Seasonal Sea Level Prediction Using Anomaly Assimilation of the Subsurface Ocean

Many seasonal forecasting systems suffer large climate drifts which can substantially reduce predictive skill. In a forecast initialized with the full observed field, the model adjusts back quickly to its own coupled climatology using the internal model physics. This adjustment can overwhelm the model’s response to the physical anomaly. To reduce inherent model drift, we use anomaly assimilation for a state-of-the-art coupled climate model initialized with three-dimensional ocean temperature and salinity. Ensembles of 12-month retrospective forecasts from 1981–present are compared using two ocean reanalysis products: NCEP’s Global Ocean Data Assimilation System (GODAS) and ECMWF’s Ocean Reanalysis System 4 (ORA-S4). Seasonal-to-interannual skill and uncertainty of the forecasts are assessed as a function of observational noise and also strength of the predicted climate signal, focusing on sea level variability in the tropical Pacific. From these experiments comes a new seasonal forecasting framework, with small model drift, applicable for real-time dynamical sea level prediction.