A coupled ocean-atmosphere mesoscale ensemble prediction system has been developed by the Naval Research Laboratory (NRL). The system is designed to take into account major sources of uncertainty in: i) non-deterministic dynamics, ii) model error, and iii) initial states. The purpose of the system is to provide mesoscale ensemble forecasts for use in probabilistic products, such as reliability and frequency of occurrence, and in risk management applications. The system components include COAMPSŪ (Coupled Ocean/Atmosphere Mesoscale Prediction System) and NCOM (Navy Coastal Ocean Model) for atmosphere and ocean forecasting, and NAVDAS (NRL Atmospheric Variational Data Assimilation System) and NCODA (Navy Coupled Ocean Data Assimilation) for atmosphere and ocean data assimilation. NAVDAS and NCODA are 3D-variational (3DVAR) analysis schemes. The ensembles are generated using separate applications of the Ensemble Transform (ET) technique in both the atmosphere (for moving or non-moving nests) and the ocean. The atmospheric ET is computed using wind, temperature, and moisture variables, while the oceanographic ET is derived from ocean current, temperature and salinity variables. Estimates of analysis error covariance, which is used as a constraint in the ET, are provided by the ocean and atmosphere 3DVAR assimilation systems. The newly developed system has been successfully tested for a variety of configurations, including differing model resolution, number of members, forecast length, and moving and fixed nest options. Results from coupled simulations of the Fukushima nuclear accident utilizing an imbedded passive tracer transport capability within COAMPS will be presented.