Lei Ren, Department of Civil Engineering, National University of Ireland, Galway, Ireland

Forecasting and hindcasting of surface currents for Galway Bay using sequential data assimilation algorithms and High Frequency CODAR data

Michael Hartnett


Accurate forecasting of surface currents in coastal waters is of great importance for operations such as search and rescue operations and marine energy extraction. In order to produce accurate information of surface currents in coastal areas, a combination technique, named data assimilation, has been applied to make the best use of radar data into models to enhance modelling performance. Surface currents measured by High Frequency CODAR (Coastal ocean dynamics applications radar) had relatively good quality cross-validated with ADCP (Acoustic Doppler Current Profile) data and wave buoy measurements. Before assimilating the radar data into models, the best “free run” (without data assimilation) of EFDC (Environmental Fluid Dynamics Code) model was set up by sequentially examining wind data, wind stress coefficient, vertical layer thickness structure and bottom roughness height. Radar currents were separately combined into the model using four kinds of sequential data assimilation algorithms: Direct Insertion, Optimal Interpolation, nudging and indirect data assimilation via correcting wind stress. Assimilation parameters in each algorithm were optimized based on generating good patterns of surface currents over hindcasting periods. To extend forecasting improvements, sensitivity tests were performed using the temporally interpolated radar data to determine the best data assimilation cycle length for each algorithm. To assess degrees of implementation complexity and improvement success in each algorithm, modelled results were intercompared with the radar data in qualitative methods and quantitative methods. According to the model forecasting performance, the best developed data assimilation model for simulating surface currents in Galway Bay was to combine the radar currents at each model timestep using nudging data assimilation algorithm.