parameters that greatly
349
Ocean Engineering 144 (2017) 340350
inuence both the spreading and mean trajectory of the oil. When the
velocity threshold is large, oil under ice will tend to stay there, and in
large ice concentrations where oil in leads also follow the ice drift, the oil
will remain concentrated in a small region. When the velocity threshold
is small, oil under ice will tend to follow the water currents below the ice,
which may be different from the ice drift. Oil under ice will then move
with the water current while oil in the leads will move with the ice drift,
and the oil will be spread out over a large area.
Regardless of the ice parameterization parameters, the oil ends up
somewhere between the end point of the trajectory based on only the ice
drift and the trajectory based only on the water current drift, which in
turn are highly dependent on the hydrodynamic forcing and the assim-
ilation of ice conditions. For example, with everything else equal, the
shift of ice thickness assimilation from AMSR-E to IceMap caused the oil
to end up 40 km further west, an effect that is almost as big as the dif-
ference between the new and old ice parameterization. In future de-
velopments of oil spill modeling in ice it is therefore important to focus
not only on the ice parameterization but also to continuously improve
and validate the HD models with regard to ice drift and the currents
below the ice. In addition, including the prediction of ice oe size in the
ice modeling would be of value for the oil spill modeling.
Acknowledgments
The work presented in this paper was primarily carried out within the
BONUS STORMWINDS project, supported by funding received from
BONUS, the joint Baltic Sea research and development programme (Art
185), funded jointly from the European Union's Seventh Programme for
research, technological development and demonstration, and from the
Swedish Research Council FORMAS, and the Academy of Finland. The
authors acknowledge this nancial support. We also acknowledge the
suggestions for improvements to Seatrack Web presented by Johan
S
oderkvist in an internal SMHI report.
References
Afenyo, M., Khan, F., Veitch, B., Yang, M., 2016a. Modeling oil weathering and transport
in sea ice. Mar. Pollut. Bull. 107 (1), 206215.
Afenyo, M., Veitch, B., Khan, F., 2016b. A state-of-the-art review of fate and transport of
oil spills in open and ice-covered water. Ocean. Eng. 119, 233248.
Ambjorn, C., Mattsson, J., 2006. A new version of the operational oil drift forecasting
system Seatrack Web. In: Dahlin, H., Flemming, N.C., Marchand, P., Petersson, S.E.
(Eds.), European Operational Oceanography: Present and Future, Proceedings of the
Fourth International Conference on EuroGOOS, pp. 1116.
Ambjorn, C., Liungman, O., Mattsson, J., Hkansson, B., 2014. Seatrack Web: the
HELCOM Tool for oil spill prediction and identication of illegal polluters. In:
Kostianoy, A.G., Lavrova, O.Y. (Eds.), Oil Pollution in the Baltic Sea. Springer-Verlag,
Berlin Heidelberg, pp. 155184.
ASCE, 1996. State-of-the-art review of modeling transport and fate of oil spills, ASCE
Committee on Modeling Oil Spills. Water Resour. Eng. Div. J. Hydraul. Eng. 594609.
November.
L. Arneborg et al.
Axell, L., 2013. BSRA-15: a Baltic Sea Reanalysis 1990-2004. Reports Oceanography, 45.
Swedish Meteorological and Hydrological Institute, Norrkoping, Sweden.
Axell, L., Liu, Y., 2016. Application of 3-D ensemble variational data assimilation to a
Baltic Sea reanalysis 1989-2013. Tellus A (68), 24220.
Brandvik, P.J., Faksness, L.G., 2009. Weathering processes in Arctic oil spills: meso-scale
experiments with different ice conditions. Cold Reg. Sci. Technol. 55 (1), 160166.
Cox, J.C., Schultz, L.A., 1981. The containment of oil spilled under rough ice. Int. Oil Spill
Conf. Proc. 203208.
Dahlgren, P., Kallberg, P., Landelius, T., Unden, P., 2014. EURO4M Project - Report, D 2.9
Comparison of the Regional Reanalyses Products with Newly Developed and Existing
State-of-the Art Systems. Technical Report. http://www.euro4m.eu/Deliverables.
html.
Delvigne, G.A.L., Sweeney, C., 1988. Natural dispersion of oil. Oil Chem. Pollut. 4 (4),
281310.
Drozdowski, D., Nudds, S., Hannah, C.G., Niu, H., Peterson, I., Perrie, W., 2011. Review of
Oil Spill Trajectory Modelling in the Presence of Ice. Fisheries and Ocean, Canada,
p. 274. Canadian Technical Report of Hydrographic and Ocean Sciences.
Faksness, L.G., Brandvik, P.J., Daling, P.S., Singsaas, I., Srstrm, S.E., 2016. The value of
offshore eld experiments in oil spill technology development for Norwegian waters.
Mar. Pollut. Bull. 111 (1), 402410.
Fay, J.A., 1969. The Spread of Oil Slicks on a Calm Sea. In: Hoult, D.P. (Ed.), Oil on the
Sea. Ocean Technology. Springer, Boston, MA.
Fingas, M.F., 1999. The evaporation of oil spills: development and implementation of new
prediction methodology. In: Proceedings of the 1999 International Oil Spill
Conference. American Petroleum Institute, Washington, D.C, pp. 281287.
Fingas, M.F., Hollebone, B.P., 2003. Review of behaviour of oil in freezing environments.
Mar. Pollut. Bull. 47, 333340.
Funkquist, L., Kleine, E., 2007. HIROMB: an Introduction to HIROMB, an Operational
Baroclinic Model for the Baltic Sea, Report Oceanography 37. Swedish
Meteorological and Hydrological Institute, Norrk oping, Sweden.
Goerlandt, F., Goite, H., Valdez Banda, O.A., H oglund, A., Ahonen-Rainio, P., Lensu, M.,
2017. An analysis of wintertime navigational accidents in the Northern Baltic Sea.
Saf. Sci. 92, 6684.
Hoglund, A., Pemberton, P., Hordoir, R., Schimanke, S., 2017. Ice conditions for maritime
trafc in the Baltic Sea in future climate. Boreal Env. Res. 22, 245265.
Kollo, M., Laanearu, J., Tabri, K., 2017. Hydraulic modelling of oil spill through
submerged orices in damaged ship hulls. Ocean. Eng. 130, 385397.
350
Ocean Engineering 144 (2017) 340350
Lee, K., Boufadel, M., Chen, B., Foght, J., Hodson, P., Swanson, S., Venosa, A., 2015. The
Behaviour and Environmental Impacts of Crude Oil Released into Aqueous
Environments. The Royal Society of Canada, Ottawa.
Lehr, W.J., 2001. Review of modelling procedures for oil spill weathering behavior. In:
Brebbia, C.A. (Ed.), Oil Spill Modelling and Processes. WIT Press, Southampton, UK,
pp. 5190.
Lindgren, C., et al., 2001. Seatrack Web - algoritmer f or vadringsprocesser p oljor ute till
havs. In: IVL Svenska Milj oinstitutet AB, report for the Swedish Coast Guard and the
Swedish Rescue Servicces Agency. September 2001 (in Swedish).
Liungman, O., Mattsson, J., 2011. Scientic Documentation of Seatrack Web; Physical
Processes, Algorithms and References, p. 32. https://stw.smhi.se/.
Rahm, L.-A., Svensson, U., 1986. Dispersion of marked uid elements in a turbulent
Ekman layer. J. Phys. Oceanogr. 16, 20842096.
Raudsepp, U., Ambj orn, C., Axell, L., Lepparanta, M., Rasmus, K., Uiboupin, R., Vainio, J.,
2010. Improved Ice Forecasts in HIROMB and Seatrack Web, p. 30. SIDA report.
Sergejeva, M., Laanearu, J., Tabri, K., 2017. In: On Parameterization of Emulsication and
Heat Exchange in the Hydraulic Modeling of Oil Spill from a Damaged Tanker in
Winter Conditions. 6th International Conference on Marine Structures
(MARSTRUCT2017), pp. 810. May 2017, Lisbon, Portugal.
Soares dos Santos, A., Daniel, P., 2000. Oil spill modelling near the Portuguese coast. In:
Rodriguez, G.R., Brebbia, C.A. (Eds.), Oil & Hydrocarbon Spills II. WIT Press,
pp. 1118.
Stanovoy, V.V., Eremina, T.R., Isaev, A.V., Neelov, I.A., Vankevich, R.E.,
Ryabchenko, V.A., 2012. Modeling of oil spills in ice conditions in the Gulf of Finland
on the basis of an operative forecasting system. Oceanology 52, 754759.
Valdez Banda, O.A., Goerlandt, F., Montewka, J., Kujala, P., 2015. A risk analysis of
winter navigation in Finnish sea areas. Accid. Anal. Prev. 79, 100116.
Valdez Banda, O.A., Goerlandt, F., Kuzmin, V., Kujala, P., Montewka, J., 2016. Risk
management model of winter navigation operations. Mar. Pollut. Bull. 108 (12),
242262.
Venkatesh, S., El-Tahan, H., Comfort, G., Abdelnour, R., 1990. Modelling the behaviour of
oil spills in ice-infested waters. Atmosphere-Ocean 28, 303329.
Wang, K., Lepparanta, M., Gastgifvars, M., Vainio, J., Wang, C., 2008. The drift and
spreading of the Runner 4 oil spill and the ice conditions in the Gulf of Finland,
winter 2006. Est. J. Earth Sci. 57, 181191.