Oil spill risk assessment based on ocean model ensemble prediction system

Accidental oil spills at sea can have severe environmental impact on the marine
environment. To help quantify the risk of petroleum activities, relevant oil
spill scenarios are simulated ahead of time, to form a picture of possible
outcomes, and to estimate needs for response equipment. Formally, an
Environmental Risk Assessment (ERA) may be carried out, in which risk is
quantified by estimating the environmental consequences of different outcomes,
weighted by the probabilities of those outcomes.

Probabilities in ERA are commonly determined by ensemble simulations with an
oil spill trajectory model. Long time series of environmental data are produced
for the relevant area, and the oil spill scenario is simulated repeatedly at
different intervals within the environmental data set. Due to differences in
wind, current and other environmental parameters, the outcome of a scenario
will be different each time, and each simulation in the ensemble constitutes a
sample from the space of possible outcomes.

In this work, we run ensembles with the OSCAR oil spill model, using half a
year of data from 24 different ensemble members of an ocean model EPS (Ensemble
Prediction System) setup for the Barents Sea. We demonstrate that in addition
to the variation in outcomes from running simulations at different times, we
also get variation across the 24 different realisations of the environmental
data. Assuming that each of the ensemble members are equally likely guesses at
the ocean state, the use of the EPS data as input to the oil spill simulations
allow us to explore a larger range of possible outcomes of the oil spill.

The use of EPS in weather forecasting is already common practice, and available
to the public through ranges of uncertainty in weather apps. Given that the
transport of oil at the sea surface is to a large degree controlled by the
wind, the use of EPS data in operational oil spill modelling of ongoing events
is already possible. Making use of such data can help predict significant, but
perhaps unlikely events, such as catastrophic oiling of sensitive beaches.