Retrospective Forecasts of Interannual Sea Surface Temperature Anomalies from 1982 to Present Using a Directly Coupled Atmosphere–Ocean General Circulation Model

Abstract

Abstract A large number of ensemble hindcasts (or retrospective forecasts) of tropical Pacific sea surface temperature (SST) have been made with a coupled atmosphere–ocean general circulation model (CGCM) that does not employ flux correction in order to evaluate the potential skill of the model as a seasonal forecasting tool. Oceanic initial conditions are provided by an ocean data assimilation system. Ensembles of seven forecasts of 6-month length are made starting each month in the 1982 to 2002 period. Skill of the coupled model is evaluated from both a deterministic and a probabilistic perspective. The skill metrics are calculated using both the bulk method, which includes all initial condition months together, and as a function of initial condition month. The latter method allows a more objective evaluation of how the model has performed in the context in which forecasts are actually made and applied. The deterministic metrics used are the anomaly correlation and the root-mean-square error. The coupled model deterministic skill metrics are compared with those from persistence and damped persistence reference forecasts. Despite the fact that the coupled model has a large cold bias in the central and eastern equatorial Pacific this coupled model is shown to have forecast skill that is competitive with other state-of-the-art forecasting techniques. Potential skill from probabilistic forecasts made using the coupled model ensemble members are evaluated using the relative operating characteristics method. This analysis indicates that for most initial condition months this coupled model has more skill at forecasting cold events than warm or neutral events in the central Pacific. In common with other forecasting systems, the coupled model forecast skill is found to be lowest for forecasts passing through the Northern Hemisphere (NH) spring. Diagnostics of this so-called spring predictability barrier in the context of this coupled model indicate that two factors likely contribute to this predictability barrier. First, the coupled model shows a too-weak coupling of the surface and subsurface temperature anomalies during NH spring. Second, the coupled-model-simulated signal-to-noise ratio for SST anomalies is much lower during NH spring than at other times of the year, indicating that the model’s potential predictability is low at this time.