Oil spill models are used to forecast the transport and fate of oil after it has been released. CranSLIK is a model that predicts the movement and spread of a surface oil spill at sea via a stochastic approach. The aim of this work is to identify parameters that can further improve the forecasting algorithms and expand the functionality of CranSLIK, while maintaining the run-time efficiency of the method. The results from multiple simulations performed using the operational, validated oil spill model, MEDSLIK-II, were analysed using multiple regression in order to identify improvements which could be incorporated into CranSLIK. This has led to a revised model, namely CranSLIK v2.0, which was validated against MEDSLIK-II forecasts for real oil spill cases. The new version of CranSLIK demonstrated significant forecasting improvements by capturing the oil spill accurately in real validation cases and also proved capable of simulating a broader range of oil spill scenarios.
Oil spills can have damaging effects on the environment and also on human
activities and infrastructure such as fishing, recreation, harbours and power
plants. Adding to the tragic direct fatalities and injuries caused, an oil
spill accident can also have an adverse long-term impact on human health. Oil
spill accidents can lead to severe financial implications too. Costs emerge
not only with the level of damage caused, but also during the implementation
of the response strategy and subsequently during the restoration phase. The
National Commission on the BP Deepwater Horizon Oil Spill and Offshore
Drilling prepared a very comprehensive and detailed report for the US
President about the Deepwater Horizon Spill, where it is reported
Planning at both the strategic and operational level can benefit from the use
of computerised models which predict the path and spread of spilt oil and
changes in its state. A substantial amount of money has been spent by
governments and the oil industry in an effort to develop the capability to
predict the fate of spilt oil. Several models have been developed over the
years
Weathering process, from
The purpose of the CranSLIK oil spill model is to fulfil this responsive
role. The first version of this model, namely CranSLIK v1.0, is described in
The primary objectives of this study are to investigate whether the
forecasting accuracy of CranSLIK v1.0 can be improved by taking into account
additional factors – specifically sub-surface currents, oil density and sea
surface temperature – and also to extend and validate the model's
functionality to a greater range of scenarios. Collecting comprehensive data
on real oil spills in the field is difficult and such data sets are not
readily available. For this reason, CranSLIK bases its development and
validation on results from MEDSLIK-II
Data from two real oil spill cases are available from the MEDSLIK-II team.
The first case is an oil spill of 680
The key steps of this work were as follows.
Identify additional parameters which could be significant for short-term and long-term oil spill prediction. Create samples of input parameter values. Run the MEDSLIK-II simulation using the samples as inputs. Based on the MEDSLIK-II output, fit regression models to map the inputs to the response variables. Incorporate the regression models into the CranSLIK prediction code. Test the developed code against MEDSLIK-II forecasts for real scenarios,
by running both models using identical inputs and comparing the results.
CranSLIK v1.0 forecasts the transport, shape and size of an oil slick in hourly time steps after instantaneous release of the oil at a point in space.
The trajectory of the slick's centre of mass is predicted using analyses of
forecasts of wind and surface current velocity which are provided by
atmospheric and oceanographic models. Wind forcing, i.e. the wind velocity
components at 10
CranSLIK v1.0 can be run in both deterministic and stochastic mode. The former uses the wind and current forecasts exactly as output by the oceanographic and atmospheric models and produces a single forecast, whereas stochastic mode recognises that there is inherent uncertainty in the forecasts and runs the model as a Monte Carlo simulation. For a Monte Carlo run, the wind and current forecasts are sampled randomly from representative probability distributions. Based on the results of multiple runs using different samples, the results can be analysed as a range of possible outcomes with associated probability estimates. This stochastic capability is believed to be one of CranSLIK's key benefits and is made practical by the speed at which it runs.
In CranSLIK v1.0 the shape of the slick is modelled as circular and the
radius forecast is based on the mass of the spill and the age of the slick.
Validation of the model was carried out using the point-mass Algeria test
case which is described in Sect.
CranSLIK v1.0 established the methodology and demonstrated the potential of using approximation methods with stochastic capability. However, it was recognised that there was scope for developing the model further, because its algorithms had been limited to using wind and surface current velocities, spill mass and slick age. It was found that the resulting trajectory forecast needed to be reset to the MEDSLIK-II forecast every few hours to achieve reasonable accuracy, so investigation was recommended to identify other significant factors which could be used to improve forecasting accuracy. The model had also been limited to a particular type of spill, namely, a release of oil at a single point in both time and space in the open sea, so extending its scope to continuous spills and interaction with coastlines, for example, was another potential area for development.
MEDSLIK-II simulates the evolution of a surface oil spill to produce
forecasts of the oil's movement and the change in its condition due to
weathering. The advection–diffusion of the oil is modelled using a
Lagrangian approach whereby the slick is represented as a collection of
constituent particles. Four weathering processes – spreading, evaporation,
dispersion and emulsification – are modelled using empirical formulae which
are largely based on the methods of
The key input parameters to MEDSLIK-II are the forecasts of wind and sea current, which can be obtained from a variety of atmospheric and oceanographic models already described in the previous section. Other important input parameters are the sea surface temperature forecast, the type and properties of the oil and the spill's location, date and time, release rate and release period.
Further details of MEDSLIK-II, including the theoretical foundations as well as the numerical validation and simulations, can be found in
In order to obtain the data required to investigate whether additional
factors should be included in the CranSLIK forecasting algorithms, multiple
scenarios were simulated in MEDSLIK-II, with each run using a different
combination of values of the input variables, viz. wind velocity, current
velocity, oil density, sea surface temperature and spill mass. The location
and time of the spill were the same for each scenario and were based on the
Algeria test case. The other input variables used values taken from the
following intervals:
wind velocity components (N and E): current velocity components (N and E): oil density: 17–45 API; sea surface temperature: 25–29 spill mass: 50–1000
For wind velocity, current velocity and sea surface temperature, the mean
and range of each interval were loosely based on the forecasts used for the
test cases. The range of oil densities was determined by MEDSLIK's
capability and the maximum spill size was set to 1000
Centre of mass trajectory forecasts (hours 1–36) for the Algeria point case using the current at different depths as the advective current.
Centre of mass trajectory forecasts (hours 1–36) for the Algeria contour case using the current at different depths as the advective current.
Distance between MEDSLIK and CranSLIK centre of mass forecasts for the Algeria test case (point and contour scenarios), using the current at different depths as the advective current.
MEDSLIK and CranSLIK forecasts of slick after 36
MEDSLIK and CranSLIK forecasts of slick after 36
Percentage of oil captured by CranSLIK's forecast of the slick's convex hull for the Algeria case (point and contour scenarios), using the current at different depths as the advective current.
Centre of mass trajectory forecasts (hours 1–600) for the Lebanon case using the current at different depths as the advective current.
Distance between MEDSLIK and CranSLIK centre of mass forecasts for the Lebanon test case, using the current at different depths as the advective current.
Percentage of oil captured by CranSLIK's forecast of the slick's convex hull for the Lebanon case, using the current at different depths as the advective current.
MEDSLIK and CranSLIK “good” (
MEDSLIK and CranSLIK “bad” (minimum % oil captured) slick forecasts for the Lebanon case, using the current at different depths as the advective current. Note that CranSLIK does not predict oil concentration.
Polynomial linear regression is a common technique used to derive a model of the form
The method is called linear regression because the model is linear in the
coefficients
The initial part of the investigation considered the effect of the following
input parameters: wind velocity, sea current velocity, oil density, sea
surface temperature and spill mass, which essentially represent the
Various different models were generated that included some or all of the
studied parameters either as linear terms, as higher power terms, or combined
as mixed terms, which essentially constitute the
Based on the above study we conclude that the wind and
the advective current are the main drivers of the MEDSLIK-II forecast. In
particular, the regression analysis showed that, in a given direction, the
slick speed is primarily a function of the wind speed and the advective
current speed and can be modelled by the following linear polynomial formula:
The sensitivity of the oil spill trajectory forecast due to the choice of
current velocity components has been assessed in
As they age, slicks tend to deform, break up and spread out over a large area
variations of wind and current over large slicks are handled more accurately; the model can be initialised using data for an observed slick; continuous spills can be modelled by adding to the collection of mini-slicks as time is advanced during the period when oil is released; stranding of oil on shorelines can be simulated by removing mini-slicks from the collection.
In addition to calculating mini-slick paths, spreading of each mini-slick was
modelled by updating its radius at each time step via a method similar to
that used in CranSLIK v1.0 but with the following revisions. Firstly, the
oil's density was added to the formula, which slightly improved its accuracy.
Secondly, this formula used the slick's age, so in cases where the model was
initialised to an observed slick, its age at the time of observation had to
be estimated. For this purpose, an additional formula based on the ratio of
the spill's mass to its average concentration was derived. Thirdly, the
revised formula for the absolute value of the slick radius was only used to
calculate the radius at the end of the first time step. Thereafter, a new
formula which expressed the radius as a multiple of this value was used. This
was because it gave similar results for a 36
Given that the new methods model a slick as a number of constituent
mini-slicks, does this increase the computational expense to unacceptable
levels? The answer is no, which is most easily demonstrated by the Algeria
test case. The processing for the two scenarios, point and contour, is
virtually identical, but the former is represented as 1 mini-slick and the
latter by more than 4000, which is probably towards the upper limit likely to
be required because the initial observed slick was spread out over a large
area. For both of these scenarios, generation of a 36
The results obtained using these proposed new methods on the test cases are presented and analysed in the next section.
The Algeria case was modelled both by initialising it as an instantaneous
spill at a point (referred to below as the point scenario) and starting from
observed slick data (referred to below as the contour scenario). In both
cases, a 36
For both Algeria scenarios, the CranSLIK and MEDSLIK-II trajectory forecasts
showed good agreement as seen in Figs.
The slick forecasts after 36
By taking the boundary of the CranSLIK slick forecast to be the convex hull
of the mini-slicks, it was straightforward to calculate a useful measure of
CranSLIK's accuracy, namely the percentage of the oil's mass captured
within this closed curve. This measure is plotted in Fig.
For the Lebanon case, it was decided to create a 600
Some example plots to illustrate “good” and “bad” forecasts are given in
Figs.
Where possible, the revised model's forecasts were compared with those given
by CranSLIK v1.0. For the Algeria point scenario, the new trajectory forecast
is significantly better than the v1.0 forecast, which had a maximum error of
approximately 8
This paper discusses further development and enhancement of the CranSLIK
stochastic model. The forecasts of oil slick trajectory, shape and size
obtained using the revised model, namely CranSLIK v2.0, showed good agreement
with MEDSLIK-II in the test cases and significant improvement in forecasting
accuracy compared with CranSLIK v1.0. For the Algeria test case, CranSLIK
v2.0 captured a minimum of 98 % of the amount of oil for a 36
CranSLIK v2.0 incorporates additional functionality which increases the model's flexibility and its ability to handle a wider range of scenarios. Firstly, it can model both an instantaneous and a continuous release of oil. Secondly, the new model can be initialised to an observed slick. Thirdly, the beaching of oil can now be modelled, which is important because it is often near the coast where the potentially damaging effects of the pollution are most keenly felt. The inclusion of oil–shoreline interaction in the new model appears to give reasonable results, but has not yet been fully developed and requires further work. In particular, the ability to automatically handle any shape of coastline needs to be incorporated into the model, and the accuracy of the beaching algorithm and the way it appears to affect the trajectory forecast needs investigation. A major strength of the developed model is its computational efficiency and the minimal time required to perform Monte Carlo simulations and thus generate maximum likelihood regions.
It is believed that CranSLIK has a role to play in both planning and operational mode and merits further development. In addition to the points mentioned above, the model would be enhanced by the development of the stochastic methods used to associate estimates of uncertainty with the forecasts.
The code for CranSLIK v2.0 is open-source code that can be downloaded from
the website
CranSLIK is available under the GNU General Public License (GNU-GPL Version 3, 29 June 2007).
We would like to thank George Zodiatis from the Oceanography Centre of the University of Cyprus for the fruitful discussions held that improved the quality of this work. We gratefully acknowledge the two anonymous reviewers and the topical editor, Ignacio Pisso, for providing constructive comments that improved this paper. The service charges for this open-access publication have been covered by Cranfield University. Edited by: I. Pisso