Ensemble Forecasting of Return Flow over the Gulf of Mexico

Errors in operational forecasts of return flow events (RFEs) over the Gulf of Mexico have dictated the search for sources of these errors. Based on earlier studies, likely candidates for these errors are: incorrect parameterization of turbulent transfer processes at the air-sea interface, uncertain vertical motion above the mixed layer, and incorrect initial conditions. We investigate these possible sources of error by performing numerical experiments with a Monte Carlo ensemble prediction model applied to a well-observed case in February 1988. In essence, we examine uncertainty in prediction due to uncertainty in the model’s elements of control. A mixed-layer model with roughly 50 elements of control is used to determine forecast uncertainty due to initial conditions alone, boundary conditions alone, parameterization alone, as well as the full complement of uncertainty in these elements of control. The uncertainty is calculated at points along a predetermined outflow trajectory that originates over shelf waters in the northeastern Gulf, passes north of the Yucatan Peninsula, and terminates in the west-central Gulf-all points along the trajectory are characterized by convective heating at the sea-air interface. Results from the numerical experiments led to the following results: 1) parameterization of physical processes exerts the greatest influence on forecast uncertainty, and 2) the water-vapor mass in the mixed-layer column is uncertain by a factor of two at the trajectory’s terminal point. The latter result confirms forecasters’ long-held view that vapor return is the most suspect product in operational prediction of RFEs. In addition to these numerical experiments with the 1988 case, a recent RFE is examined in the context of operational model performance at the National Center for Environmental Prediction (NCEP). The paper ends with discussion of steps to be taken that hold promise for improved operational prediction of RFEs over the Gulf of Mexico.