Comparison of predictive skills offered by Prognocean, Prognocean Plus and MyOcean real-time sea level forecasting systems

Abstract

We use three real-time forecasting systems (MyOcean, Prognocean, Prognocean Plus) which provide 7-day predictions of gridded sea level anomaly (SLA) data, and carry out two independent comparisons: Prognocean with MyOcean and Prognocean Plus with MyOcean. MyOcean uses the physically based model, while the other two systems utilise a few data-based models. Because of different spatial resolutions of gridded SLA forecasts, two experiments are conducted (grid sizes of 1°×1° and 1/4°×1/4°). The spatial resolutions of SLA forecasts provided by the MyOcean system are artificially decreased from 1/12°×1/12° to the common resolution of 1°×1° and 1/4°×1/4°, for each experiment. The data span the time intervals: 28/04/2013–25/09/2014 and 08/08/2014–31/01/2015, for low and high spatial resolutions, respectively. The predictions from each system are separately compared with SLA data using: root mean square error, mean absolute error, Nash–Sutcliffe efficiency coefficient, coefficient of determination and index of agreement. A rigorous statistical comparison allows us to infer that the data-based approaches may reveal better performance in forecasting sea level variability than the physically based one. It is found that Prognocean and Prognocean Plus perform better than MyOcean in terms of prediction errors, however MyOcean resolves irregular SLA changes better since its prognoses highly correlate with data.