Oceanic ensemble forecasting in the Gulf of Mexico: An application to the case of the Deep Water Horizon oil spill

•Bred Vectors used for the generation of perturbed ocean initial conditions.•Ensemble mean has lower error than the control forecast.•In the coastal region, the lower error is due to accounting for forcing uncertainty.•Accounting for initial condition uncertainty is more important in the open ocean.

This paper demonstrates the potential of ocean ensemble forecasting in the Gulf of Mexico (GoM). The Bred Vector (BV) technique with one week rescaling frequency is implemented on a 9  km resolution version of the Regional Ocean Modelling System (ROMS). Numerical experiments are carried out by using the HYCOM analysis products to define the initial conditions and the lateral boundary conditions. The growth rates of the forecast uncertainty are estimated to be about 10% of initial amplitude per week. By carrying out ensemble forecast experiments with and without perturbed surface forcing, it is demonstrated that in the coastal regions accounting for uncertainties in the atmospheric forcing is more important than accounting for uncertainties in the ocean initial conditions. In the Loop Current region, the initial condition uncertainties, are the dominant source of the forecast uncertainty. The root-mean-square error of the Lagrangian track forecasts at the 15-day forecast lead time can be reduced by about 10−50 km using the ensemble mean Eulerian forecast of the oceanic flow for the computation of the tracks, instead of the single-initial-condition Eulerian forecast.