Use of ensemble prediction systems in operational oil spill modelling

Models play an important role in oil spill response and contingency planning. They are used ahead of time, to analyse possible scenarios and risks, and they are used during an active spill, to help direct response measures. Given the chaotic nature of both the ocean and the atmosphere, predictions are fundamentally challenging, particularly beyond a time window of a few days. To deal with this challenge, weather forecasters have over the last decade been moving towards ensemble prediction systems (EPS), rather than simply focusing on improved model resolution. The idea is that an ensemble of different, but (ideally) equally likely, predictions will provide useful information about the forecast uncertainty.

Given that EPS weather forecast data is increasingly available, these should be routinely used in operational oil spill modelling. In this talk, we present numerical experiments of hypothetical oil spill scenarios off the coast of Norway, using the OSCAR oil spill model. For ocean input data, we use the operational NorKyst800 ROMS setup provided by MET Norway, and for wind we use the MetCoOp Ensemble Prediction System (MEPS), a collaborative product from the Meteorological institutes of Norway, Sweden and Finland. Due to operational availability, we use only a single realisation of the ocean model, but 30 different ensemble members for the atmosphere model. We investigate the potential for the combination of EPS wind data and deterministic current data to provide added value in a response situation, even though the ocean and atmosphere data will not always be dynamically consistent. Since wind, more than currents, drive stranding and entrainment of oil, the use of a wind ensemble will provide some information about the uncertainty in predictions of surface slick as well as locations and amounts of beached oil.

Additionally, we compare the use of deterministic ocean data and EPS wind data with another setup, Barents 2.5 km, where we use a dynamically consistent set of EPS data for both the ocean and the atmosphere. We show that this latter setup has larger variation between ensemble members, indicating the potential added value for oil spill response in moving towards operational EPS forecasting also for ocean models.